Test int vs float performance with simplest operations.

TestIntFloat

#include "stdafx.h"
#define WIN32_LEAN_AND_MEAN
#include <Windows.h>
#include <array>
#include <vector>
#include <iostream>
#include <sstream>
#include <time.h>

std::ostringstream oss;

size_t tick()
{
	LARGE_INTEGER li;
	::QueryPerformanceCounter(&li);
	return (size_t&)li;
}

template<typename T>
size_t test_one(std::vector<T>& values)
{
	for (T& v : values) v = T(::rand());

	T sum = T(0);
	size_t start = tick();
	for (auto v : values) sum += v;

	//for (auto v : values) sum *= v; // halves win, floaters 60% slower than fixed
	//for (auto v : values) sum += v; // halves win, floaters suck
	// 
	// changes stuff diametrically, chars win, floaters are not left behind
	//for (auto v : values) sum += v*values.front();

	size_t delta = tick() - start;
	oss << sum;
	return delta;
}

template<typename T>
void test(const char msg[], size_t loops, size_t count)
{
	std::vector<size_t> times(loops);
	std::vector<T> values(count);
	size_t avg = 0;
	for (size_t& t: times) avg += t = test_one<T>(values);
	for (size_t t : times) printf(" %i", t);
	printf("|%s %i\n", msg, avg / loops);
}

void test_all(size_t loops, size_t count)
{
	test<char>("c", loops, count);
	test<short>("h", loops, count);
	test<unsigned short>("uh", loops, count);
	test<int>("i", loops, count);
	test<unsigned int>("ui", loops, count);
	test<float>("f", loops, count);
	test<double>("d", loops, count);
	test<size_t>("s", loops, count);
}

void main()
{
	::srand((int)::time(nullptr));
	size_t loops = 10;
	for (auto count : { 1000, 10*1000, 100 * 1000, 1000 * 1000 })
	{
		printf("\nTesting %i values %i times\n", count, loops);
		test_all(loops, count);
	}
	printf("%i", oss.str().length());
}