[Bug c++/112905] New: [missed optimization] fail to reduce a loop that increments two locations of the same buffer

[Bug c++/112905] New: [missed optimization] fail to reduce a loop that increments two locations of the same buffer

[AceSrc at outlook dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112905

            Bug ID: 112905
           Summary: [missed optimization] fail to reduce a loop that
                    increments two locations of the same buffer
           Product: gcc
           Version: 13.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: AceSrc at outlook dot com
  Target Milestone: ---

Previously, I posted a question in Stackoverflow about loop optimization
(https://stackoverflow.com/questions/77618348/why-does-gcc-fail-to-reduce-a-loop-that-increments-two-locations-of-the-same-buf),
and the answer suggests me to post a [missed optimization] report here.

The question is about GCC does not optimize the following C++ codes into
`counter[A] += n; counter[B] += n;`. In contrast, GCC optimizes the loop into
`counter[A] += n` if `++counter[B]` is deleted. 

```
unsigned int getid();
void foo(unsigned int *counter, unsigned int n) {
        unsigned int A = getid();
        unsigned int B = getid();
        for (unsigned int i = 0; i < n; i++) {
            ++counter[A];
            ++counter[B];
        }
}
```

LLVM has the same behavior.

[Bug tree-optimization/112905] [missed optimization] fail to reduce a loop that increments two locations of the same buffer

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112905

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-12-07
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
          Component|c++                         |tree-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that store-motion cannot cope with this as A could be equal to B
which would invalidate the transform it would do.  And that's the enabler
and why it works with just a single update.

Confirmed.

[Bug tree-optimization/112905] [missed optimization] fail to reduce a loop that increments two locations of the same buffer

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112905

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu.org
           Severity|normal                      |enhancement