recusion in cuda

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear all
this is code fragment for my program
I made this function as a recursive function, it compiles correctly but don’t give the right answer.
please help me.

//assume P=8
__device__ void FinalMerge(int *a, int *temp, int rec_num)
{

	int count = 1;
	int threadId = threadIdx.x + blockIdx.x*blockDim.x;
	
	if (0 <= threadId && threadId < (P/2)){
	
		int Lindex1 =  rec_num * (threadId);
		int LendIndex1 = Lindex1 + rec_num/2-1;
                int Lindex2 = LendIndex1 + 1; 
                int LendIndex2 = Lindex2 + rec_num/2-1;
		//---------------
		int LtargetIndex = Lindex1; 
		int LsortedSize = 0;

		
		while (Lindex1 <= LendIndex1 && Lindex2 <= LendIndex2){
			if (LsortedSize == rec_num/2)
				break;

			if (a[Lindex1] <= a[Lindex2]){
				temp[LtargetIndex] = a[Lindex1];
	                        ++LsortedSize; 		
                                ++Lindex1; 	
                        	++LtargetIndex;
			}

			else{
				temp[LtargetIndex] = a[Lindex2]; 
	                       ++Lindex2;  	
                              ++LsortedSize; 		
                              ++LtargetIndex;
			}


		}
	


	++count;
	__syncthreads();
	if(count<=(P/2)){
		rec_num = (2 ^ count)*1024;
		FinalMerge(temp, temp,temp10,(rec_num));
		
	}
	]
	
}

FinalMerge<< <P, 1024 >> >(dev_a, dev_temp);
cudaMemcpy(a, dev_temp, N*sizeof(int), cudaMemcpyDeviceToHost);
2

please define “the right answer”.

3

What you have shown here doesn’t make sense. A device function cannot be configured:

FinalMerge<< <P, 1024 >> >(dev_a, dev_temp);

You’re likely to get better help if you post a short, complete program, rather than fragments or snippets.

4

my program shall made merge sort
the function here is the final stage of sorting which merges the sorted array from each block

here my function can’t merge .
the final array not sorted.