Update: reload register allocation patch

[Bernd Schmidt]

On Tue, 23 Nov 1999, Joern Rennecke wrote:
> However, when I look at the code that sets spill_cost and spill_add_cost,
> I see that you don't take the width of the spilled register into account.
> The more hard register are allocated to a pseudo, the more expensive it
> is to spill it, since usually you have to reload all the hard registers
> of the pseudo.

Good point.  I don't think the current code takes this into account either,
though.

> Or does it turn out to be advantagous to be a little more mean to multi-
> hard-reg pseudos because that frees up multiple spill registers at once?

No idea.  I'll experiment a little.  The obvious way to adjust the cost would
be to multiply by nregs; do you have any other ideas how to tune it?

Bernd