Matlab engine in Fortran code (gfortran problem)

Hi,

I followed the instructions http://www.mathworks.nl/help/techdoc/matlab_external/f38569.html to interface some matlab models in my fortran code. When compiling the project with Intel Fortran compiler I use:

-fpp -I/usr/local/MATLAB/R2011b/extern/include -I/usr/local/MATLAB/R2011b/simulink/include -fexceptions -DMX_COMPAT_32

for compiling and:

-Wl,-rpath-link,/usr/local/MATLAB/R2011b/bin/glnxa64 -L/usr/local/MATLAB/R2011b/bin/glnxa64 -leng -lmx -lm

for linking and everything works fine!

When I compile with gfortran, I use the same commands (except -fpp which becomes -cpp) and I get a segmentation fault on the last line of:

mwPointer ep, mVx0
double precision Vx0
Vx0=1.d0
ep = engOpen('matlab ')
if (ep .eq. 0) then
  write(log,*) 'Can''t start MATLAB engine'
  stop
endif
mVx0 = mxCreateDoubleMatrix(1, 1, 0)
call mxCopyReal8ToPtr(Vx0, mxGetPr(mVx0), 1)

Anyone can help out?

Petros

Answer 1

[James Tursa]

In Fortran you should never pass literal constant arguments to the API functions since they can vary in size among different platforms. You can get away with this in C/C++ because arguments are passed by value and are automatically converted to the correct type, but you can't get away with this practice in Fortran. Instead of doing this:

mVx0 = mxCreateDoubleMatrix(1, 1, 0)
call mxCopyReal8ToPtr(Vx0, mxGetPr(mVx0), 1)

you should be doing this:

mwPointer, external :: mxCreateDoubleMatrix, mxGetPr
mwSize :: one = 1
integer*4 :: complexity = 0
    :
mVx0 = mxCreateDoubleMatrix( one, one, complexity )
call mxCopyReal8ToPtr( Vx0, mxGetPr(mVx0), one )

That way the integer arguments should always be the correct type regardless of platform.

You might also be interested in this Fortran 95 interface package:

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