x86 backend update

[Matthew Fluet]

> I tried to do a self compile with the new native backend, and it seems
> at the very least to be a lot slower, at the worst, nonterminating. 
> It's been running for 2.5 hours, and is still sitting in the backend. 
> This is a G0 building a G1, so asserts will be turned on.  Do you think
> asserts are prohibitively slow for a self compile?  Any other ideas? 
> Here is the transcript. 

Steve, did that self-compile ever finish?  I thought I had recreated the
problem: the native backend spit out 8 .s files then didn't add anything
to the last one for a long time.  So, I added some Control.messageStr
printing to try to pinpoint where it wasn't making progress, but that
self-compile finished; unfortunately, I didn't timed it, because I assumed
it wasn't going to complete.  But it was < 5 hours on a PentiumII 266MHz.

The only issue with the asserts might be if some basic block was very,
very large.  That's never occured before, except in the initGlobals block,
but I special case that.  The other oddity is that it is occuring at
something like the very last block of the program (because the last
self-compile I did a couple of days ago needed 8 .s files).  Very strange.