[Beowulf] A start in Parallel Programming?

Robert G. Brown rgb at phy.duke.edu
Tue Mar 20 16:03:40 PDT 2007


On Tue, 20 Mar 2007, Greg Lindahl wrote:

> On Tue, Mar 20, 2007 at 10:58:00AM -0400, Robert G. Brown wrote:
>
>> I think that this does them a disservice.  From the BLAS support
>> section:
>
> OK, so I only read the FFT section of the docs. But the compaint was
> about lack of parallelism, and if ATLAS can be linked, you get
> threaded parallelism for free. See the power of a standardized
> interface?
>
> And they _have_ changed the user interface unnecessarily. When a
> community has exactly one interface for a library, not inventing a new
> one is a good thing. Here, they are only using the standard interface
> under the hood. So they want fancy data types. Well, you can still use
> the standard interface with minimal changes to get your fancy data
> types -- and they didn't. But while we're on the topic:
>
>> So you can use gsl_vector_get(v,i) to get the ith element of the
>> gsl_vector v with all sorts of range and bounds checking,
>
> For those from Planet Fortran, we have this thing called a "one
> dimensional array" (a part of the standardized language) which works
> with "compiler bounds checking" (available in all compilers). Gah.  If
> I use ordinary arrays in C, it sounds like it's a lot of work to use
> the GSL in my code.

As I said, I frankly agree -- I think that they bent over too far to try
to fortranize and typecheck vectors and matrices at the expense of being
able to extend transparently to tensors, etc (although you actually CAN
get to the vectors etc in old fashioned form in the structs, it is just
more than a bit of a PITA to have to decon-struct to do so).

However, it isn't a lot of work to use all the GSL in part of my code,
or part of the GSL in all of my code, or something like that.  Their
special functions are excellent.  Their RNGs are superior and
well-implemented with a consistent interface.  Their quadrature and ODE
solvers are decent.  Their linear algebra and FFT stuff may or may not
be up to snuff, but I rarely use this at a megascale.

I still like the IDEA of the GSL far better than many of the
alternatives.  There is room for improvement, but there is also the
standard open source pathway towards improvement...

>> This is unfortunately not that easy to consistently DO -- as the list
>> has discussed, MPI doesn't really have an ABI yet.
>
> Hey! That's my line!

Yeah, well... where do you think I learned it...;-)

    rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list