[Bug tree-optimization/115602] New: [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643

[Bug tree-optimization/115602] New: [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

            Bug ID: 115602
           Summary: [15 Regression]  ICE on liblapack-3.12.0: in
                    vect_schedule_slp_node, at tree-vect-slp.cc:9643
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: slyfox at gcc dot gnu.org
  Target Milestone: ---

Possibly recent regression observed on liblapack-3.12.0 on
r15-1566-gfd536b8412d4da.

Reduced down to the following:

* $ cat ztpmv.f.f
      SUBROUTINE ZTPMV_64(AP,X)
      COMPLEX*16 AP(*),X(*)
      COMPLEX*16 TEMP
      LOGICAL NOCONJ
      LOGICAL LSAME_64
      IF (LSAME_64(UPLO,'U') .AND.LSAME_64(UPLO,'L')) THEN
      ELSE IF (LSAME_64(TRANS,'N') .AND.LSAME_64(DIAG,'N')) THEN
          IF (LSAME_64(UPLO,'U')) THEN
              IF (INCXINCXINCX.EQ.1) THEN
                      TEMP = X(JX)
                      IF (NOCONJ) THEN
                          TEMP = TEMP*AP(KK)
                      ELSE
                          TEMP = TEMP*DCONJG0
                          DO 130 K = 1,J
                              TEMP = TEMP + AP(K)*X(IX)
  130                     CONTINUE
                      END IF
                      X(JX) = TEMP
                      END IF
          END IF
      END IF
      END

Crashing:

$ gfortran -O2 -frecursive -fdefault-integer-8 -c ztpmv.f.f -o bug.o -Wall
during GIMPLE pass: slp
ztpmv.f.f:1:25:

    1 |       SUBROUTINE ZTPMV_64(AP,X)
      |                         ^
internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9643
0x1f2a3be diagnostic_impl(rich_location*, diagnostic_metadata const*, int, char
const*, __va_list_tag (*) [1], diagnostic_t)
        ???:0
0x1f2aa98 internal_error(char const*, ...)
        ???:0
0x767124 fancy_abort(char const*, int, char const*)
        ???:0
0x74b0c3 vect_schedule_slp_node(vec_info*, _slp_tree*, _slp_instance*) [clone
.part.0] [clone .cold]
        ???:0
0x10d34f2 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d2989 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d2989 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d2989 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d2989 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d2989 vect_schedule_scc(vec_info*, _slp_tree*, _slp_instance*,
hash_map<_slp_tree*, slp_scc_info,
simple_hashmap_traits<default_hash_traits<_slp_tree*>, slp_scc_info> >&, int&,
vec<_slp_tree*, va_heap, vl_ptr>&)
        ???:0
0x10d39f0 vect_schedule_slp(vec_info*, vec<_slp_instance*, va_heap, vl_ptr>
const&)
        ???:0
0x10d61ca vect_slp_bbs(vec<basic_block_def*, va_heap, vl_ptr> const&, loop*)
        ???:0
0x10d7675 vect_slp_function(function*)
        ???:0
0x10deef5 (anonymous namespace)::pass_slp_vectorize::execute(function*)
        ???:0

$ gfortran -v
Using built-in specs.
COLLECT_GCC=/<<NIX>>/gfortran-15.0.0/bin/gfortran
COLLECT_LTO_WRAPPER=/<<NIX>>/gfortran-15.0.0/libexec/gcc/x86_64-unknown-linux-gnu/15.0.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../source/configure --prefix=/<<NIX>>/gfortran-15.0.0
--with-gmp-include=/<<NIX>>/gmp-6.3.0-dev/include
--with-gmp-lib=/<<NIX>>/gmp-6.3.0/lib
--with-mpfr-include=/<<NIX>>/mpfr-4.2.1-dev/include
--with-mpfr-lib=/<<NIX>>/mpfr-4.2.1/lib --with-mpc=/<<NIX>>/libmpc-1.3.1
--with-native-system-header-dir=/<<NIX>>/glibc-2.39-52-dev/include
--with-build-sysroot=/
--with-gxx-include-dir=/<<NIX>>/gfortran-15.0.0/include/c++/15.0.0/
--program-prefix= --enable-lto --disable-libstdcxx-pch
--without-included-gettext --with-system-zlib --enable-checking=release
--enable-static --enable-languages=fortran --disable-multilib --enable-plugin
--disable-libcc1 --with-isl=/<<NIX>>/isl-0.20 --build=x86_64-unknown-linux-gnu
--host=x86_64-unknown-linux-gnu --target=x86_64-unknown-linux-gnu
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 15.0.0 99999999 (experimental) (GCC)

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #1 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
Created attachment 58501
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58501&action=edit
ztpmv.f.f

Also attaching unreduced 'ztpmv.f.f' in case cvise mangled it too much.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |15.0

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-unknown-linux-gnu

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It worked at 00848ba8d1602c2e3ff9fb1cb40f232b4c2b2452 . So maybe the fix for PR
115597  .

