#include "cuda_runtime.h"
#include "device_launch_parameters.h"

#include <stdio.h>
#include <math.h>
#include <iostream>
#include <iomanip>
#include <ctime>

#define M_PI       3.14159265358979323846

static const int blockSize = 1024;
static const int gridSize = 500;
static const int size = gridSize * blockSize;

void sumWithCuda(double* table, double cpu_time);


__global__ void sum(double* in, double* out) {
	__shared__ double sh_sum[blockSize];
	unsigned int tid = threadIdx.x;
	unsigned int i = blockIdx.x * blockSize + tid;

	sh_sum[tid] = in[i];
	__syncthreads();

	for (int s = blockSize / 2; s > 0; s >>= 1) {
		if (tid < s) {
			sh_sum[tid] += sh_sum[tid + s];
		}
		__syncthreads();
	}

	if (tid == 0) out[blockIdx.x] = sh_sum[0];
}


int main()
{
	double org_pi = M_PI;
	double sum = 0, pi;
	double* a = new double[size];
	a[0] = 0;
	for (int i = 1; i < size; i++) {
		a[i] = 1 / (pow(i, 2));
	}

	clock_t begin = clock();
	//CPU
	for (int i = 1; i < size; i++) {
		sum += a[i];
	}
	clock_t end = clock();
	double elapsed_time = double(end - begin) / CLOCKS_PER_SEC;
	pi = sqrt(6 * sum);
	double accuracy = 100 * pi / org_pi;
	std::cout << std::setprecision(20) << "CPU: " << pi << "\tczas w sekundach: " << elapsed_time << "\t\tdokladnosc: " << accuracy << "%\n";

	//GPU
	sumWithCuda(a, elapsed_time);

	return 0;
}


// Helper function for using CUDA to add vectors in parallel.
void sumWithCuda(double* table, double cpu_time)
{
	double org_pi = M_PI;
	double* dev_table, * dev_out;
	double* out = new double[size];

	cudaMalloc((void**)&dev_table, size * sizeof(double));
	cudaMemcpy(dev_table, table, size * sizeof(double), cudaMemcpyHostToDevice);

	cudaMalloc((void**)&dev_out, size * sizeof(double));

	clock_t begin = clock();

	sum << <gridSize, blockSize >> > (dev_table, dev_out);
	sum << <1, blockSize >> > (dev_out, dev_out);

	clock_t end = clock();
	double elapsed_time = double(end - begin) / CLOCKS_PER_SEC;


	cudaMemcpy(out, dev_out, size * sizeof(double), cudaMemcpyDeviceToHost);

	cudaFree(dev_table);
	cudaFree(dev_out);

	double pi_gpu = sqrt(6 * out[0]);
	double accuracy = 100 * pi_gpu / org_pi;
	std::cout << std::setprecision(20) << "GPU: " << pi_gpu << "\tczas w sekundach: " << elapsed_time << "\t\tdokladnosc: " << accuracy << "%\n";
	double speed_up = cpu_time / elapsed_time;
	std::cout << std::setprecision(5) << "SpeedUp: " << speed_up << "\n";
}