Re: [Boost-users] [Boost.SIMD] Interest in library

Subject: Re: [Boost-users] [Boost.SIMD] Interest in library
From: dariomt (dariomt_at_[hidden])
Date: 2014-12-05 11:39:18

Mathias Gaunard <mathias.gaunard <at> ens-lyon.org> writes:
> 
> On 17/11/2014 10:09, dariomt <at> gmail.com wrote:
> > Hi,
> >
> > I'm interested in trying out Boost.SIMD. Where can I download the 
latest
> > stable version?
> 
> Latest stable version is 3.1.0, that you can download at
> <http://nt2.numscale.com/downloads/>
> 
> Documentation is here:
> <http://nt2.numscale.com/doc/html/boost_simd.html>
> 
Thanks for the info!

> > Any idea when it might get proposed for Boost review? I think it 
would
> > be beneficial if it is added to the Boost Incubator until then.
> 
> We're working on improving compilation times, which requires some 
> important redesign of the library.
> The library will be submitted for review after that.
> 
Do you know if it will support C++03 or will it require C++11?

> It's been going on for the past few months, but those efforts are 
> finally coming to an end!
> Expect something soon-ish.
> 
> > Also, in http://meetingcpp.com/tl_files/mcpp/slides/12/simd.pdf slide 
36
> > I see mention of a vectorized Thomas Tri-diagonal Solver.
> > I thought the data dependency made the algorithm unfeasible for
> > vectorization, so I'm *very* interested to learn how Boost.SIMD helps 
in
> > this case. Is there more information available about this?
> 
> You can try looking at the paper on this topic,
> "A Parallel Solver for Incompressible Fluid Flows" by Yushan Wang.
> 
> Hopefully Joel Falcou, who also worked on this project and also browses 
> this mailing list, can tell you more about this.
> 
>From that paper: "The SIMD opportunity in our algorithm stands from the 
fact that solving the tridiagonal systems with multiple RHS can be 
performed in parallel."

Darn! I only have a single RHS, too bad!

Thanks!