Re: boost.numeric (was Re: Re: matrix library?)

From: Dean Calver (deano_at_[hidden])
Date: 2001-03-20 13:04:43

>Jeremy Siek wrote:
>
> I'd challenge those interested in boost numerics libraries to come forward
> with concrete lists of problems and algorithms for a particular domain.
> With those in hand we'll be able to determine scope and priority for
> various peices, and it will provide a firmer ground for discussing
> interface and implementation issues
>
> Cheers,
> Jeremy
>

I'm primarly interested in 'small' scale numerics with performance being
key, just about anything to do with vectors and matrices of dimension 8 or
less.

I've got more of a grounding in hardware/performance of workstation machines
than numerical analysis.

My particular field (Computer graphics) we tend to only use 'small'
numerical operations. We create matrices and apply them to thousands of
vectors, invert matrices, solve eigen systems, use functional approximations
and use lots of geometrical algorithms.

Performance and precision are our main bug bears (sometimes have to use
integer and fixed point and usually use floats rather than doubles).

A typical use might be creating a transformation matrix by multipling a
quaternion rotatation with a translation matrix, then inverting it,
extracting eigenvectors and values, then using these to transform thousands
of vectors before sending them to the screen. Another use would be iDCT for
MPEG decompression (integer matrix operations, often very limited
precision).

I'll volunteer myself for any of these type things if there is any interest.

Bye,
    Deano

Dean Calver