x86 update

[Stephen Weeks]

> I integrated it and ran some tests over the weekend.  The self compile failed
> during the x86 code gen pass due to running out of memory.  Something really bad 
> must have happened, since the max live with the old x86-codegen was about 60M.
> I had made some changes to the closure converter, but I can't imagine how they
> would have caused this.  Anyways, I'll revert back to the old x86-codegen and
> see if I can do a self-compile there.

That self compile finished fine, so it is likely to be a bug in the new codegen
(I suppose we shouldn't rule out the possibility of an SML/NJ bug).

> > Matthew, do you want a snapshot?
> 
> Sure.  If you encounter this some more, you might try changing the
>   val tracer =
> definitions at the top of x86-{translate,simplify,allocate-registers}.fun.
> That will use the original traceCall which might pinpoint where the memory
> explosion is occuring.

I put a snapshot at http://www.star-lab.com/sweeks/mlton.tgz.

I'll start a compile with this change and the new codegen.