Boost logo

Ublas :

Subject: Re: [ublas] Project using ublas and the lapack bindings
From: Jesse Perla (jesseperla_at_[hidden])
Date: 2009-03-28 20:17:15

On Sat, Mar 28, 2009 at 7:41 PM, Thomas Klimpel <Thomas.Klimpel_at_[hidden]
> wrote:

> This doesn't make it easy to help Jesse. So I decided to test the new
> automatically generated numeric_bindings... I started by modifying the
> automatically generated CMake build system such that it works on my system,
> and typed "make -k". Some bindings failed to compile, but most passed at
> least the "does a file including the automatically generated binding
> compiles?" test. So I thought about making the test more challenging ("does
> code using the automatically generated binding compiles?"), but noticed that
> the checked in bindings generator doesn't reproduce the checked in
> automatically generated bindings. Could you check in the latest version of
> the bindings generator? (The difference is quite small, so I could also live
> with checked in bindings generator for what I'm trying to do.)

Thanks for all your help guys,
It appears from some of the discussions recently that it may be a little
premature for me to try to use the trickier functions. Especially if there
are traits classes in progress for things like triangular matrices, packed
storage, etc. that need some time to think through and since they are
generic numeric traits/decorations questions that would be used in a variety
of bindings.

I think I have tracked down enough Fortran code that solves my most
immediate linear algebra problems, and yesterday I think I successfully
bound to it with dense ublas matrices, so I think that will keep me going
for the next month or so. Speaking of which, are there good utilities to
take fortran interfaces/modules and generate a bunch of C headers? I will
be using a bunch of functions inside of a control/filtering library called and will contribute them at some point. I think I am starting to
really understand the beauty of the numeric traits.

So I won't need to use zgges, seqr, and all these other very specific
routines for now. I may even have bought enough time to just use high level
solve interfaces, etc. and the most likely after that will probably be
eigenvectors/values and a linear least squares. I don't want all of the
people I am convincing to use the bindings to get confused with the v1
bindings, so I think I will still test routines as I need them from the
trunk when you tell me they are ready. The routines I need will probably be
pretty straightforward, general dense routines for the immediate future,
with an emphasis on the high level routines.