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #3 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
bisect landed at r15-1565-g2a345214fc332b

commit 2a345214fc332b6f0821edf394ff8802b768db1d (HEAD)
Author: Richard Biener <rguenther@suse.de>
Date:   Sun Jun 23 11:26:39 2024 +0200

    tree-optimization/115597 - allow CSE of two-operator VEC_PERM nodes

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #4 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
If C example is easier to debug graphviz-10.0.1 exhibits the same ICE.
Minimized example:

// $ cat gvevent.c.c
typedef struct {
  double x, y;
} pointf;
struct {
  pointf focus;
  double zoom;
  pointf devscale;
  char button;
  pointf oldpointer;
} gvevent_motion_job;
char gvevent_motion_job_4;
double gvevent_motion_pointer_1, gvevent_motion_pointer_0;
void gvevent_motion() {
  double dx = (gvevent_motion_pointer_0 - gvevent_motion_job.oldpointer.x) /
              gvevent_motion_job.devscale.x,
         dy = (gvevent_motion_pointer_1 - gvevent_motion_job.oldpointer.y) /
              gvevent_motion_job.devscale.y;
  if (dx && dy < .0001)
    return;
  switch (gvevent_motion_job_4)
  case 2: {
    gvevent_motion_job.focus.x -= dy / gvevent_motion_job.zoom;
    gvevent_motion_job.focus.y += dx / gvevent_motion_job.zoom;
  }
}

Crashing:

$ gcc -O2 -c gvevent.c.c -o bug.o
during GIMPLE pass: slp
gvevent.c.c: In function 'gvevent_motion':
gvevent.c.c:13:6: internal compiler error: in vect_schedule_slp_node, at
tree-vect-slp.cc:9643
   13 | void gvevent_motion() {
      |      ^~~~~~~~~~~~~~

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #5 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
Created attachment 58502
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58502&action=edit
gvevent.c.c

Attaching unreduced gvevent.c.c in case cvise mangled it too much.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-06-24

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Mine.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rsandifo at gcc dot gnu.org
           Keywords|                            |missed-optimization

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ah, interesting - somehow we managed to create a self-referencing cycle!?

t.c:13:6: note: node 0x4ab34e0 (max_nunits=1, refcnt=3) vector(2) double
t.c:13:6: note: op: VEC_PERM_EXPR
t.c:13:6: note:         stmt 0 _11 = gvevent_motion_job.zoom;
t.c:13:6: note:         stmt 1 _11 = gvevent_motion_job.zoom;
t.c:13:6: note:         lane permutation { 0[1] 0[0] }
t.c:13:6: note:         children 0x4ab34e0

that's because this permute is the same as the load that was originally
feeding it:

t.c:13:6: note: node 0x4ab3690 (max_nunits=2, refcnt=1) vector(2) double
t.c:13:6: note: op template: _11 = gvevent_motion_job.zoom;
t.c:13:6: note:         stmt 0 _11 = gvevent_motion_job.zoom;
t.c:13:6: note:         stmt 1 _11 = gvevent_motion_job.zoom;
t.c:13:6: note:         load permutation { 2 2 }

that's a missed optimization caused by the SLP optimize pass which inserts
this permute as compensation.

Richard - can you look where to best see that uniform nodes (through a
uniform load permute) do not require a permute?

I'll see to somehow make CSE robust against this.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

David Binderman <dcb314 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dcb314 at hotmail dot com

--- Comment #8 from David Binderman <dcb314 at hotmail dot com> ---
Created attachment 58504
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=58504&action=edit
C++ source code

A C++ example. -O3 required. From project box2d.

cvise $ ~/gcc/results/bin/gcc -c -O3 bug1039.cc
during GIMPLE pass: slp
bug1039.cc: In function ‘void b2Distance()’:
bug1039.cc:27:6: internal compiler error: in vect_schedule_slp_node, at
tree-vect-slp.cc:9644
   27 | void b2Distance() {
      |      ^~~~~~~~~~
0x14c966b vect_schedule_slp_node(vec_info*, _slp_tree*, _slp_instance*)
        /home/dcb40b/gcc/working/gcc/../../trunk/gcc/tree-vect-slp.cc:9643

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #9 from GCC 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:c43c74f6ec795a586388de7abfdd20a0040f6f16

commit r15-1583-gc43c74f6ec795a586388de7abfdd20a0040f6f16
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Jun 24 09:52:39 2024 +0200

    tree-optimization/115602 - SLP CSE results in cycles

    The following prevents SLP CSE to create new cycles which happened
    because of a 1:1 permute node being present where its child was then
    CSEd to the permute node.  Fixed by making a node only available to
    CSE to after recursing.

            PR tree-optimization/115602
            * tree-vect-slp.cc (vect_cse_slp_nodes): Delay populating the
            bst-map to avoid cycles.

            * gcc.dg/vect/pr115602.c: New testcase.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

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

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

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Should be fixed, but the inefficiency in creating the permute is still present.
I guess I'll create another PR for that.

[Bug tree-optimization/115602] [15 Regression] ICE on liblapack-3.12.0: in vect_schedule_slp_node, at tree-vect-slp.cc:9643 since r15-1565-g2a345214fc332b

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115602

--- Comment #11 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
The change fixed both liblapack-3.12.0 and graphviz-10.0.1 builds for me. Thank
you!