[Bug tree-optimization/112819] New: rearrange branches to improve code generation

[Bug tree-optimization/112819] New: rearrange branches to improve code generation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112819

            Bug ID: 112819
           Summary: rearrange branches to improve code generation
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
bool src(char *a, int l) {
    if(l < 1) {
        return false;
    }
    if(*a != 'a') {
        return false;
    }
    if(l < 2) {
        return false;
    }
    return true;
}

bool tgt(char *a, int l) {
    if(l < 1) {
        return false;
    }
    if(l < 2) {
        return false;
    }
    if(*a != 'a') {
        return false;
    }
    return true;
}
```
You would expect these 2 to produce both the same code but only tgt produces
decent code. That is because we can merge `l < 1` with `l < 2` into just `l <
2` (or `l <= 1`).

This might be an reassociation problem or something else.

[Bug tree-optimization/112819] rearrange branches to improve code generation

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112819

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
                 CC|                            |fkastl at suse dot cz,
                   |                            |hubicka at gcc dot gnu.org,
                   |                            |rguenth at gcc dot gnu.org
   Last reconfirmed|                            |2023-12-04
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
We don't "re-associate" branches.  But we might want to turn if-to-switch
into sth that does (relaxing the restriction on the ifs testing the same var).

We should have a PHI node with incoming vals predicated by the ifs, free to
re-order otherwise.

The *a deref might impose some limitations for re-ordering, but we can always
handle it last in this case.

Might be also interesting to order branches which are predictable earlier.

Related to switch-conversion/if-to-switch IMHO.