void cudaCublas(){
     float elapsedTime;
    cudaEvent_t start, stop;
    float *cx_d, *cx_h, *cy_h, *cy_d;
	cudaEventCreate(&start);
	cudaEventCreate(&stop);

	cudaMallocHost((void**)&cx_h, N * sizeof(float));
	cudaMallocHost((void**)&cy_h, N * sizeof(float));
	cudaMalloc((void**)&cx_d,  N * sizeof(float));
	cudaMalloc((void**)&cy_d,  N * sizeof(float));

	for(int i = 0; i < N; i++) {
		cx_h[i] = (float) i;
		cy_h[i] = 0.87f;
	}

	cublasHandle_t cublas_handle;
	cublasCreate(&cublas_handle);

	const int num_rows = N;
	const int num_cols = 1;
    const int stride = 1;
	//const size_t elem_size = sizeof(float);

	//cublasSetMatrix(num_rows, num_cols, elem_size, cx_h, num_rows, cx_d, num_rows);
	//cublasSetMatrix(num_rows, num_cols, elem_size, cy_h, num_rows, cy_d, num_rows);

    cublasSetVector(num_rows, sizeof(float), cx_h, stride, cx_d, stride);
	cublasSetVector(num_rows, sizeof(float), cy_h, stride, cy_d, stride);


	float alpha = 3.0f;

	cudaEventRecord(start, 0);
	cublasSaxpy(cublas_handle, N, &alpha, cx_d, stride, cy_d, stride);
	cudaEventRecord(stop, 0);

	cudaEventSynchronize(stop);
	cudaEventElapsedTime(&elapsedTime, start, stop);

	cublasGetVector(num_rows, sizeof(float), cy_d, stride, cy_h, stride);
	printf("cuBLAS Time:\n \t %f ms\n", elapsedTime);
	cublasDestroy(cublas_handle);
	cudaFreeHost(cx_h);
	cudaFreeHost(cy_h);
	cudaFree(cx_d);
	cudaFree(cy_d);
}