[Bug tree-optimization/115579] New: wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu

[Bug tree-optimization/115579] New: wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu

[zhendong.su at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

            Bug ID: 115579
           Summary: wrong code at -Os with "-fno-tree-sra" on
                    x86_64-linux-gnu
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zhendong.su at inf dot ethz.ch
  Target Milestone: ---

It appears to be a very recent regression as it doesn't reproduce with 14.1. 

Compiler Explorer: https://godbolt.org/z/8qsq5zbeE


[572] % gcctk -v
Using built-in specs.
COLLECT_GCC=gcctk
COLLECT_LTO_WRAPPER=/local/suz-local/software/local/gcc-trunk/libexec/gcc/x86_64-pc-linux-gnu/15.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../gcc-trunk/configure --disable-bootstrap
--enable-checking=yes --prefix=/local/suz-local/software/local/gcc-trunk
--enable-sanitizers --enable-languages=c,c++ --disable-werror --enable-multilib
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 15.0.0 20240620 (experimental) (GCC) 
[573] % 
[573] % gcctk -Os small.c; ./a.out
[574] % 
[574] % gcctk -Os -fno-tree-sra small.c
[575] % ./a.out
Aborted
[576] % 
[576] % cat small.c
int printf(const char *, ...);
int a, b = 1, c;
int main() {
  int d[2], *e = &d[1];
  while (a) {
    int *f = &b;
    d[1] = 0;
    *f = 0;
  }
  if (c)
    printf("%d\n", *e);
  if (b != 1)
    __builtin_abort();
  return 0;
}

[Bug tree-optimization/115579] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- Comment #1 from Sam James <sjames at gcc dot gnu.org> ---
*** Bug 115578 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|wrong code at -Os with      |[15 regression] wrong code
                   |"-fno-tree-sra" on          |at -Os with "-fno-tree-sra"
                   |x86_64-linux-gnu            |on x86_64-linux-gnu
            Version|unknown                     |15.0
   Target Milestone|---                         |15.0
                 CC|                            |sjames at gcc dot gnu.org
           Keywords|                            |wrong-code

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu.org
            Summary|[15 regression] wrong code  |[15 regression] wrong code
                   |at -Os with "-fno-tree-sra" |at -Os with "-fno-tree-sra"
                   |on x86_64-linux-gnu         |on x86_64-linux-gnu since
                   |                            |r15-1391-g4b75ed33fa5fd6

--- Comment #2 from Sam James <sjames at gcc dot gnu.org> ---
very recent indeed: r15-1391-g4b75ed33fa5fd6

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[zhendong.su at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- Comment #3 from Zhendong Su <zhendong.su at inf dot ethz.ch> ---
Perhaps the same or a related issue.  It reproduces for -O{s,2,3}. 

Compiler Explorer: https://godbolt.org/z/x779P9b5a

[570] % gcctk -v
Using built-in specs.
COLLECT_GCC=gcctk
COLLECT_LTO_WRAPPER=/local/suz-local/software/local/gcc-trunk/libexec/gcc/x86_64-pc-linux-gnu/15.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../gcc-trunk/configure --disable-bootstrap
--enable-checking=yes --prefix=/local/suz-local/software/local/gcc-trunk
--enable-sanitizers --enable-languages=c,c++ --disable-werror --enable-multilib
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 15.0.0 20240620 (experimental) (GCC) 
[571] % 
[571] % gcctk -O2 small.c; ./a.out
[572] % 
[572] % gcctk -O2 -fno-tree-fre -fno-tree-sra -fno-tree-ch -fno-tree-dce
small.c
[573] % ./a.out
Floating point exception
[574] % 
[574] % cat small.c
int a = 1;
int main() {
  int b[1], c;
  for (c = 0; c < 2; c++) {
    int d = a, e = 1 / d;
    while (!e) {
      b[0] = 0;
      while (b[0])
        ;
      a = 0;
    }
  }
  return 0;
}

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-06-21
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed. LIM is doing it wrong:
```
  <bb 2> [local count: 118111600]:
  a.0_1 = a;
...
  b_lsm.10_17 = _16(D);
  goto <bb 4>; [100.00%]
...
  <bb 4> [local count: 1073741824]:
  # d_I_lsm.9_13 = PHI <d_I_lsm.9_15(2), d_I_lsm.9_18(3)>
  # b_lsm.10_14 = PHI <b_lsm.10_17(2), b_lsm.10_19(3)>
  if (a.0_1 != 0)
    goto <bb 3>; [89.00%]
  else
    goto <bb 5>; [11.00%]

  <bb 5> [local count: 118111600]:
  # d_I_lsm.9_23 = PHI <d_I_lsm.9_13(4)>
  # b_lsm.10_24 = PHI <b_lsm.10_14(4)>
  d[1] = d_I_lsm.9_23;
  b = b_lsm.10_24;
...
```

Why was the store to b an unconditional one here?



Before LIM:
```
  <bb 2> [local count: 118111600]:
  goto <bb 4>; [100.00%]

  <bb 3> [local count: 955630224]:
  d[1] = 0;
  b = 0;

  <bb 4> [local count: 1073741824]:
  a.0_1 = a;
  if (a.0_1 != 0)
    goto <bb 3>; [89.00%]
  else
    goto <bb 5>; [11.00%]
```

It looks like it is pulling out d[1] of the loop unconditionally and that also
causes b to be done unconditionally ...

So -Os is needed because we can't see the loop is an infinite one too ...

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Zhendong Su from comment #3)
> Perhaps the same or a related issue.  It reproduces for -O{s,2,3}. 

It is the same due to -fno-tree-ch . -Os has a very restrictive version of
that.

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So looking into the code. The problem is the ordering.
That is if the store to b is first things will just work.

The loop to call execute_sm:
          execute_sm (loop, ref, aux_map, true, !first_p);

Except use_other_flag_var is wrong if the first one didn't have a store flag.

  if (multi_threaded_model_p && !use_other_flag_var)
    aux->store_flag
      = execute_sm_if_changed_flag_set (loop, ref, &aux->flag_bbs);
  else
    aux->store_flag = NULL_TREE;

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Mine.  This is probably a latent issue.

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ah, ok - the situation is new in that store-flag re-use is only considered for
loops which have all stores in a single BB and of course there always_stored
is always the same but now ref_can_have_store_data_races can be different for
stores.

I have a fix, it's to some extent not optimal as we emit unnecessary loads
up to the first var that needs the MT model.

There's the general optimization opportunity missing to share flag vars
for stores that are stored in the same BBs.

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

--- 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:8a1795bddcd34284936af4706f762d89c60fc69c

commit r15-1564-g8a1795bddcd34284936af4706f762d89c60fc69c
Author: Richard Biener <rguenther@suse.de>
Date:   Sat Jun 22 14:59:09 2024 +0200

    tree-optimization/115579 - fix wrong code with store-motion

    The recent change to relax store motion for variables that cannot have
    store data races broke the optimization to share flag vars for stores
    that all happen in the same single BB.  The following fixes this.

            PR tree-optimization/115579
            * tree-ssa-loop-im.cc (execute_sm): Return the auxiliary data
            created.
            (hoist_memory_references): Record the flag var that's eventually
            created and re-use it when all stores are in the same BB.

            * gcc.dg/pr115579.c: New testcase.

[Bug tree-optimization/115579] [15 regression] wrong code at -Os with "-fno-tree-sra" on x86_64-linux-gnu since r15-1391-g4b75ed33fa5fd6

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115579

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> ---
Fixed.