[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [help-GIFT] Optimizing

From: risc
Subject: Re: [help-GIFT] Optimizing
Date: Tue, 19 Sep 2006 13:27:51 -0500
User-agent: Mutt/1.4.1i

On Tue, Sep 19, 2006 at 07:00:12PM +0100, David Squire wrote:
> address@hidden wrote:
> > I agree completely. once we're done with your plan, I'll be the first to
> > advocate removing 60*. however, for the time being, we've got X version,
> > and X+1 version thats faster. which would you prefer new users are 
> > running, and judging gift based on?
> >
> > None of the rest of my work involves hard-coded fixed sizes. Rest
> > assured, the rest is much more complex. ;)
> >   
> Good, good.
> > have you read through the 70* and 80* patches? i'm commiting those as 
> > well, and would like a technical commentary.
> >
> >   
> I'm afraid that I have not. I am working on a bunch of other projects at
> present (Damocles, new IR research, etc.) and keep swapping between
> Perl, Java/Tomcat, C++ also. I am finding it hard to keep up :) I hope
> to get to it soon, but I don't think I should be the bottleneck. If the
> feature files you create are binary-identical, then the main issue
> really is ability to compile across a variety of systems.
> BTW, have you looked at asking for aligned memory when allocating
> (*posix_memalign)*? I am told by the people here who need video
> frame-rate processing here that it can give you a big speed-up. It would
> be interesting to hear about.
> Cheers,
> David
> -- 
> Dr. David McG. Squire, Senior Lecturer. On sabbatical in 2006.
> Caulfield School of Information Technology, Monash University, Australia
> CRICOS Provider No. 00008C
I've tried and failed mizerably. :)

I've been told its a major speedup as well, but despite pulling every 
trick i can to align everything, I havent seen any speedups as a result.

At the moment, i'm concentrating on making my 'memory in order' patches
hit the tree, and then we'll be working toward the same goal (a 
single-invoke feature extractor), because I think we'll have more speed
improvements from that than from me continuing to bang my head on 
alignment issues. Its on the TODO, however. :)

Julia Longtin <address@hidden>

reply via email to

[Prev in Thread] Current Thread [Next in Thread]