Re: [boost] Accelerating algorithms with SIMD - Segmented iterators and alternatives

Subject: Re: [boost] Accelerating algorithms with SIMD - Segmented iterators and alternatives
From: David Abrahams (dave_at_[hidden])
Date: 2010-10-13 11:07:03

At Wed, 13 Oct 2010 15:52:47 +0100,
Mathias Gaunard wrote:
> 
> On 13/10/10 12:18, Mathias Gaunard wrote:
> 
> > Since those could be different segments, you wouldn't need to check
> > which segment you are in at each iteration.
> > Actually, all local iterators for a a segmented iterator must be the
> > same for all segments, but since a local iterator can be a segmented
> > iterator itself, it can also have a local iterator which can be
> > different. By recursion, you can therefore use different iterator types
> > for each segment, but that's somewhat convoluted.
> 
> That's wrong actually. Since all the local iterators are the same,
> then their local iterators must be the same too.
> 
> So segmented iterators wouldn't help to only check for alignment once
> it has been reached I think. I would have to properly try to implement
> one, but it's not particularly practical to do.

I think the exercise would be worth your time

> Anyway, my belief is that my alternative to segmented iterators is
> more interesting.

Not to me, until you can begin to address something like std::merge().

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com