|
|
Ublas : |
Subject: Re: [ublas] [bindings] Initial submission plan?
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2009-02-11 06:30:52
Indeed BLAS seems to be a lot of work, but luckily its code base and
bindings structure are similar to that of LAPACK. I've split the
generators, and put their shared codebase in seperate python modules.
Please find attached a file that the a blas-generator managed to create. It
will probably take some extra templating (engine), but I guess not as much
as has been needed for LAPACK.
I agree the Traits part is the core. I'm all in favour of the incremental
approach, let's start by having the Traits being "good to go", and let's
see how far we get with BLAS/LAPACK by then.
As you might have read in another post, to maintain usability for current
bindings users, sandbox/numeric_bindings is created. Could you be persuaded
to create a quickbook-file for the traits in Numeric_Bindings at
sandbox/numeric_bindings?
Are MUMPS and UMFPACK complete in the sense that they fully cover the
underlying libraries?
Cheers,
Rutger
Karl Meerbergen wrote:
> Dear All,
>
> Several people have asked to have the bindings in Boost. (Some on this
> list, some in private communication with me.)
>
> A few months ago I suggested to only have the traits classes in an
> official repository since the bindings algorithms (blas, lapack, etc)
> were far from complete. (In the meantime, hard work on lapack has been
> done.)
>
> So I would opt for an incremental process where the traits should be
> part of it from the beginning and bindings to packages can be added once
> they are complete. The lapack bindings now are complete and could be
> added in this first stage. The blas bindings need a tremendous amount of
> work. MUMPS and UMFPACK are OK I think and could also be added. New
> bindings could then be created in the sandbox and added after approval
> (I am thinking of an FFT package, ARPACK, BLZPACK).
>
> Best,
>
> Karl
>
>