https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111763

            Bug ID: 111763
           Summary: `(a & ~1) | 2` could be done as `(a & ~(1 | 2)) + 2`
                    which allows to use leal
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---
            Target: x86_64-linux-gnu

Take:
```
int f1(int in) {
  in = (in & ~(unsigned long)1);
  in = in | 2;
  return in;
}


int f2(int in) {
  in = (in & ~(unsigned long)(1|2));
  in = in + 2;
  return in;
}
```

We currently get:
```
f1:
        movl    %edi, %eax
        andl    $-2, %eax
        orl     $2, %eax
        ret
f2:
        andl    $-4, %edi
        leal    2(%rdi), %eax
        ret
```

The leal version is better because it saves more move due to leal not being a 2
operand but 3 operand instruction so it could improve register allocation ...

I noticed this whole looking into PR 111762 (and PR 111282) and looking at
clang/LLVM's code generation here .

Also I don't know how often this shows up though.

Reply via email to