Subject: Re: [boost] different matrix library?
From: DE (satan66613_at_[hidden])
Date: 2009-08-14 14:44:16
on 14.08.2009 at 22:34
joel wrote :
>> you might be right. i just thought 'why i need arbitrary size
>> _matrix_?' it might turn out there is no such cases
> There is. Small matrix like rotation matrix or 3D vector can benefit
> from a total unrolling of their evaluation.
> Knowing size of matrix at compile-time opens for a lot of such
> optimization (tiling, prefethcing, unrolling, strip mining)
> that make such a library able to compete with hand written C.
but by the phrase 'arbitrary size' i mean varying at runtime
for 3d rotation one should consider quaterninons - not a matrix
and explain please what you mean by unrollin? loop unrolling?
so far i suggest separate classes for vector and matrix and a general
abstraction for order >2
vector (column) and matrix would have common public interface while
less common high order entities would be represented by some tensor<>
template or whatever with general interface
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk