Just *out of curiosity*:

I *think* part of the machinery for supporting Eigen is already in place.

Exporting the env var "PYOP2_BLAS_DIR" to the Eigen's main folder should do the trick (it's not just finding the headers because Eigen is c++, and you need to wrap your kernel's signature with extern "C" for linking to work), provided that one makes the condition "if self._kernel._applied_blas:" - line 655 in host.py - more general.

2015-02-10 10:03 GMT+00:00 Ham, David A <david.ham@imperial.ac.uk>:


On 9 February 2015 at 18:15, Mitchell, Lawrence <lawrence.mitchell@imperial.ac.uk> wrote:

On 9 Feb 2015, at 18:11, Colin Cotter <colin.cotter@imperial.ac.uk> wrote:

> Dear Firedrakers,
>       David has been helping me by implementing a way of assembling the inverse of a local tensor. This involves calling LAPACK routines, and we realised that the big problem is how to locate clapack, and how to link against it.
>
>       The only approach we could think of was that scipy.linalg.lapack builds a clapack.so. Then, we could make our own header file that contains the interfaces to the LAPACK routines that we want to use, and compile aginst clapack.so. We really don't want to replicate some kind of CMAKE approach to locating libraries for COFFEE.

Use Eigen, which is header only?

I don't understand why that helps - now we have to find Eigen (finding headers versus finding libraries would not seem to be a huge distinction).
 

Lawrence



--
Dr David Ham
Departments of Mathematics and Computing
Imperial College London

http://www.imperial.ac.uk/people/david.ham