Subject: Re: code optimization relevant?
From: Jonathan Aldrich (jonal@cs.washington.edu)
Date: Wed Mar 14 2001 - 12:07:17 PST
> I was pointed to this talk by Bill Pugh by Mike Hind at IBM yesterday.
>
> http://www.cs.umd.edu/~pugh/IsCodeOptimizationRelevant.pdf
I think there's a lot of truth in this talk. As his example, Pugh
discusses my own previous research area, eliminating unnecessary
synchronization, saying we've gone about it all wrong. I basically
agree...one of the key themes of my papers in the area is that the
analyses are most useful to enable better concurrent language designs
(such as one where every object is synchronized). I get the feeling that
no-one was listening, they were all more impressed with the small (in most
cases) improvements we showed on small benchmarks. Other papers covering
the same problem didn't even mention other programming paradigms.
In retrospect, a better way to have gone about this research would be to
think hard about what the right model for synchronization should be, then
do the language design work, then show how to compile the language
efficiently. The resulting compilation techniques are likely to be more
innovative, and result in much larger bottom-line speedups. Yes, it's
great to have done a compilers paper that shows small but useful speedups
on real, existing programs...but publishing a paper that could improve the
way people write programs is a bigger deal in the long run.
I think there's a lot of good things about the way we're doing
research--coming up with new frameworks that make it easier to write good
and correct compilers, solving representation problems so people don't
have to use nasty object models to get performance, and doing langauge
designs that support more powerful and orthogonal features, from EML and
multimethods to advanced type systems and software architecture. These
are all important things that go beyond the next 5% in compiler speedups.
It's always good think of how to do better, though.
Jonathan :-)
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Wed Mar 14 2001 - 12:08:04 PST