[Trilinos-Users] templated lapack?

[Pat Notz]

On Tue, Mar 29, 2011 at 9:55 AM, Baker, Christopher G. <bakercg at ornl.gov> wrote:
> Ahhh.... perhaps our grandchildren will enjoy a templated LAPACK.
> Unfortunately, every time the LAPACK is rewritten (CULA, PLASMA, etc), it
> is done without the benefit of templating. Players always be hatin' on
> templates.

No kidding. Even the MPACK that Mike referred to (Thanks Mike!)
doesn't use templates. Instead, they have hard coded routines for a
four different high precision libraries. Ugh.

>
> One short-term solution is to manually implement methods in
> Teuchos::LAPACK from the reference C/LAPACK. We already have this
> capability mostly in place for the BLAS. This involves, at least,
> templating the code from CLAPACK. At best, it should involve some testing
> as well. GESV is pretty straightforward, as far as LAPACK routines go. A
> warning: templating the reference LAPACK should be done carefully; when
> the BLAS methods were ported, there was no consideration for complex
> types, so that the generic methods in Teuchos::BLAS are incorrect for
> complex scalar types.

It does seem like the best approach. It's *oh so* tempting to try and
modernize the code a bit (string literals like "No transpose" as
flags?!). Anyway, I'll take a look at that.

> Chris
>
>
>
> On 3/29/11 11:43 AM, "Pat Notz" <patnotz at gmail.com> wrote:
>
>>Hi,
>>
>>I'm looking at using Tpetra for some problems where I'd like higher
>>precision (e.g., quad). I'm currently using lapack's dgesv for a small
>>(4x4 or 5x5) dense linear solve on the inside of my algorithm.
>>
>>Can anyone recommend a replacement for dgesv that supports templated
>>scalar types?
>>
>>Thanks in advance ~ Pat
>>
>>_______________________________________________
>>Trilinos-Users mailing list
>>Trilinos-Users at software.sandia.gov
>>http://software.sandia.gov/mailman/listinfo/trilinos-users
>
>