[Rd] [R] Use of .Fortran

[Göran Broström]

On 06/20/2010 04:02 AM, David Scott wrote:
> Thanks very much to all who replied. I went with Brian's approach, and
> eventually, despite all my attempts to foul it up, I did get it to work
> successfully. For the record here are the details.
>
> The subroutine is:
>
>         subroutine SSFcoef(nmax,nu,A)
>         implicit double precision(a-h,o-z)
>         implicit integer (i-n)

These 'implicit' declarations are dangerous. Use 'implicit none' instead 
and declare explicitly every variable (like you already are doing). You 
get much better error checks (eg, misspelled variable names are caught).

Göran

>         integer k,i,nmax
>         double precision nu,A(0:nmax,0:nmax)
>         A(0,0) = 1D0
>         do k=1,nmax
>         	do i=1,k-1
>         		A(k,i) = (-nu+i+k-1D0)*A(k-1,i)+A(k-1,i-1)
>         	end do
>         	A(k,0) = (-nu+k-1D0)*A(k-1,0)
>         	A(k,k) = 1D0
>         end do
>         return
>         end
>
> This was in the file SSFcoef.f95 and was made into a dll with
>
> R CMD SHLIB SSFcoef.f95
>
> Then calling it in R went like this:
>
> ### 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.double(nu),
> 	    A = matrix(0, nmax+1, nmax+1)
>               )$A
> }
>
> SSFcoef(10,2)
>
>
> There are a number of comments I should make.
>
> Yes, Brian, should have gone to R-devel. I had forgotten about that.
>
> I recognised from my faintly recalled past Fortran experience that the
> code was different and suspected a later Fortran, so good to be advised
> it was 95.
>
> I actually gave a wrong version of the Fortran subroutine, one I had
> been messing around with and had added some extra arguments (nrowA and
> ncolA). As pointed out these were unnecessary.
>
> Something which then caused me a bit of grief before I noticed it.
> Despite the 'implicit integer (i-n)' declaration in the subroutine, nu
> is later declared to be double so has to be specified as double in the R
> code.
>
> Many thanks again, I at least learnt something about calling other
> language code from R.
>
> David
>
> Prof Brian Ripley wrote:
>> On Sat, 19 Jun 2010, David Scott wrote:
>>
>>> I have no experience with incorporating Fortran code and am probably doing
>>> something pretty stupid.
>>
>> Surely you saw in the posting guide that R-help is not the place for
>> questions about C, C++, Fortran code?  Diverting to R-devel.
>>
>>> I want to use the following Fortran subroutine (not written by me) in the
>>
>> Well, it is not Fortran 77 but Fortran 95, and so needs to be given a
>> .f95 extension to be sure to work.
>>
>>> 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
>>> }
>>
>> That does not match.  nrowA and ncolA are unused, so you need
>> SSFcoef<- function(nmax, nu){
>>     .Fortran("SSFcoef",
>>              as.integer(nmax),
>>              as.integer(nu),
>>              A = matrix(0, nmax+1, nmax+1),
>>              0L, 0L)$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?
>>>
>>> David Scott
>>>
>
>

-- 
Göran Broström               phone: 46 90 786 5223
Department of Statistics     fax: 46 90 786 6614
Umeå University              email: gb at stat.umu.se
SE-90187 Umeå, Sweden        http://www.stat.umu.se