[patch] ppc prologue scheduling/ -fPIC and -msched-prolog

[Geoff Keating]

This patch is OK, subject to Richard's suggested changes.

> Geoff, what happened to the method you and I discussed in which
> no scanning was necessary at all? 

I'm not now so sure about this.  It makes the code nicer, but
slower...

I'm strongly tempted to simply add the relocations needed.  Then it
would be as easy as

	bl	.LCFx
.LCFx:
	mflr	t1
	addis	t2,t1,(.LCTOC0 - .LCFx)@ha
	addi	toc,t2,(.LCTOC0 - .LCFx)@lo

which is faster, smaller, and generally nicer in every way than the
alternatives.

-- 
- Geoffrey Keating <geoffk@cygnus.com>