MATLAB: How to set all one gpuArray in mex

gpuarrayinitialMATLABones

Hi,

I need to set a all-one double gpuArray. I think I could use mxGPUCreateGPUArray to initial it. But it seems mxGPUCreateGPUArray could only initial a all-zero array like below.

mxGPUArray *x = mxGPUCreateGPUArray(ndim, dims, mxDOUBLE_CLASS, mxREAL, MX_GPU_INITIALIZE_VALUES);

Is there any good way to do this? Do I need to write a kernel function to do it with cuda function? Thank you.

Best Answer

You basically have two options: either create and uninitialized mxGPUArray and fill the values using CUDA code (either a kernel, or one of the cuMemset* variants if your data type is small enough), or pass in to your MEX function an array of ones that you create in MATLAB using ones(...,'gpuArray').

Related Solutions

MATLAB: GPU coder : When I use GPU coder to any function, Do I need to type coder.gpu.kernelfun in all subfunctions in the main function

Hello Lim,

I think, Dynamic kernel invocation i.e calling kernels from within kernels is not supported by GPU coder yet. Therefore my guess is even if you write the kernel fun pragma in 'colChk_new' GPU threads will ignore that pragma. But, if the computations involved in the for loop of 'fcn_PRM_DH5' then the function call of 'colChk_new' will be parallel.

~Pravin

MATLAB: Problems declaring static in Mex

Hi,

your code looks pretty bad. Here is one way to do it:

#include "mex.h"
static double *t = NULL;
static int count = 0;
//we need to free the allocated memory, we do this when the mex file is
//released from memory
static void CleanUp(void)
{
  mexPrintf("cleaning up\n");
  //mxFree(t); this wont work
  free(t);
  count = 0;
}
void timestwo(double A, double *y, double *x)
{
    int i,j;
    //first call?
    if (count == 0)
    {  
        //if so allocate memory
        //t = (double*) mxMalloc(80*sizeof(double)); wont work
        t = (double*) malloc(80*sizeof(double));
        for(i=0;i<80;i++)
        {
            //fill up t, which will never change
            t[i] = (double) i;
            //calculate y
            y[i]=A*t[i];
       }
    }else
    {
        //timestwo was called before, so memory is allocated and t is filled,
        //simple calaculte y again
        for(i=0;i<80;i++)
            y[i]=A*t[i];
    }
  count = count+1;
  //copy the t vector into x
  memcpy(x,t,80*sizeof(double));
  mexPrintf("%i\n",count);
  mexPrintf("%f\n",x[50]);
}
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    //first call? 
    if (t == NULL){
        //if so set the cleanup routine
        mexAtExit(CleanUp);
    }
    plhs[0] =mxCreateDoubleMatrix(80,1,mxREAL);
    plhs[1] =mxCreateDoubleMatrix(80,1,mxREAL);
    timestwo(mxGetScalar(prhs[0]),mxGetPr(plhs[0]),mxGetPr(plhs[1]));
}

Best Answer

Related Solutions

MATLAB: GPU coder : When I use GPU coder to any function, Do I need to type coder.gpu.kernelfun in all subfunctions in the main function

MATLAB: Problems declaring static in Mex

Related Question