I've had trouble with matric inversion with Slatec. I've noticed that
a test in PDL::Stats::GLM fails on 64-bit systems using Slatec matrix
inversion. The same test runs fine on 32-bit systems and with
PDL::MatrixOps matrix inversion on 64-bit systems.

I meant to send an example to the list but I haven't got a chance to
do so since I'm mostly on 32-bit systems.


Maggie


On Mon, Nov 9, 2009 at 10:09 AM, Chris Marshall <[email protected]> wrote:
> * We use our own copy of a *subset* of SLATEC to
>  build the PDL bindings.  Issues with the full
>  SLATEC release should be minimal.
>
> * f2c works fine for computational routines.
>  The hard work is in the IO stuff which I
>  don't believe we are using and if you are
>  trying to interoperate with fortran.
>
> * Fixing the f77/fortran compiler issues for
>  PDL builds for such a basic functionality
>  as matrix inversion would be a big win.
>
> * The conversion should be straightforward
>  and would be a nice fix until something
>  better comes along....
>
> Regards,
> Chris
>
> David Mertens writes:
>>
>> I've sometimes wondered about how we can get
>> slatec support when using gcc-4.x.x (no g77)
>> as the compiler. Is there some general advice
>> (ie not specific to any particular OS) on how
>> this can be achieved?
>>
>>
>> I believe that Chris has suggested running
>> f2c on the Slatec code and then including
>> that in our source code. I don't know how
>> well the f2c code works, though I think this
>> could potentially solve the problem. If it
>> did 90% of the work for us, we could probably
>> get the last 10% done by hand. Normally I
>> wouldn't suggest tweaking auto-generated code
>> by hand, but the Slatec code base hasn't
>> changed since the early 90s, so I think we
>> could safely do this once and never have to
>> touch it again.
>>
>> I'm not sure if we want to do this, however.
>> It's not clear to me if the Slatec code is
>> thread-safe: if it uses global variables or
>> even function-static variables, this would
>> impose a major hurdle to creating a true
>> multi-threaded PDL. I believe GSL provides
>> all all of the capabilities of Slatec and
>> much more, and is thread safe, but this would
>> require that we rewrite some of the basic PDL
>> modules (linear fitting is one I'm aware of)
>> to use GSL code instead of Slatec code.
>>
>> But, as a first-solution to the problem,
>> running f2c on the Slatc code might work.
>>
>> David
>
> _______________________________________________
> Perldl mailing list
> [email protected]
> http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
>
>

_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl

Reply via email to