sum.cpp

#include <iostream>
#include <chrono>

const int N = 4000;
int data[N][N];

void generate_data() {
    for (int i = 0; i < N; i++)
        for (int j = 0; j < N; j++)
            data[i][j] = (i + j) & 3;
}

int sum1() {
    int result = 0;
    // add row-by-row
    for (int x = 0; x < N * N; x+=4 /* 4 values at a time */) {
        int row = x / N;
        int col = x % N;
        result += data[row][col] + data[row][col + 1] +
                  data[row][col + 2] + data[row][col + 3];
    }
    return result;
}
int sum2() {
    int result = 0;
    // add column-by-column
    for (int x = 0; x < N * N; x+=4 /* 4 values at a time */) {
        int row = x % N;
        int col = x / N;
        result += data[row][col] + data[row + 1][col] +
                  data[row + 2][col] + data[row + 3][col];
    }
    return result;
}

void measure_sum1() {
    std::chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
    int result = sum1();
    std::chrono::steady_clock::time_point end = std::chrono::steady_clock::now();

    std::cout << "sum1: result=" << result << ", duration=" << std::chrono::duration_cast<std::chrono::microseconds>(end - begin).count() << "[µs]" << std::endl;
}

void measure_sum2() {
    std::chrono::steady_clock::time_point begin = std::chrono::steady_clock::now();
    int result = sum2();
    std::chrono::steady_clock::time_point end = std::chrono::steady_clock::now();

    std::cout << "sum2: result=" << result << ", duration=" << std::chrono::duration_cast<std::chrono::microseconds>(end - begin).count() << "[µs]" << std::endl;
}

int main() {
    generate_data();
    measure_sum1();
    measure_sum2();
    return 0;
}