[Bug tree-optimization/102138] New: t = a==0 and a = PHI<0, t> should be done earlier than PRE

[Bug tree-optimization/102138] New: t = a==0 and a = PHI<0, t> should be done earlier than PRE

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102138

            Bug ID: 102138
           Summary: t = a==0 and a = PHI<0, t> should be done earlier than
                    PRE
           Product: gcc
           Version: 12.0
            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:
int len(int f, int l) {
 return l < f ? 0 : l - f + 1;
}

int lenzero(int f, int l) {
 return len(f, l) == 0;
}

int g(); int h();
int lenzero1(int f, int l) {
 if ( len(f, l) == 0)
   return g();
  return h();
}

we should be able to optimize lenzero at -O1 but don't.  lenzero1 is optimized
at -O1 because dom is able to jump thread.
The difference in IR is:
  # iftmp.0_9 = PHI <0(2), iftmp.0_8(3)>
  _2 = iftmp.0_9 == 0;

vs:
  # iftmp.0_13 = PHI <0(2), iftmp.0_12(3)>
  if (iftmp.0_13 == 0)

at -O2 PRE is enabled which does the optimization.
This is similar to spaceship_replacement in tree-ssa-phiopt but a more general
statement of the problem.

[Bug tree-optimization/102138] t = a==0 and a = PHI<0, t> should be done earlier than PRE

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102138

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
that is we should be able to optimize lenzero at -O1.

[Bug tree-optimization/102138] t = a==0 and a = PHI<0, t> should be done earlier than PRE

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102138

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-08-31
             Status|UNCONFIRMED                 |NEW

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  I attached some propagation through PHI folding for forwprop (WIP)
to the spaceship PR.  There's also tree-ssa-phiprop.c which does sth like
PHI-translation for loads that could be generalized.

In general it opens the same issues as PRE - possibly extra copies due to the
constants in PHIs and possibly less optimized code because readily available
constants (in registers) are explicitely represented and need to be
re-materialized.

[Bug tree-optimization/102138] t = a==0 and a = PHI<0, t> should be done earlier than PRE

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102138

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
mine.

[Bug tree-optimization/102138] t = a==0 and a = PHI<0, t> should be done earlier than PRE

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102138

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is the generic solution to what was done to fix PR 104639.