MATLAB encountered an internal error and needs to close -MEX file

I debugged my code based on Mathwork post https://www.mathworks.com/help/matlab/matlab_external/debugging-on-microsoft-windows-platforms.html and crashed MATLAB. This my code:

#include "fintrf.h"

        subroutine mexFunction(nlhs, plhs, nrhs, prhs) ! Interface 
        IMPLICIT REAL*8(A-H,O-Z)
        mwpointer plhs(*), prhs(*)
  C      integer nlhs, nrhs
        mwpointer mxCreateDoubleMatrix, mxGetPr
        double precision mxGetScalar

C     Pointers to input/output mxArrays:
      mwPointer TT

C     Get the size of the input array.
      m_in = 20
      n_in = 1
      size = m_in * n_in

C     Create Fortran array from the input argument.
C        IZ  = mxGetScalar(prhs(1));
C        EEAR = mxGetScalar(prhs(2));
C        DDIA = mxGetScalar(prhs(3));
C        RRPM = mxGetScalar(prhs(4));
        NN  = mxGetScalar(prhs(1));
C     Create matrix for the return argument.
      plhs(1) = mxCreateDoubleMatrix(m_in, n_in, 0)
      TT = mxGetPr(plhs(1))

        call LOOPS1(%VAL(TT),%VAL(NN)) ! call the computation
        return
        end

        subroutine LOOPS1(TT, nn)
        IMPLICIT REAL*8(A-H,O-Z)

        n1 = nn+1
        call LOOPS(TT, n1)
        return
        end

The error is in the line (n1 = nn+1). I don't know what the problem is. Please help me.

[iman AHMADI]

Could you please explain me when we use %VAL exactly? Just for outputs or for arrays? I just know it is used instead of mxcopy.

Answer 1

[James Tursa]

Some comments:

----------------------------------------------------------------------------------------

        IMPLICIT REAL*8(A-H,O-Z)

You need to delete this from your brain and never use it again. Ever. One of the few helps you get with a Fortran compiler for spotting typing issues and variable name misspellings etc is with the following, so use it in all of your code from now on:

        IMPLICIT NONE

This will force you to type everything, but it is worth it in the long run.

----------------------------------------------------------------------------------------

C      integer nlhs, nrhs

Which means of course you will need to uncomment this line (as well as add typing statements for all of your other untyped variables).

----------------------------------------------------------------------------------------

      m_in = 20
      n_in = 1
         :
      plhs(1) = mxCreateDoubleMatrix(m_in, n_in, 0)

Unlike C/C++, you do not get automatic argument promotion in Fortran ... especially for implicit interfaces. So you should always use the exact types specified in the documented signature and never use explicit constants in argument lists. E.g., the signature for mxCreateDoubleMatrix is:

mwPointer mxCreateDoubleMatrix(m, n, ComplexFlag)
mwSize m, n
integer*4 ComplexFlag

So your code should follow this signature exactly. E.g.,

      mwSize m_in, n_in
      integer*4 :: ComplexFlag = 0
          :
      m_in = 20
      n_in = 1
         :
      plhs(1) = mxCreateDoubleMatrix(m_in, n_in, ComplexFlag)

----------------------------------------------------------------------------------------

      size = m_in * n_in

size is the name of an intrinsic function in Fortran. You should pick a different name for your variable.

----------------------------------------------------------------------------------------

        call LOOPS1(%VAL(TT),%VAL(NN)) ! call the computation

TT is holding a pointer value, but NN is just an integer variable (not holding a pointer to a integer variable). So you should not be passing NN by value here (this error is what is causing your crash). E.g., you should be doing this instead:

        call LOOPS1(%VAL(TT),NN) ! call the computation

----------------------------------------------------------------------------------------

[iman AHMADI]

Thank you very much James. It worked! Could you please tell me when we should exactly use %VAL? Just for outputs(arrays and scalars) or in another place? I just know that it is used instead of mxcopy.

[James Tursa]

The following API functions return mwPointer types that hold memory addresses: mxGetPr, mxGetPi, mxGetData, mxGetImagData, mxGetIr, mxGetJc, mxGetDimensions. Anytime you are passing the result of these calls to another Fortran routine that is expecting a normal intrinsic type (like REAL, REAL(8), INTEGER, INTEGER(8), etc), you should use the %VAL( ) construct in the calling routine. Basically, you want to put the address of the data on the stack since that is what the other Fortran routine is expecting (for implicit interfaces) to get on the stack.

Answer 2

[iman AHMADI]

Thank you again. I write another code but the code just get the first output!! This is my code:

 #include "fintrf.h"

 #if 0

 C     
C     Ship.F
C     .F file needs to be preprocessed to generate .for equivalent
C     
#endif
C=======================================================================

 C     Gateway subroutine

      subroutine mexfunction(nlhs, plhs, nrhs, prhs)

C Declarations

      implicit none

C mexFunction arguments:

      mwPointer plhs(*), prhs(*)
      integer nlhs, nrhs

C Function declarations:

      mwPointer mxGetPr
      mwPointer mxCreateDoubleMatrix
      mwPointer mxGetM, mxGetN
      real*8    mxGetScalar

C Pointers to input/output mxArrays:

      mwPointer pr_out1, pr_out2

C Array information:

      integer m_in1, n_in1
      integer m_in2, n_in2

      integer*4 :: ComplexFlag = 0
      real*8 pr_in1

      m_in1 = 1
      n_in1 = 1
      m_in2 = 1
      n_in2 = 1

C Create Fortran array from the input argument.

        pr_in1 = mxGetScalar(prhs(1));

C Create matrix for the return argument.

      plhs(1) = mxCreateDoubleMatrix(m_in1, n_in1, ComplexFlag)
      pr_out1 = mxGetPr(plhs(1))
      plhs(2) = mxCreateDoubleMatrix(m_in2, n_in2, ComplexFlag)
      pr_out2 = mxGetPr(plhs(2))

      call image1(%VAL(pr_out1),%VAL(pr_out2),pr_in1) ! call the computation

      return
      end

      subroutine image1(C_Res, Fr_N, pr_in1)
      IMPLICIT REAL*8(A-H,O-Z)
      COMMON /FNCW/FNN, CWW, Fr_in
       .
       .
       .
       Fr_in = pr_in1
       .
       .
       .
      CALL FSPNLP2(NSPC)
       .
       .
       .
       CALL WAVE1
       C_Res = CWW
       Fr_N = FNN
       .
       .
       .
      RETURN
      END

      SUBROUTINE FSPNLP2(NSPC)
      IMPLICIT REAL*8(A-H,O-Z)
      COMMON /CONDP/FN
      COMMON /FNCW/FNN,CWW,Fr_in
      FN = Fr_in
       .
       .
       .
      RETURN
      END

      SUBROUTINE WAVE1
      IMPLICIT REAL*8(A-H,O-Z)
      COMMON /CONDP/ FN
      COMMON /FNCW/FNN,CWW,Fr_in
       .
       .
       .
C After calculating CW
      CWW=CW
      FNN=FN 
      RETURN
      END

I wrote in command window f=0.2;image(fn) and just got the first output(CW). the input and second output(FN) are identical. I need FN to be sure that the input won't change during several CALLs. Do you think this is because of using IMPLICIT REAL*8(A-H,O-Z)?

[James Tursa]

I am not sure I understand the problem. You show the following for how you are calling the function:

f=0.2;image(fn)

If "image" is the name of your mex routine and you want to get both outputs back to the workspace, you need to call it with two outputs (just like you would call any other function that has two outputs). E.g.,

f=0.2;
[out1,out2] = image(fn);  % <-- provide two variable names to store the result