using mosml as a frontend for MLton

[Stephen Weeks]

> Can you enlighten me on what mosml is?  

Moscow ML (http://www.dina.kvl.dk/~sestoft/mosml.html).  It's been around for a
while, but until very recently, only compiled the core language.  It now claims
to compile SML 97, although as we saw from my earlier mail that should probably
be taken with a grain of salt.

> Also, I spoke with Matthew today about future (short-term) directions.  I
> suggested he look at building a simple profile tool to get some estimate of (a)
> what percentage of expressions in basic blocks refer to common subexpressions,
> and (b) what percentage are loop invariant. I think it'll be relatively easy to
> gather this info (a couple of days max.) and should either number prove
> interesting, it shouldn't take much time to implement during the remainder of
> the time he's here.

Common subexpressions may not be so easy on the X86 IL since you need alias
information (for the same reason gcc does).  The aliasing is more obvious on the
CPS IL.  Unfortunately, not everything will be doable in CPS, since array
addressing computations aren't exposed.  Similar comments apply to loop
invariants.  Probably the right thing to do, for now, is to do as much as we can
on the CPS IL.  Someday, more will be doable on RCPS.

I'd vote for getting doubles in before this, because I'd like a complete
system before most anything else.