// parloop_cpp0x_openmp.cpp : Defines the entry point for the console application.
//
#include "stdafx.h"

#include <stdio.h>
#include <stdlib.h>

#include <iostream>
#include <cmath>
#include <vector>
#include <limits>

#ifndef _OPENMP
#error OpenMP support required!
#endif
#include <omp.h>



double compute(double val, const int numiterations)
{
	for (int i = 0; i < numiterations; i++)
	{
		val = std::sin(val) * std::cos(val);
	}
	return val;
}


void run_serial(int iNumRepetitions, int iNumElements, int iNumIterations)
{
	std::vector<double> vec(iNumElements);

	double dBest = std::numeric_limits<double>::max();
	int iBest = 0;
	double dWorst = std::numeric_limits<double>::min();
	int iWorst = 0;

	double dStart, dEnd;
	for (int rep = 0; rep < iNumRepetitions; rep++)
	{

		dStart = omp_get_wtime();
		for (int i = 0; i < iNumElements; i++)
		{
			vec[i] = compute(vec[i], iNumIterations);
		}
		dEnd = omp_get_wtime();

		double dTime = dEnd - dStart;
		if (dBest > dTime)
		{
			dBest = dTime;
			iBest = rep;
		}
		if (dWorst < dTime)
		{
			dWorst = dTime;
			iWorst = rep;
		}

	}

	std::cout << "Results of 'run_serial':" << std::endl;
	std::cout << "  +best:  " << dBest << " in repetition " << iBest << std::endl;
	std::cout << "  -worst: " << dWorst << " in repetition " << iWorst << std::endl;
	std::cout << std::endl;
}


void run_openmp_for(int iNumRepetitions, int iNumElements, int iNumIterations)
{
	std::vector<double> vec(iNumElements);

	double dBest = std::numeric_limits<double>::max();
	int iBest = 0;
	double dWorst = std::numeric_limits<double>::min();
	int iWorst = 0;

	double dStart, dEnd;
	for (int rep = 0; rep < iNumRepetitions; rep++)
	{

		dStart = omp_get_wtime();
#pragma omp parallel for shared(iNumElements, vec, iNumIterations)
		for (int i = 0; i < iNumElements; i++)
		{
			vec[i] = compute(vec[i], iNumIterations);
		}
		dEnd = omp_get_wtime();

		double dTime = dEnd - dStart;
		if (dBest > dTime)
		{
			dBest = dTime;
			iBest = rep;
		}
		if (dWorst < dTime)
		{
			dWorst = dTime;
			iWorst = rep;
		}

	}

	std::cout << "Results of 'run_openmp_for':" << std::endl;
	std::cout << "  +best:  " << dBest << " in repetition " << iBest << std::endl;
	std::cout << "  -worst: " << dWorst << " in repetition " << iWorst << std::endl;
	std::cout << std::endl;
}


template<typename F>
void omp_pfor(int start, int end, F x)
{
#pragma omp parallel for
	for(int __i = start; __i < end; __i++)
	{
		x(__i);
	}
}


void run_pfor(int iNumRepetitions, int iNumElements, int iNumIterations)
{
	std::vector<double> vec(iNumElements);

	double dBest = std::numeric_limits<double>::max();
	int iBest = 0;
	double dWorst = std::numeric_limits<double>::min();
	int iWorst = 0;

	double dStart, dEnd;
	for (int rep = 0; rep < iNumRepetitions; rep++)
	{

		dStart = omp_get_wtime();
		omp_pfor (0, iNumElements, [&](int i)
		{
			vec[i] = compute(vec[i], iNumIterations);
		});
		dEnd = omp_get_wtime();

		double dTime = dEnd - dStart;
		if (dBest > dTime)
		{
			dBest = dTime;
			iBest = rep;
		}
		if (dWorst < dTime)
		{
			dWorst = dTime;
			iWorst = rep;
		}

	}

	std::cout << "Results of 'run_ompfor':" << std::endl;
	std::cout << "  +best:  " << dBest << " in repetition " << iBest << std::endl;
	std::cout << "  -worst: " << dWorst << " in repetition " << iWorst << std::endl;
	std::cout << std::endl;
}


void printUsage()
{
	std::cout << "Usage:   parloop  num_repetitions  num_elements  num_iterations" << std::endl;
	std::cout << "   - num_repetitions: number of test repetitions" << std::endl;
	std::cout << "   - num_elements:    dimension of work array" << std::endl;
	std::cout << "   - num_iterations:  number of iterations per element" << std::endl;
	std::cout << std::endl;
}


int main(int argc, char* argv[])
{
	if (argc != 4)
	{
		printUsage();
		return EXIT_FAILURE;
	}
	int iNumRepetitions = atoi(argv[1]);
	int iNumElements = atoi(argv[2]);
	int iNumIterations = atoi(argv[3]);

	run_serial(iNumRepetitions, iNumElements, iNumIterations);
	run_openmp_for(iNumRepetitions, iNumElements, iNumIterations);
	run_pfor(iNumRepetitions, iNumElements, iNumIterations);

	
	return EXIT_SUCCESS;
}