Re: [boost] A possible GSOC 2011 proposal idea

Subject: Re: [boost] A possible GSOC 2011 proposal idea
From: Dave Abrahams (dave_at_[hidden])
Date: 2011-03-21 13:45:09

On Mon, Mar 21, 2011 at 7:02 AM, Mathias Gaunard
<mathias.gaunard_at_[hidden]> wrote:
> On 21/03/2011 14:50, Chad Seibert wrote:
>>
>> Yes, it does. However, it seems inappropriate to write an LP library for
>> Boost that uses LAPACK to do the linear algebra stuff. If we wanted to go
>> down that road, we could just write a wrapper for an existing lp library
>> instead, like what OpenOpt does.
>
> LAPACK is a common interface for which there exists many optimized
> implementations from various vendors. Those are of very high quality, you
> are simply never going to beat them in terms of performance, even if you are
> a numerical computation expert.
> There are also open-source implementations available with liberal licenses.
>
> OpenOpt, on the other hand, is a large framework, and albeit it has a
> liberal license, there is still interest for alternatives.
> While it is good, it's not quite as good as it could be.
>
> But I don't think rewriting the LAPACK routines you need makes any sense at
> all.  It's really state of the art software.

Typically what you do with a generic library is that you write generic
algorithms for arbitrary datatypes and then specialize them to
dispatch to LAPACK when the types can be handled directly.  That way
you can still get a reasonably-performant matrix-of-matrices or
matrix-of-rationals and get screamin'-hot performance for floats and
doubles.

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