#include <iostream>
#include <cstdlib>
#include <cmath>
#include <omp.h>
#include "Ttiming.h"
#include <opencv2/opencv.hpp>

using namespace std;
using namespace cv;

#define RX 5
#define RY 5
int main(int argc, char *argv[])
{
	int i,j,k,l,m,n;
	TTiming tt;
	char *endptr;
	
	int ratio[RX][RY] =
	{
		{1, 4, 7, 4, 1},
		{4, 16, 26, 16, 4},
		{7, 26, 41, 26, 7},
		{4, 16, 26, 16, 4},
		{1, 4, 7, 4, 1}
	};
	
	if (argc < 4)
	{
		cerr << "Usage: " << argv[0] << " <n> <input_image> <output_image>" << endl;
		exit(1);
	}
	
	int threads_num = strtol(argv[1], &endptr, 10);
	
	if (*endptr)
	{
		cerr << "Invalid number of threads format" << endl;
		exit(1);
	}	  

	if (threads_num <= 0)
	{
		cerr << "The number of threads must be positive" << endl;
		exit(1);
	}
	
	int suma_wag = 0;
	for (i=0; i<RX; ++i)
		for (j=0; j<RY; ++j)
			suma_wag += ratio[i][j];
		
	//ustawienie odpowiedniej ilosci watkow
	omp_set_num_threads(threads_num);	
		
	Mat img = cv::imread(argv[2]);
	if(!img.data )
	{
		cerr << "File " << argv[2] << " does not exist" << endl;
		exit(1);
	
	}
	
	Mat img_out;
	img_out.create(img.rows, img.cols, img.type());
	
	int sumka_r, sumka_g, sumka_b;
	
		//#pragma omp parallel for default(shared)	

	tt.Begin();
	//#pragma omp parallel for default(shared)	
	#pragma omp parallel for private(i, j, k, l, m, n, sumka_r, sumka_g, sumka_b)
	for (i=0; i<img.rows; ++i)
	{
		for (j=0; j<img.cols; ++j)
		{
			if (i<2 || i>img.rows-3 || j<2 || j>img.cols-3)
			{
				img_out.at<Vec3b>(i, j)[0] = img.at<Vec3b>(i, j)[0];
				img_out.at<Vec3b>(i, j)[1] = img.at<Vec3b>(i, j)[1];
				img_out.at<Vec3b>(i, j)[2] = img.at<Vec3b>(i, j)[2];
			}
			else
			{
				sumka_r = 0;
				sumka_g = 0;
				sumka_b = 0;
				m=i-2;
				for (k=0; k<RX; ++k,++m)
				{
					n=j-2;
					for (l=0; l<RY; ++l,++n)
					{
						sumka_b += ratio[k][l] * img.at<Vec3b>(m, n)[0]; 
						sumka_g += ratio[k][l] * img.at<Vec3b>(m, n)[1]; 
						sumka_r += ratio[k][l] * img.at<Vec3b>(m, n)[2]; 
					}	
				}
				img_out.at<Vec3b>(i, j).val[0] = sumka_b / suma_wag;
				img_out.at<Vec3b>(i, j).val[1] = sumka_g / suma_wag;
				img_out.at<Vec3b>(i, j).val[2] = sumka_r / suma_wag;
			}
			
			//cout << img_out.at<Vec3b>(i, j)[0].val[0];
		}
	}
	long elapsed = tt.End();	
	cout << "Time: " << elapsed << " ms" << endl;

	imwrite(argv[3], img_out);
	
	exit(0);
}