Thread Subject: various CUDA scripts in one m script failure

hi

i ve just wrote first cuda [2.3] script, but discovered after i put it in one m file with example fft [cuda for matlab 1.1] that they cant work at one time -> second script canot connect with device [however he sees it]. i also run my function two times in one empty m script and i got same msg, however calculation were donne correctly.

#include "mex.h"
#include "cuda.h"
#include "stdio.h"
#include "cuda_runtime.h"


/*makes .* operation plus multiplays with 1 integer
 * (alpha, trans N, N, transM, M)
 */
//nlhs number of outputs
//nrhd number of inputs
//prhs inputs



__global__ void GPU2D(float *n, float *m, float *nm, int nx, int ny)
{
    int i = blockIdx.x* blockDim.x + threadIdx.x ;
    int j = blockIdx.y* blockDim.y + threadIdx.y ;
    if ((i < ny) && (j < nx))
    {
     
       nm[i+j*ny] = n[i+j*ny] * m[i+j*ny];
    }

}

int iDivUp(int a, int b)
 { return (a % b != 0) ? (a / b + 1) : (a / b); }

void mexFunction( int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    
    int MX, MY, NX, NY; //matrix dims
    int MMX, MMY, NNY, NNX; //dims for translation
    float alpha; //alpha real multiplicator
    int tn,tm; // transponing flags
    float *n, *m, *nm; //host malloc
    float *gn, *gm; //device malloc matrix
    float *gnm; // device
    
    //data check START-->
    if(nrhs != 5){
           mexErrMsgTxt("a*m*.n requires 5 args");
    }else if (nlhs !=1 )
            mexErrMsgTxt("a*m*.n requires 1 out");
            
    if( !mxIsSingle(prhs[2]) || !mxIsSingle(prhs[4]) )
        mexErrMsgTxt("inputs not in single precision mode");
            
     tn= (int) mxGetScalar(prhs[1]);
     tm= (int) mxGetScalar(prhs[3]);
     alpha= (float) mxGetScalar(prhs[0]);
     
     
     NX= mxGetM(prhs[2]); //gets number of rows in N
     NY= mxGetN(prhs[2]); //gets number of colums in N
     MX= mxGetM(prhs[4]); //gets number of rows in M
     MY= mxGetN(prhs[4]); //gets number of columns in M
     
  

     if(tn!=0)
     {
           NNX=NX;
           NNY=NY;
     }else
     {
           NNX=NY;
           NNY=NX;
     }
    if(tm!=0)
     {
           MMX=MY;
           MMY=MX;
     }else
     {
           MMX=MX;
           MMY=MY;
     }


//get 2 arrays single precision (float*)for arrays (float** for matrix)
     n = (float*) mxGetData(prhs[2]);
     m = (float*) mxGetData(prhs[4]);

     //initation of left handed array
    plhs[0] = mxCreateNumericMatrix(NNY, NNX, mxSINGLE_CLASS, mxREAL);
     nm = (float*) mxGetData(plhs[0]);

         


int device = 0;
cudaDeviceProp deviceProp;

    cudaGetDevice(&device);

printf("device %d \n", device);

cudaGetDeviceProperties(&deviceProp, device);
printf("Using device %d: %s \n", device, deviceProp.name);

    /*if(cudaSetDevice(device) !=CUBLAS_STATUS_SUCCESS )
    {
        printf ("!!!! Set Device error\n");
            
    }*/

    cudaMalloc( &gn, NNX*NNY*sizeof(float) );
    cudaMalloc( &gm, NNX*NNY*sizeof(float) );
    cudaMalloc( &gnm, NNX*NNY*sizeof(float) );
 
cudaMemcpyHostToDevice );
     cudaMemcpy ( gn, n, NNX*NNY*sizeof(float),cudaMemcpyHostToDevice);

cudaMemcpyHostToDevice );
    cudaMemcpy ( gm, m, NNX*NNY*sizeof(float),cudaMemcpyHostToDevice);

    cudaMemcpy ( gnm, nm, NNX*NNY*sizeof(float),cudaMemcpyHostToDevice);
  
  dim3 dimBlock(8, 8);
   dim3 dimGrid(iDivUp(NNX , dimBlock.x ), iDivUp(NNY ,dimBlock.y));
   GPU2D<<<dimGrid, dimBlock>>>(gn, gm, gnm, NNX, NNY);

    cudaMemcpy ( nm, gnm, NNX*NNY*sizeof float),cudaMemcpyDeviceToHost);


    free(n);
    free(m);
    cudaFree(gn);
    cudaFree(gm);
    cudaFree(gnm);
    
}

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

i cleared all memory i allocated, also tried without setting device [i got only one card so i suppose its unnecessary] but its still no go. or maybe i cant use two different cuda apis in one thread?

keneth