[Bug target/38134] [4.6/4.7/4.8 Regression] speed regression with many loop invariants

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38134

--- Comment #21 from Steven Bosscher <steven at gcc dot gnu.org> 2013-03-13 20:09:42 UTC ---
(In reply to comment #20)
> Ok, the goal would be to have all !targetm.legitimate_constant_p () 
> constants assigned to a pseudo (and in GIMPLE to an SSA name).

I'm not sure this would be a very useful thing to do.  What is a legitimate
constant for one insn may be non-legit for another.  You may end up doing
more harm than good...