Re: [Help-gsl] Re: gsl c++ wrapper ?

[Rohit Garg]

On Sat, Sep 12, 2009 at 3:19 AM, Leo Razoumov <address@hidden> wrote:
> On 9/10/09, Rohit Garg <address@hidden> wrote:
>>  [..snip..]
>> But that still begs the question. Why would you want to use GSL's
>>  linear algebra subset?
>
> Here are few:
> (1) Simplicity of GSL design
> (2) No expression templates !!
I do not agree with you here, but if you like GSL's approach better
here, then good for you.

> (3) support for BLAS with vectorized optimizations (SSE2 and later)
I was under the impression that gsl's blas was fairly generic and you
had to use ATLAS and the like to get good performance and
vectorization. Or are you referring to gsl compiled with
auto-vectorization?

> (4) total manual control of resources

I certainly have less experience than you, and I _demand_ total manual
control over my code when I am optimizing it. But since eigen is a
purely template library, I think it offers you more choices.

BTW, I mostly link with fedora's gsl packages, so our POV may be
different. How do you use gsl ( in terms of compiling/compile
options/linking etc. ) ?

>
> I do not want to derogate your project. It could well be very good,
> indeed. But my 15+ years of large scale numerical simulations taught
> me few lessons. I would pick a simple, clean library with manual
> resource management numeric and flat debugging anytime over a "smart"
> library like Boost or eigen for that matter.
Agreed, overall choice depends on many factors. smartness is one of them.

> more crucial. And when you optimize algorithms you want to be very
> close to your hardware and have full manual control. In numerics I see
> people going from C to asm more often than to C++.

How about this for a manual control over optimization

http://listengine.tuxfamily.org/lists.tuxfamily.org/eigen/2009/01/msg00234.html

The product would have been auto vectorized if the size was a multiple
of 2. (3 in this case).

Just my 1 cent. :)

PS: GSL's non linear algebra subset is great, and I am really thankful
to the devs for this.
-- 
Rohit Garg

http://rpg-314.blogspot.com/

Senior Undergraduate
Department of Physics
Indian Institute of Technology
Bombay