Illustrating cache line sharing penalty

Cacheline.cs

using System.Diagnostics;

// Change to 128 on ARM64 macOS
const int CACHELINE = 64;
const int iterations = 10_000_000;
var coreCount = Environment.ProcessorCount;

Console.WriteLine($"""
    Time to increment an int {iterations} times on {coreCount} cores.
    Cache line sharing coefficient is the count of cores that have to compete for the same cache line.
    --------------------------------------------------------------------------------------------------
    """);

var arrayLength = coreCount;
var maxPadding = CACHELINE * coreCount / sizeof(int);
while (arrayLength <= maxPadding)
{
    var bytesPerInt = arrayLength / sizeof(int);

    Console.WriteLine(
        $"Padding: {bytesPerInt} bytes\n" +
        $"Cache line sharing coefficient: {(float)CACHELINE / bytesPerInt}x");

    var counters = new int[arrayLength];
    var start = Stopwatch.GetTimestamp();

    Parallel.For(0, coreCount, core =>
    {
        var stride = arrayLength / coreCount;
        for (var i = 0; i < iterations; i++)
        {
            counters[stride * core]++;
        }
    });

    Console.WriteLine(
        $"Time: {Stopwatch.GetElapsedTime(start).TotalMilliseconds}ms\n");

    arrayLength += coreCount;
}