[Bug tree-optimization/112464] New: [14 Regression] ICE avx512 with -ftrapv since r14-5076

[Bug tree-optimization/112464] New: [14 Regression] ICE avx512 with -ftrapv since r14-5076

[mjires at suse dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

            Bug ID: 112464
           Summary: [14 Regression] ICE avx512 with -ftrapv since r14-5076
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mjires at suse dot cz
                CC: rdapp at gcc dot gnu.org
  Target Milestone: ---

Following reduction of testcase avx-vandpd-1.c causes ICE with -mavx512dq
-ftrapv
Bisection points to r14-5076

$ cat included.c
long *e;
int n, i, err;
void fn() {
  for (; i < n; i++)
    if (e[i])
      err++;
}

$ gcc included.c -Ofast -mavx512dq -ftrapv
during GIMPLE pass: vect
included.c: In function ‘fn’:
included.c:3:6: internal compiler error: in vect_finish_replace_stmt, at
tree-vect-stmts.cc:1353
    3 | void fn() {
      |      ^~
0x91fc95 vect_finish_replace_stmt(vec_info*, _stmt_vec_info*, gimple*)
        /home/mjires/git/GCC/master/gcc/tree-vect-stmts.cc:1353
0x1296bd6 vectorize_fold_left_reduction
        /home/mjires/git/GCC/master/gcc/tree-vect-loop.cc:7191
0x1296bd6 vect_transform_reduction(_loop_vec_info*, _stmt_vec_info*,
gimple_stmt_iterator*, gimple**, _slp_tree*)
        /home/mjires/git/GCC/master/gcc/tree-vect-loop.cc:8456
0x1fac0f5 vect_transform_stmt(vec_info*, _stmt_vec_info*,
gimple_stmt_iterator*, _slp_tree*, _slp_instance*)
        /home/mjires/git/GCC/master/gcc/tree-vect-stmts.cc:13069
0x1287d1f vect_transform_loop_stmt
        /home/mjires/git/GCC/master/gcc/tree-vect-loop.cc:11325
0x12acf13 vect_transform_loop(_loop_vec_info*, gimple*)
        /home/mjires/git/GCC/master/gcc/tree-vect-loop.cc:11770
0x12ee5a1 vect_transform_loops
        /home/mjires/git/GCC/master/gcc/tree-vectorizer.cc:1006
0x12eebac try_vectorize_loop_1
        /home/mjires/git/GCC/master/gcc/tree-vectorizer.cc:1152
0x12eebac try_vectorize_loop
        /home/mjires/git/GCC/master/gcc/tree-vectorizer.cc:1182
0x12eef34 execute
        /home/mjires/git/GCC/master/gcc/tree-vectorizer.cc:1298
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.


$ gcc -v
Using built-in specs.
COLLECT_GCC=/home/mjires/built/master/bin/gcc
COLLECT_LTO_WRAPPER=/home/mjires/built/master/libexec/gcc/x86_64-pc-linux-gnu/14.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /home/mjires/git/GCC/master/configure
--prefix=/home/mjires/built/master --disable-bootstrap
--enable-languages=c,c++,fortran,lto --disable-multilib --disable-libsanitizer
--enable-checking
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.0 20231109 (experimental) (GCC)

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

--- Comment #1 from Robin Dapp <rdapp at gcc dot gnu.org> ---
We fail at:

void
vect_finish_replace_stmt (vec_info *vinfo,
                          stmt_vec_info stmt_info, gimple *vec_stmt)
{
  gimple *scalar_stmt = vect_orig_stmt (stmt_info)->stmt;
  gcc_assert (gimple_get_lhs (scalar_stmt) == gimple_get_lhs (vec_stmt));

where

scalar_stmt = _ifc__40 = .COND_ADD (_22, err_lsm.9_10, 1, err_lsm.9_10);
and
patt_7 = stmp_patt_7.23_123 + stmp_patt_7.23_124;

It happens when we expand the reduction into separate unconditional statements.

