Function.cu

// The code that runs on CPU Properly.
void CPU_function(int row,int col,int window_sizes[],int num_window)
{

	int i;
	long int d=0;
	int win;
	if(col!=0)
	{
	// Variable window loop
	for(win=0;win<num_window;win++)
	{
		for(i=first_index[win];i<col;i++)
		{
	d=(key_column[row*MAX_LIMIT+first_index[win]])-(key_column[row*MAX_LIMIT+col]);
			if(d<=t_window[win])
			{
				if(key_column[row*MAX_LIMIT+last_index[win]]!=0)
				{
					printf("\nScore=%d",score[win]);
					first_index[win]++;
					last_index[win]=first_index[win]+window_sizes[win]-1;
				}
				break;
			}
			else
			{
			   first_index[win]++;
		           last_index[win]=first_index[win]+window_sizes[win]-1;
			}
		}

	}//Variable window loop end

	}

}


//Thread Organization set

dim3 block(num_winow,num_threads);
dim3 grid(1,1)
/*
My Question below

Usually num_window will be a small value say 3 and usually less than 30 and the number of threads I need is keep on varies for each num_window.

Example: when num_window[0]=3 (window of size= 3), I need 45 threads
	when num_window[1]=5 (window of size=4 ) I need 35 threads.

How to get such variable number of threads for each num_window values.????

*/

// The corresponding CUDA code for running on GPU that I wrote.

__global__ void GPU_Function(int row,int col,long int *d_key_column,int *d_num_window,int *d_first_index,int *d_last_index,int *d_t_window,int *d_score,int *d_window_sizes,int D_MAX_LIMIT,long int * d_update_score,int *d_x,int *d_y)
{
	long int i,j,k=0;
	long int d=-1;
	i=threadIdx.x;
	j=threadIdx.y;
	k=threadIdx.y+d_window_sizes[i];// For each window_sizes
	if(col!=0)
	{
		if(i<*d_num_window)
		{
			if(j<D_MAX_LIMIT-d_window_sizes[i]+1)
			{
		d=( d_key_column[row*D_MAX_LIMIT+k]) - (d_key_column[row*D_MAX_LIMIT+j]);
		if(d>=0&& d<=d_t_window[i])
				d_update_score[j]=d_score[i];
			}
		}
	}
}