SSE_CompareEqual_ushort.cs

using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Jobs;
using BenchmarkDotNet.Running;

#pragma warning disable CS8618

var summary = BenchmarkRunner.Run<MainTest>();

[SimpleJob(RuntimeMoniker.Net70)]
[WarmupCount(2)]
[IterationCount(2)]
// [MemoryDiagnoser]
public class MainTest
{
    const int ITERATIONS = 10000;

    [Params(128, 256, 512, 1024 * 1, 1024 * 2, 1024 * 3, 1024 * 4, 1024 * 5, 1024 * 6, 1024 * 7, 1024 * 8)]
    public int N { get; set; }

    ushort[] ints;
    ushort   findValue;

    [IterationSetup]
    public void Setup()
    {
        ints      = Enumerable.Range(0, N).Select(p => (ushort) p).ToArray();
        findValue = ints[^3];
    }

    [Benchmark]
    public void FindSimple()
    {
        for (var iteration = 0; iteration < ITERATIONS; iteration++)
        {
            FindSimple_(ints, findValue);
        }
    }

    [Benchmark]
    public void FindSSE_128()
    {
        for (var iteration = 0; iteration < ITERATIONS; iteration++)
            FindSSE_128_(ints, findValue);
    }

    [Benchmark]
    public void FindSSE_256()
    {
        for (var iteration = 0; iteration < ITERATIONS; iteration++)
            FindSSE_256_(ints, findValue);
    }


    static int FindSimple_(ushort[] data, int value)
    {
        foreach (var i in data)
            if (i == value)
                return i;
        return -1;
    }

    static int FindSSE_128_(ushort[] data, ushort value)
    {
        var ints = MemoryMarshal.Cast<ushort, Vector128<ushort>>(data);

        var compareValue = Vector128.Create(value);
        var vectorLength = Vector128<ushort>.Count;

        for (var i = 0; i < ints.Length; i++)
        {
            var result = Sse2.CompareEqual(ints[i], compareValue);
            if (result == Vector128<ushort>.Zero) continue;

            for (var k = 0; k < vectorLength; k++)
                if (result.GetElement(k) != 0)
                    return i * vectorLength + k;
        }

        for (var i = ints.Length * vectorLength; i < data.Length; i++)
            if (data[i] == value)
                return i;

        return -1;
    }

    static int FindSSE_256_(ushort[] data, ushort value)
    {
        var ints = MemoryMarshal.Cast<ushort, Vector256<ushort>>(data);

        var compareValue = Vector256.Create(value);
        var vectorLength = Vector256<int>.Count;

        for (var i = 0; i < ints.Length; i++)
        {
            var result = Avx2.CompareEqual(ints[i], compareValue);
            if (result == Vector256<ushort>.Zero) continue;

            for (var k = 0; k < vectorLength; k++)
                if (result.GetElement(k) != 0)
                    return i * vectorLength + k;
        }

        for (var i = ints.Length * vectorLength; i < data.Length; i++)
            if (data[i] == value)
                return i;

        return -1;
    }
}