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

--- Comment #2 from Robin Dapp <rdapp at gcc dot gnu.org> ---
I tested

diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
index a544bc9b059..257fd40793e 100644
--- a/gcc/tree-vect-loop.cc
+++ b/gcc/tree-vect-loop.cc
@@ -7084,7 +7084,7 @@ vectorize_fold_left_reduction (loop_vec_info loop_vinfo,
                                         opmask, &vec_opmask);
     }

-  gimple *sdef = scalar_dest_def_info->stmt;
+  gimple *sdef = vect_orig_stmt (scalar_dest_def_info)->stmt;

overnight on x86 and aarch64.  This seems to help for the reduced testcase.

It appears as if we "look through" a statement pattern in
vect_finish_replace_stmt but not before when we replace the newly created
vector statement's lhs.  Then the lhs is the statement pattern's lhs while in
vect_finish_replace_stmt we assert that it's from the statement the pattern
replaced.

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-11-10
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Robin Dapp from comment #2)
> I tested
> 
> diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> index a544bc9b059..257fd40793e 100644
> --- a/gcc/tree-vect-loop.cc
> +++ b/gcc/tree-vect-loop.cc
> @@ -7084,7 +7084,7 @@ vectorize_fold_left_reduction (loop_vec_info
> loop_vinfo,
>                                          opmask, &vec_opmask);
>      }
>  
> -  gimple *sdef = scalar_dest_def_info->stmt;
> +  gimple *sdef = vect_orig_stmt (scalar_dest_def_info)->stmt;
> 
> overnight on x86 and aarch64.  This seems to help for the reduced testcase.
> 
> It appears as if we "look through" a statement pattern in
> vect_finish_replace_stmt but not before when we replace the newly created
> vector statement's lhs.  Then the lhs is the statement pattern's lhs while
> in vect_finish_replace_stmt we assert that it's from the statement the
> pattern replaced.

Hmm, that will only work if the orig stmt is the root of the pattern
sequence.  But otherwise it looks correct to me.

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

--- Comment #4 from Robin Dapp <rdapp at gcc dot gnu.org> ---
Is there another way to make it more robust?

Or does the existing

void
vect_finish_replace_stmt (vec_info *vinfo,
                          stmt_vec_info stmt_info, gimple *vec_stmt)
{
  gimple *scalar_stmt = vect_orig_stmt (stmt_info)->stmt;

have the same problem?

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Robin Dapp from comment #4)
> Is there another way to make it more robust?
> 
> Or does the existing
> 
> void
> vect_finish_replace_stmt (vec_info *vinfo,
> 			  stmt_vec_info stmt_info, gimple *vec_stmt)
> {
>   gimple *scalar_stmt = vect_orig_stmt (stmt_info)->stmt;
> 
> have the same problem?

Well, yeah, it also implicitly assumes that.  I couldn't come up with
the "correct" assert quickly, so I didn't suggest any.  That is, just
go with your patch.

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

--- Comment #6 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:d7b971cf61ae0091dbc777c9aa056e6f1293da2f

commit r14-5322-gd7b971cf61ae0091dbc777c9aa056e6f1293da2f
Author: Robin Dapp <rdapp@ventanamicro.com>
Date:   Fri Nov 10 08:56:18 2023 +0100

    vect: Look through pattern stmt in fold_left_reduction.

    It appears as if we "look through" a statement pattern in
    vect_finish_replace_stmt but not before when we replace the newly
    created vector statement's lhs.  Then the lhs is the statement pattern's
    lhs while in vect_finish_replace_stmt we assert that it's from the
    statement the pattern replaced.

    This patch uses vect_orig_stmt on the scalar destination's definition so
    the replaced statement is used everywhere.

    gcc/ChangeLog:

            PR tree-optimization/112464

            * tree-vect-loop.cc (vectorize_fold_left_reduction): Use
            vect_orig_stmt on scalar_dest_def_info.

    gcc/testsuite/ChangeLog:

            * gcc.target/i386/pr112464.c: New test.

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

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

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, can this be closed as fixed?

[Bug tree-optimization/112464] [14 Regression] ICE avx512 with -ftrapv since r14-5076

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112464

Robin Dapp <rdapp at gcc dot gnu.org> changed:

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

--- Comment #8 from Robin Dapp <rdapp at gcc dot gnu.org> ---
Fixed.