Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.javascript > #32656 > unrolled thread

a few functions for fast linear algebra

Back to article view | Back to comp.lang.javascript

Contents

  a few functions for fast linear algebra glathoud <glathoud@yahoo.fr> - 2017-02-28 07:58 -0800
    Re: a few functions for fast linear algebra "Michael Haufe (TNO)" <tno@thenewobjective.com> - 2017-03-04 10:12 -0800
      Re: a few functions for fast linear algebra glathoud <glathoud@yahoo.fr> - 2017-03-06 10:34 -0800

#32656 — a few functions for fast linear algebra

Hello,

I'd like to share these functions for fast linear algebra:
http://glat.info/flatorize/lib/flatmat_speedtest.html

There are a few speed tests showing the interest of generating code, for a given matrix dimensions.

Best regards,
Guillaume

[toc] | [next] | [standalone]

#32680

glathoud wrote:
> Hello,
> 
> I'd like to share these functions for fast linear algebra:
> http://glat.info/flatorize/lib/flatmat_speedtest.html
> 
> There are a few speed tests showing the interest of generating code, for a given matrix dimensions.

How does this compare with other libraries such as: <http://glmatrix.net/> ?

[toc] | [prev] | [next] | [standalone]

#32701

On Saturday, March 4, 2017 at 7:12:37 PM UTC+1, Michael Haufe (TNO) wrote:
> [...]
> How does this compare with other libraries such as: <http://glmatrix.net/> ?

glMatrix requires a GPU, requires WebGL support, is limited by what
the shading language can do, i.e. only certain operations, and
especially only on square matrices and vectors of dimensions 2,3,4.
The benefit of those restrictions is the GPU speed, provided that it
largely exceeds the cost of transmitting the data to the GPU and back.

flatmat does not impose any restriction. It runs on the CPU, permits
any matrix or vector dimension, and can be extended at will.

One can see that the application domains differ quite a bit.

Two examples come to my mind:

In a graphical application context (e.g. 3D), both could be useful,
running as many computations as possible on the GPU, and implementing
the remaining ones using flatmat to have as little CPU-induced latency
as possible.

In a generic statistical analysis context (e.g. dimension > 4),
flatmat can make a lot of sense, providing speed and flexibility.

Best regards,
Guillaume

[toc] | [prev] | [standalone]

Back to top | Article view | comp.lang.javascript

csiph-web