[R] Use of .Fortran

[Duncan Murdoch]

On 18/06/2010 9:59 AM, David Scott wrote:
> I have no experience with incorporating Fortran code and am probably 
> doing something pretty stupid.
>
> I want to use the following Fortran subroutine (not written by me) in 
> the file SSFcoef.f
>
>        subroutine SSFcoef(nmax,nu,A,nrowA,ncolA)
>        implicit double precision(a-h,o-z)
>        implicit integer (i-n)
>        integer l,i,nmax
>        double precision nu,A(0:nmax,0:nmax)
>        A(0,0) = 1D0
>        do l=1,nmax
>        	do i=1,l-1
>        		A(l,i) = (-nu+i+l-1D0)*A(l-1,i)+A(l-1,i-1)
>        	end do
>        	A(l,0) = (-nu+l-1D0)*A(l-1,0)
>        	A(l,l) = 1D0
>        end do
>        return
>        end
>
>
> I created a dll (this is windows) using R CMD SHLIB SSFcoef.f
>
> Then my R code is:
>
> ### Load the compiled shared library in.
> dyn.load("SSFcoef.dll")
>
> ### Write a function that calls the Fortran subroutine
> SSFcoef <- function(nmax, nu){
>    .Fortran("SSFcoef",
>             as.integer(nmax),
>             as.integer(nu)
>             )$A
> }
>
> SSFcoef(10,2)
>
> which when run gives
>
>  > SSFcoef(10,2)
> NULL
>
> I am pretty sure the problem is that I am not dealing with the matrix A 
> properly. I also tried this on linux and got a segfault.
>
> Can anyone supply the appropriate modification to my call (and possibly 
> to the subroutine) to make this work?

Two problems:

Your subroutine takes 5 arguments, you're only passing two. 
You didn't name your arguments, but are trying to retrieve A by name.

So this will get you closer (it's untested, so there might still be 
problems...):

SSFcoef <- function(nmax, nu){
   .Fortran("SSFcoef",
            as.integer(nmax),
            as.integer(nu),
	    A = numeric((nmax+1)^2),
	    nrowA = as.integer(nmax+1), # These are unused...
	    ncolA = as.integer(nmax+1)
            )$A
}


Duncan Murdoch