Parallel Programming using CUDA

Any CUDA program broadly consist of the following components:

1)          Include header files

2)          Kernel that executes on the CUDA device, e.g:

     //__global__ void MatrixMulKernel(float *Md, float *Nd, float *Pd, int Width)

3)          main( ) routine, the CPU must find. 

    3.1:-    Define pointer to host and device arrays

    3.2:-   Define other variables used in the program e.g. arrays etc.

    3.3:-   Allocate array on the host

                    /e.g. a_h=(float*)malloc(size)

    3.4:-    Allocate array on device (DRAM of the GPU)
                   /e.g. cudaMalloc ((void**) (a_d,size))

    3.5:-  Copy the data from host array to device array.
              // cudaMemcpy(Md_d,Md_h,size,cudaMemcpyHostToDevice);

    3.6:-   Kernel Call, Execution Configuration  //  e.g add_array<<<n block,p size>>>(…..)

    3.7:-    Retrieve result from device to host in the host memory, e.g;

           cudaMemcpy(Pd_h,Pd_d,size,cudaMemcpyDeviceToHost);

    3.8:-    Print result // for (i=0,………)

               printf(“%f “,,a_h[i])  ;

    3.9:-   Free  allocated device and host memories  // e.g

        free(a_h);
        cudaFree(a_d);

Using the above programming steps,  the following program calculates and prints  the square of first 

1000 integers. 

                                                                     
// 1) Include header files   

#include <stdio.h>
#include <cuda.h>
#include <conio.h>

// 2)          Kernel that executes on the CUDA device

__global__ void square_array(float*a,int N)

{
int idx=blockIdx.x*blockDim.x+threadIdx.x;
if(idx<N)a[idx]=a[idx]*a[idx];
}

// 3)          main( ) routine, the CPU must find

 int main(void)
{

// 3.1:-    Define pointer to host and device arrays

float*a_h,*a_d;    

// 3.2:-   Define other variables used in the program e.g. arrays etc.

const int N=100;

size_t size=N*sizeof(float);

// 3.3:-   Allocate array on the host

a_h=(float*)malloc(size);

// 3.4:-    Allocate array on device (DRAM of the GPU) 

cudaMalloc((void**)&a_d,size);

for(int i=0;i<N;i++)a_h[i]=(float)i;

// 3.5:-  Copy the data from host array to device array.

cudaMemcpy(a_d,a_h,size,cudaMemcpyHostToDevice);

// 3.6:-   Kernel Call, Execution Configuration  

int block_size=4;
int n_blocks=N/block_size+(N%block_size==0);
square_array<<<n_blocks,block_size>>>(a_d,N);

// 3.7:-    Retrieve result from device to host in the host memory, e.g;  

cudaMemcpy(a_h,a_d,sizeof(float)*N,cudaMemcpyDeviceToHost);

// 3.8:-    Print result  

for(int i=0;i<N;i++)
printf("%d\t%f\n",i,a_h[i]);

// 3.9:- Free allocated memories on the device and host 

free(a_h);
cudaFree(a_d);
getch();
}

)