compile .mexglx with ifort glib detected error

Hello,

I am trying to compile with ifort 11.1. I have gotten this mex file to compile and run on a Windows machine with a Lahey compiler. Now I would like it to work on my Ubuntu machine.

Here is the mexfunction:

    #include <fintrf.h>
      SUBROUTINE MEXFUNCTION(NLHS, PLHS, NRHS, PRHS)
      IMPLICIT NONE
      INTEGER  PLHS(*), PRHS(*)
      INTEGER  NLHS, NRHS
      INTEGER  mxCreateDoubleMatrix,mxMalloc,flag_complex
      INTEGER  mxIsComplex
      INTEGER  mxGetPr,mxGetM,mxGetN, mxGetPi,indvec_p, indveci_p
      INTEGER  A_pr,A_pi, INPUT2_p, T_pr, T_pi, rnorms_p
      INTEGER  m, n, k, flag_fixed_k
      REAL*8   INPUT2(1), acc

c Check that we have two inputs and two or three outputs

      IF (NRHS .NE. 2) THEN
        CALL mexErrMsgTxt('id_decomp requires two input arguments')
      ELSEIF (NLHS .LT. 3) THEN
        CALL mexErrMsgTxt('id_decomp requires 2 or 3 output arguments')
      ELSEIF (NLHS .GT. 4) THEN
        CALL mexErrMsgTxt('id_decomp requires 2 or 3 output arguments')
      ENDIF

c Extract the problem dimensions from the first input argument.

      m = MXGETM(PRHS(1))
      n = MXGETN(PRHS(1))

c Assign a pointer to the second input argument.

      INPUT2_p = MXGETPR(PRHS(2))
      CALL mxCopyPtrToReal8(INPUT2_p,INPUT2,1)

c Check the nature of INPUT2: c If INPUT2 < 1, then set acc = INPUT2 and use the "specified accuracy" version of id_idd. c If INPUT2 >= 1, then set k = INPUT2 and use the "specified rank" version of id_idd.

      IF (INPUT2(1) .LT. 1) THEN
         flag_fixed_k = 0
         k            = min(m,n)
         acc          = INPUT2(1)
      ELSE
         flag_fixed_k = 1
         k            = min(int(INPUT2(1)), min(m,n))
      END IF

c Check whether the input matrix A is complex.

      flag_complex = mxIsComplex(PRHS(1))
      IF (flag_complex .EQ. 0) THEN
        CALL mexErrMsgTxt('comp_id_decomp cannot yet handle real input')
      END IF

c Create the output matrix T and copy A onto T.

      IF (flag_complex .EQ. 1) THEN
        A_pr     = mxGetPr(PRHS(1))
        A_pi     = mxGetPi(PRHS(1))
        PLHS(1) = mxCreateDoubleMatrix(m,n,0)
        T_pr     = mxGetPr(PLHS(1))
        PLHS(2) = mxCreateDoubleMatrix(m,n,0)
        T_pi     = mxGetPr(PLHS(2))
        CALL mxCopyPtrToReal8(A_pr,%VAL(T_pr),m*n)
        CALL mxCopyPtrToReal8(A_pi,%VAL(T_pi),m*n)
      END IF

c Create the vector "rnorms". c If there are only 2 output arguments, then it is c created simply as a temporary array, otherwise it c is created as an output.

      IF (NLHS .EQ. 2) THEN
        rnorms_p = mxMalloc(8*n)
      ELSE
        PLHS(3)  = mxCreateDoubleMatrix(1,n,0)
        rnorms_p = mxGetPr(PLHS(3))
      ENDIF

c Create an array for the index vector. This is only c a temporary work array since the actual output has c to consist of reals to follow matlab standards.

      indveci_p = mxMalloc(4*n)

c Do the actual computations.

CALL id_pass(%VAL(T_pr),%VAL(T_pi),acc, m,n,k, 1 %VAL(indveci_p),%VAL(rnorms_p),flag_fixed_k)

        CALL mxSetM(PLHS(1),k)
        CALL mxSetN(PLHS(1),n-k)
        CALL mxSetM(PLHS(2),k)
        CALL mxSetN(PLHS(2),n-k)
  
c     Create the REAL VALUED output index vector indvec.
c     Then copy the integer vector indveci to the real valued output indvec.
      PLHS(3)  = mxCreateDoubleMatrix(1,n,0)
      indvec_p = mxGetPr(PLHS(3))
      CALL COPY_IND2REAL(%VAL(indveci_p), %VAL(indvec_p), n)

c Free the temporary work arrays.

      CALL mxFree(indveci_p)
      IF (NLHS .EQ. 3) THEN
        CALL mxFree(rnorms_p)
      ENDIF
      RETURN
      END

It compiles both in Matlab and in the terminal but both ways lead to the following error.

glibc detected * /usr/local/bin/glnx86/MATLAB: double free or corruption (!prev): 0x0a0f39e0 *

I don't know where the problem may be. Maybe flags? Or missing libraries?

Any help is much appreciated.

Thanks.

Answer 1

[Kaustubha Govind]

From the error message, it seems like you have a memory leak in your function (from my experience, Linux is more robust at detecting runtime memory corruption, than Windows - this could be why you don't see any error on Windows).

I see at least two instances of memory leaks in your file:

1. rnorms_p is allocated when (NLHS .EQ. 2), but you release the memory when (NLHS .EQ. 3) - there is probably a typo in one of these conditions (they must be the same condition).
2. PLHS(3) is being allocated twice using mxCreateDoubleMatrix when (NLHS .EQ. 2) is false.

[Adrianna]

Thank you for your help. This fixed the problem.