Re: [abc] paper updated

Prof. Laurie HENDREN wrote:

>Pavel, are there any updated performance numbers? I remember that you
>were working on a further improvement, did that work out. Also,
>Jennifer sent e-mail a bit back saying that the newer version of Soot
>fixes some j2j inefficiencies.
>
>
I have been struggling with the optimisations me and Oege mentioned in
the last two mails about 'simplifying codegen for tracematches' -- the
whole thing to be very frustrating, with both of them seemingly
increasing both time and memory usage rather than reducing it. After
much nail-biting I am now leaning to the hypothesis that my method of
taking measurement skewes the results. Both optimisations have been
implemented for some time now, but I'm fiddling with them trying to make
the benefits more convincing. Just half an hour ago I measured a 25%
speedup (for the first time ever ;-) ). So, while there are no updated
numbers, there may be some as soon as I've finally figured out what's
going on.

I'll have to remind myself how to update soot (and hope it breaks
nothing), as those fixes sound rather promising.

>When we think we are not adding any more, we should make a complete
>read through.
>
>Who is submitting the final version? Will Oege be back by then?
>
>
It was my impression that Oege would do it, although he is strictly
speaking still on holiday on the 12th -- it would be nice if he could
confirm or deny that. :-)

- P
Received on Mon Aug 8 13:42:14 2005