x86-codegen bug

[Stephen Weeks]

Matthew, there is a bug in the x86-codegen that you sent yesterday.  It causes
segfaults on many of the regressions (date, fft, kitqsort, real, word, word8,
world6).  If I use the C codegen or back out to the x86-codegen from Jan 15,
there are no problems with any of the regressions (other than the usual real
problem).

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

It has the Jan 15 codegen built in.  You can switch to 
	src/mlton/codegen/x86-codegen-new
to use the buggy codegen.