Re: [abc] abc release

On Wed, 25 Jul 2007, Eric Bodden wrote:

>> Another issue is the fact that our cflow results are invalid
>> for the ants benchmark due to a bug fix in Paddle. We're looking
>> into this with Ondrej (who is visiting from tomorrow), so perhaps
>> we can wait for the outcome of that too. It's pretty embarassing
>> to have published results that look good because of a bug :-(
>
> I see. Do you think you can make it by Friday next week? (My talk is
> on Friday.) Probably in that case Ondrej could do the Soot release,
> too. I would have done it myself but I don't know if it's gonna be
> possible from a potentially flaky internet connection in Berlin (plus
> from my experience I think I won't have much time during ECOOP).

Uh, I cannot speak for Ondrej, obviously, who is on his way to Oxford
and then ECOOP right now. We'll try to be ready with the abc stuff,
and if we have to admit the cflow experiments were flawed, that's just
the way it is.

>> Finally, we'd be very grateful if you could use the latest build
>> from cvs for numbers in your ECOOP presentation, because of the
>> big improvements in tm codegen.
>
> I did new runs just yesterday exactly for that reason. For some
> benchmarks, the values look indeed a quite a bit better. The only
> benchmark that still refuses to run reasonably fast (even after all
> our optimizations) is bloat/FailSafeIter.

Great, thanks. When you apply the gold standard AspectJ version of
FailSafeIter, does it run fast? If so, stay tuned for Julian's
ITD optimisation...

On irc, I was asking earlier today whether you could give some advice
on how to run your optimisations on our benchmarks, so we can
include some numbers when discussing your papers in related work.
Is it tough to run the optimisations? Most of our benchmarks are
multi-threaded, so I guess the POPL stuff is not safe to apply?

-O
Received on Wed Jul 25 2007 - 13:48:19 BST