IRA improvements 3/4

[Andi Kleen]

Vladimir Makarov <vmakarov@redhat.com> writes:
>>
> Yes, usually I expect that this kind of optimizations (moving
> invariant and removing living pseudo by loop reversal by this case)
> are done by gcc.  But when I check the assembler code, it is sometimes
> not what I expected.  As I remember that was the case for this
> critical loop when I compiled by gcc4.3.

This was fixed some time ago with PR40886. It also worked in gcc 3.x.
Should work again with gcc 4.5.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.