[Bug tree-optimization/104018] New: Comparison against 0 propagates into other statement causing no-CSE from happening

[Bug tree-optimization/104018] New: Comparison against 0 propagates into other statement causing no-CSE from happening

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104018

            Bug ID: 104018
           Summary: Comparison against 0 propagates into other statement
                    causing no-CSE from happening
           Product: gcc
           Version: 12.0
            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:
int f(unsigned int a, unsigned int b) {
    if (a > 0) {
        return a == b;
    } else {
        return a == b;
    }
}
int f2(unsigned int a, unsigned int b) {
    if (a > 1) {
        return a == b;
    } else {
        return a == b;
    }
}

They should produce the same results but don't currently as in the first case
the >0 is turned into != 0 and then the 0 is propagated into the comparison and
then the comparison in the other BBSs are not able to be CSE'ed with the other
one.

[Bug tree-optimization/104018] Comparison against 0 propagates into other statement causing no-CSE from happening

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104018

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |12.0, 4.1.2
           Severity|normal                      |enhancement

[Bug tree-optimization/104018] Comparison against 0 propagates into other statement causing no-CSE from happening

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104018

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-01-14
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The "unfortunate" thing is that FRE1 figures out they are equal but there's
no SSA name available with the value to replace the PHI node with ... it needs
PRE hoisting to apply the CSE which is too late.

It's DOM doing the equivalence propagation btw. (I think DOM is even the only
pass doing this).