Tamas K Papp wrote:
On Fri, Apr 06, 2007 at 05:11:27PM -0400, Raymond Toy wrote:
"Tamas" == Tamas K Papp <Tamas> writes:
Tamas> Hi,
Tamas> Is it possible to call a Fortran subroutine from Lisp? I have seen
Tamas> the C language FFI in the docs, but I wonder if it would be possible
Tamas> to call Fortran somehow. There is a subroutine for banded sparse
Tamas> matrices in LAPACK I would like to use.
As mentioned by others, there are already libraries that can do this.
If you want to cook your own, you do need to exercise some care. All
Fortran compilers I know of have, essentially, a C-compatible
interface. The only thing you need to know about is how Fortran
handles strings.
If the existing libraries don't do what you want, please ask again
here, and we can help with setting up a call to a Fortran routine.
Hi Raymond,
First, a short explanation why I am not using the existing libraries:
I need the ability to reshape a more than 2 dimensional array into a
matrix, do some matrix manipulation (eg matrix product, solving linear
system), reshape the result into an array, etc. I don't know how to
do this in either Matlisp or rfi's libraries.
Matlisp only handles 2-D matrices. You can mix and match Lisp arrays
and Matlisp matrices, with some care.
I tried calling dgemm to do a simple matrix multiplication first.
Fortunately Atlas has a C interface, so I can use that. But I got
stuck because I don't know how to extract the pointer of a simple
array, the sys:vector-sap mentioned in the user's guide doesn't work
here. What should I use instead?
vector-sap only works on specialized arrays. For other array types, you
need to get to the actual storage area that hold the data.
(%array-data? or with-array-data?) Once you have that object,
vector-sap will do what you want.
If you have any advice on what to do, or how to improve this code in
other ways, please tell me.
[snip code]
What happens when you run this code? I can't easily run this myself
right now.
One note, however. LAPACK matrices are in column-major order but Lisp
is row-major order. You might need to transpose your matrices before
calling LAPACK routines. This is why Matlisp stores its matrices as a
1D vector internally. Perhaps the C interface doesn't need this.
Ray
