Boost logo

Ublas :

From: Nico Galoppo (nico_at_[hidden])
Date: 2007-10-01 12:23:14

Hi Gunter,

I like the efforts your are putting into having some sort of sparse
vector support in ublas. Just as a heads-up, is it possible that you
are using the wrong nomenclature when calling your functions
scatter/gather? I believe that gather is a better name for converting
a sparse set (like a sparse vector) into a dense set (like your
working set). Scatter is a better name for taking elements from a
dense set (working set) and putting them into (or 'scattering' them
into) a sparse vector.

Just my 2c.


On 10/1/07, Gunter Winkler <guwi17_at_[hidden]> wrote:
> Hallo,
> Am Montag, 1. Oktober 2007 05:07 schrieb choon:
> > Actually, I will use sparse matrix M in both left and right hand
> > mat-vec multiplication, so, storing M as column major will affect the
> > performance of the other one. My application is a cutting-plane
> > method, so it need to store some vectors (I am happy to describe this
> > in more details, if necessary).
> Ok, I see. So you get an additional vector on each iteration, right?
> Maybe you should have a look at operation_sparse.hpp which contains
> sparse_prod(). This is an optimized version of prod for multiplication
> of two sparse matrices. In order to speed up the left and right
> multiplication we have to specialize axpy_prod() for compressed_matrix
> and compressed_vector. Unfortunately my time is quite limited, so I can
> only help with comments.
> > Could you please tell what do you mean by compressing dense vector
> > into archive? Also, I am thinking of doing the operation with dense
> > vector but keep the same vector in sparse format. Do you have any
> > efficient way of doing that of it is just a sparse vector
> > construction with dense vector as argument?
> I just added my scatter/gather example on
> Maybe this helps. The code is still experimental and ugly. Maybe you
> find the time to optimize it. If it is really useful we should consider
> to include it into uBLAS.
> mfg
> Gunter
> _______________________________________________
> ublas mailing list
> ublas_at_[hidden]

Nico Galoppo ::