[Bug target/38134] [4.4/4.5/4.6/4.7 Regression] speed regression with many loop invariants

[pinskia at gcc dot gnu.org]

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

--- Comment #16 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-01-05 00:35:27 UTC ---
(In reply to comment #6)
> OTOH, loop optimizer should detect invariant loads and move them out of the
> loop.
Just for someone else looking into this bug and thinking the last part of
comment #6 was answered:
It does move them out of the loop.  Then the RA comes along and moves them back
into the loop because of register pressure.