[MLton] genetic algorithms for SSA optimization phase ordering

Matthew Fluet fluet@cs.cornell.edu
Wed, 10 Dec 2003 09:36:13 -0500 (EST)


> There were a couple of what looked to be typos.  I mention them to
> make sure they aren't also in the code.

They were just typos in my writeup.

> Only two simple ideas: emphasize run time and emphasize larger
> benchmarks.

Absolutely.  But, have we moved to the point where size doesn't matter?
Are you willing to have a 2X executable for a 1.3X speedup?

> It might be nice to keep track of the chromosome that produced the
> minimum (run time, say) for each benchmark.  It would be interesting
> to see how that compares to what the baseline chromosome produces, as
> well as to any better chromosome you find.

Yes.  I need to improve the output.  I'd also like to set things up so
that I can save complete chromosome information and read it back in to
resume an interrupted evolution.