Mex Fortran Gateway Function

Hi,

Can someone help me write a mex gateway function for making a call to some fortran subroutines. The routines are written in fortran 77 and I'm using Microsoft Visual Studio 2005 with Intel Visual Fortran Compiler 10.1. The subroutines compile fine, but I'm having difficulty with the mex gateway function.

I'm trying to make a mex gateway function that will support a call to the following fortran subroutine:

subroutine crvfit(xfit,yfit,xfite,yfite,imode,yfit2,nptsf,delfac)

or

call crvfit(xfit,yfit,xfite,yfite,imode,yfit2,nptsf,delfac)

Inputs:

xfit, yfit - x and y data vectors to be fitted

xfite, yfite - the errors for the x and y vectors

imode = 1 (integer)

nptsf - the number of points to be fitted, i.e. length of xfit, yfit, xfite, and yfite

delfac = 500 (integer)

Outputs:

xfit, yfit

xfite, yfite

imode

yfit2 - temporary fit vector

nptsf2

delfac

So given these inputs, I want to return all of these outputs. The inputs will be passed to the function in MATLAB like:

[xfit, yfit, xfite, yfite, imode, yfit2, nptsf, delfac] = crvfit(xfit, yfit, xfite, yfite, imode, yfit2, nptsf, delfac)

I've tried to do this myself using example fortran mex gateway functions, but I'm very confused with using mex functions and having pointers to data types being passed through mex functions and returning them to MATLAB.

Thanks!

[Jan]

Although we could try to create the mexFunction for you, but as long as you are too confused by the details, you have no chance to adjust any details. Therefire I assume the most efficient method to solve your problem is to learn how to use pointers. Have you inspected the examples shipped with Matlab?

[Aaron]

Yes, I have experimented with the examples shipped with Matlab. I have a good understanding of pointers in other languages, so I will not be confused with adjusting the dynamics of the code.

If you think I might be confused, comments in your answer never hurt :)

Answer 1

[James Tursa]

OK, here is some code you can try (CAUTION: UNTESTED). Uses the Fortran 95 MATLAB interface code from the FEX I mentioned in an earlier post (compile this first). Just call this file crvfit.f90 and then compile it like this:

mex crvfit.f90 MatlabAPImx.obj

If you run into problems let me know. It is horribly inefficient with regards to memory copying, but I am simply following your strict interface requirements.

#include "fintrf.h"

!---------------------------------------------------------------------

        subroutine mexFunction(nlhs, plhs, nrhs, prhs)
        use MatlabAPImx
        implicit none
  !-ARG
        mwPointer plhs(*), prhs(*)
        integer*4 nlhs, nrhs
  !-LOC
        real(8), pointer :: xfit(:), yfit(:), xfite(:), yfite(:), yfit2(:)
        integer imode, nptsf
        real delfac
  !-----
        ! insert some input argument checks here
        xfit => fpGetPrCopy1(prhs(1))
        yfit => fpGetPrCopy1(prhs(2))
        xfite => fpGetPrCopy1(prhs(3))
        yfite => fpGetPrCopy1(prhs(4))
        imode = mxGetScalar(prhs(5))
        yfit2 => fpGetPrCopy1(prhs(6))
        nptsf = mxGetScalar(prhs(7))
        delfac = mxGetScalar(prhs(8))
        ! check association status of xfit etc here
        call crvfit(xfit,yfit,xfite,yfite,imode,yfit2,nptsf,delfac)
        plhs(1) = mxArray(xfit(1:nptsf))
        plhs(2) = mxArray(yfit(1:nptsf))
        plhs(3) = mxArray(xfite(1:nptsf))
        plhs(4) = mxArray(yfite(1:nptsf))
        plhs(5) = mxArray(dble(imode))
        plhs(6) = mxArray(yfit2(1:nptsf))
        plhs(7) = mxArray(dble(nptsf))
        plhs(8) = mxArray(dble(delfac))
        call fpDeallocate(xfit)
        call fpDeallocate(yfit)
        call fpDeallocate(xfite)
        call fpDeallocate(yfite)
        call fpDeallocate(yfit2)
        end subroutine mexFunction

        subroutine crvfit(xfit,yfit,xfite,yfite,imode,yfit2,nptsf,delfac)
        double precision xfit(*),yfit(*),xfite(*),yfite(*),yfit2(*)
        integer imode
        integer nptsf
        real delfac
        ! etc
        end subroutine

[Aaron]

Thank you so much for your help! I'm sorry it took me so long to respond. After debugging some of the crvfit and lower level fortran 77 routines, this interface worked wonderfully. I modified it slightly, but it was general enough to easily adapt. Thank you for your help! A+!

Answer 2

[James Tursa]

Please provide the exact interface for this routine:

subroutine crvfit(xfit,yfit,xfite,yfite,imode,yfit2,nptsf,delfac)

i.e., give the type, rank, and size specifications for each of the arguments as they are listed in the subroutine. Then it will likely not be too hard to create the mex gateway for this using my Fortran 95 Interface package from the FEX:

http://www.mathworks.com/matlabcentral/fileexchange/25934-fortran-95-interface-to-matlab-api-with-extras

I can help you once I know the exact interface for the routine.

[Aaron]

double precision xfit(*),yfit(*),xfite(*),yfite(*),yfit2(*)

integer imode

integer nptsf

real delfac

The size of xfit, yfit, xfite, yfite, and yfit2 are determined by the value of nptsf, input by the user. They are defined as vectors/arrays.

[James Tursa]

Follow-up question about your proposed calling syntax:

[xfit, yfit, xfite, yfite, imode, yfit2, nptsf, delfac] = crvfit(xfit, yfit, xfite, yfite, imode, yfit2, nptsf, delfac)

Why is *everything* listed as both an input and an output? Aren't some of these just inputs and some just outputs? e.g., does the crvfit routine change the value of nptsf? If not, then why is it an output? And is xfite really an input, or is this really just an output that is calculated by crvfit? Please clarify what, exactly, are the inputs and outputs.

[Aaron]

crvfit is a top-level function that calls many lower level functions to modify the values of all of the variables, xfit, xfite, nptsf, etc...

Given the values of xfit, yfit, etc..., it returns new values for these input variables based upon some input that the user is prompted for once the routine is exectued.

[James Tursa]

Sorry if I keep pestering you with questions, but the interface code will depend on the answers you give. My interpretation of what you have said thus far is that xfit, yfit, xfite, yfite, and yfit2 are all pre-populated vectors of size nptsf on input, but that the crvfit routine will return a *different* value of nptsf (presumably *less* than the input value of nptsf?) and that you want the returned size of these vectors to be this new value of nptsf. Is that correct?

[Aaron]

Exactly! Yes, this is correct.