[Bug tree-optimization/64700] Sink common code through PHI

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64700

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

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

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
  <bb 7> [local count: 958878293]:
  if (dir_lsm.26_39 == 1)
    goto <bb 8>; [34.00%]
  else
    goto <bb 9>; [66.00%]

  <bb 8> [local count: 326018623]:
  _11 = arr1.6_10 + _5;
  _12 = *_11;
  goto <bb 10>; [100.00%]

  <bb 9> [local count: 632859670]:
  _14 = arr2.9_13 + _5;
  _15 = *_14;

  <bb 10> [local count: 958878293]:
  # cstore_45 = PHI <_12(8), _15(9)>

PHI-OPT improvements that I am working on might be able to handle this case
too.
MEM_EXPR is a little harder than the normal expression as you need to check the
access type for aliasing reasons.

Note PHI-OPT does handle the casting case already (but it is not listed here).