Matrix shapes and performance

[p.vander...@gmail.com]

LS,

I am considering switching to NT2 doing linear algebra. Some things I could not deduce from the documentation so far is whether special matrix shapes are supported (such as diagonal and upper/lower triangular matrices), both for efficient storage and computational performance.

Would someone be able to shed some light on this? And would general performance comparisons with similar libraries be available somewhere?

Many thanks,

Pieter

[Joel Falcou]

On 27/04/2014 21:31, p.vander...@gmail.com wrote:
> LS,
>
> I am considering switching to NT2 doing linear algebra. Some things I
> could not deduce from the documentation so far is whether special
> matrix shapes are supported (such as diagonal and upper/lower
> triangular matrices), both for efficient storage and computational
> performance.

Hi

this is actually under work right now. The plan is to first support
special matrices shapes with no other benefit that selecting proper
linear algebra kernels specialization but no memory advantages. Actual
"packed" representation will be added later.

>
> Would someone be able to shed some light on this? And would general
> performance comparisons with similar libraries be available somewhere?

We have such benchmarks done for a paper we submitted recently. The
actual code have to be moved to master.