Re: [ublas] Dense matrix * sparse vector efficiency

From: Gunter Winkler (guwi17_at_[hidden])
Date: 2006-11-06 04:07:51

On Monday 06 November 2006 05:30, liu chang wrote:
> Thank you very much Gunter for the very helpful reply. As I'm very new
> to ublas, I hope I've not misunderstood you. To rephrase, all I need
> to do to get the speed up is to replace
>
> dv = prod(dm, sv);
>
> by
>
> dv = axpy_prod(dm, sv);

syntax: axpy_prod(dm,sv,dv,true)
(means dv = dm*sv)

syntax: axpy_prod(dm,sv,dv,false)
(means dv += dm*sv)

If this is still not fast enough, you can look at operation.hpp and create a 
new (partial) specialization of axpy_prod.

mfg
Gunter