Boost logo

Boost :

Subject: Re: [boost] [iterator] std::sort on zip_iterator (repost)
From: Edouard A. (edouard_at_[hidden])
Date: 2009-05-10 11:37:34

> Scratch that; as template parameters they shouldn't hurt performance at
> all.
> That's a good idea; the user needs to pass lots of template parameters,
> but
> it's easy to create overloads that pass in defaults.
> Theoretically you should be able to have a version that is introsort,
> with
> identical performance.

Yes, inlining + copy elision should make the performance impact < epsilon.
Last profiling I made on sorting, I don't recall spending too much on
function calls.

> If someone else wants to write it, I'd be willing to maintain it in the
> Boost.Algorithm.Sorting library (assuming it is accepted); not that I'd
> try
> to stop the author from maintaining it if they wanted to.

I can do it. I can add a wrapper to spread sort as well. The idea is to give
as many bricks as possible with some default constructions for the user who
isn't interested in dwelling to much in the details. That's an approach I
like. Simple and satisfactory default behavior with the possibility for
heavy customization.

I'll have some time next week and guess what, I have a new rig to benchmark!

When I returned from Aspen I found on the threshold of my door an abandoned
Core i7 and 12 GB of RAM with a letter saying "take good care of them".

Oh yeah. >)


Boost list run by bdawes at, gregod at, cpdaniel at, john at