[Bug tree-optimization/51879] Missed tail merging with non-const/pure calls

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51879

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-18 10:45:01 UTC ---
(In reply to comment #2)
> I'm afraid a lot would break.  It really depends on what you use VN for and on
> what code.
> If you have:
>   D.12345_1 = bar (7);
>   D.12346_2 = bar (7);
> and bar isn't const/pure call, then if VN equivalences D.12345_1 and D.12346_2,
> it is wrong.  Of course if you have:
> <bb7>:
>   D.12345_1 = bar (7);
>   goto bb9;
> <bb8>:
>   D.12346_2 = bar (7);
> <bb9>:
>   D.12347_3 = PHI <D.12345_1(7), D.12346_2(8)>
> (this case), you could VN them the same.

Yes, but not for

  D.12345_1 = bar (7);
  D.12346_2 = bar (7);

so you can't really value-number the calls the same.  As we are working
on SSA SCCs and not on a CFG (and thus do not do predicated value-numbering)
that ability is useless.