Segmentation fault in MEX files when parallelized with openmp

I'm trying to parallelize a mex code written in FORTRAN90 using OpenMP directives. The code works perfectly when I don't use any OpenMP but when I parallelize it I receive segmentation faults. I debugged the code running MATLAB in debugger mode. It reaches to the end of mex function without any problem and it yields this error

"Program received signal SIGSEGV, Segmentation fault. [Switching to Thread 0x7fff7afe5700 (LWP 9631)] 0x0000003c0e210410 in ?? ()"

when passing the result to MATLAB workspace I guess. Does anyone has any idea for a possible reason for such an error? I use R2014a. And here is the code:

    subroutine qtest(srcRe,srcIm,nF,nS,freq,srcPos,nDim,lstPos,nL,c0,resp)
    use omp_lib
    implicit none
    integer*8, intent(in) :: nF, nS, nDim, nL
    real*8, intent(in)    :: srcRe(nF,nS),srcIm(nF,nS),freq(nF)
    real*8, intent(in)    :: srcPos(nS,nDim), lstPos(nL,nDim), c0
    real*8, intent(out)   :: resp(nF,nL)
    real*8                      :: pi, k(nF), r(nS), x2(nS)
    complex*16                  :: icplx, G(nF), ikr(nF), p(nF)
    integer*4                   :: i,j,m

    icplx   = (0.d0,1.d0)
    pi      = 4.d0*datan(1.d0)
    k       = 2.d0*pi*freq/c0
    resp    = 0.d0
    G       = 0.d0
    !$OMP PARALLEL
    do i = 1,nL
        !$OMP SINGLE
        x2  = (lstPos(i,1) - srcPos(:,1))**2.d0
        r   = x2
        do m = 2,nDim
            r   = r + (lstPos(i,m) - srcPos(:,m))**2.d0
        enddo
        r   = dsqrt(r)
        p   = 0.d0
        !$OMP END SINGLE
        !$OMP DO PRIVATE(j,ikr,G) REDUCTION(+:p)
        do j = 1,nS
            ikr = icplx*k*r(j)
            G   = cdexp(-ikr)/(4.d0*pi*r(j))
            p   = p + (srcRe(:,j) + icplx*srcIm(:,j))*G
        enddo
        !$OMP END DO
        !$OMP SINGLE
        resp(:,i)   = resp(:,i) + 20.d0*dlog10(cdabs(p)/20.d-6)
        !$OMP END SINGLE
    enddo
    !$OMP END PARALLEL
    end subroutine qtest

And the mexfunction I use to transfer the variables from/to Matlab workspace is this:

    #include "fintrf.h"

    !
    SUBROUTINE MEXFUNCTION(NLHS, PLHS, NRHS, PRHS)
    use mexf90
    implicit none
    integer*8, intent(in)   :: PRHS(*)      ! Pointers carrying the input data
    integer*8, intent(out)  :: PLHS(*)      ! Pointers carrying the output data
    integer*4, intent(in)   :: NLHS,NRHS    ! REMAINS THE SAME FOR 64BIT system
    !-----------------------------------------------------------------------
    integer*8               :: err, nF, nF1, nS, nS1, nDim, nDim1, nL
    real*8                  :: c0
    integer*8, pointer      :: srcRe, srcIm, freq, srcPos, lstPos, resp
    character(200)          :: errMsg
    integer*4               :: txt, classId
    integer*4, external     :: mexprintf
    ! ASSIGN POINTERS TO THE VARIOUS PARAMETERS
    srcRe   =>MXGETPR(PRHS(1))
    nF      = MXGETM(PRHS(1))
    nS      = MXGETN(PRHS(1))
    srcIm   =>MXGETPR(PRHS(2))
    freq    =>MXGETPR(PRHS(3))
    nF1     = MXGETM(PRHS(3))
    srcPos  =>MXGETPR(PRHS(4))
    nS1     = MXGETM(PRHS(4))
    nDim    = MXGETN(PRHS(4))
    lstPos  =>MXGETPR(PRHS(5))
    nL      = MXGETM(PRHS(5))
    nDim1   = MXGETN(PRHS(5))
    c0      = MXGETSCALAR(PRHS(6))

    plhs(1) = mxCreateDoubleMatrix(nF,nL,0)
    resp    =>mxGetPr(plhs(1))
    call qtest(srcRe, srcIm,nF,nS,freq,srcPos,nDim,lstPos,nL,c0,resp)

    END SUBROUTINE MEXFUNCTION

[James Tursa]

Sorry for the late reply. I just saw this post. I haven't looked at your code in any detail, but will tell you that MATLAB used to be incompatible with Fortran and OpenMP. Perfectly valid code using the simplest OpenMP constructs would crash MATLAB. This was several years ago, so this issue may have been fixed by now and may not be the cause of your current problem.