https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126294

            Bug ID: 126294
           Summary: a*3/2 -> a + a/2 if a*3 is known not to overflow/wrap
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
unsigned f(unsigned short b)
{
  unsigned t = b;
  return t*3/2;
}
```

This could be done as `t+t/2` as we know t*3 does not overflow/wrap around.
LLVM does this optimization.

The same is true for signed (which LLVM does not do currently).

Reply via email to