EgorBot for hamarb123 in #99596

[EgorBot]

Processing dotnet/runtime#99596 (comment) command:

Command

-intel -amd

using BenchmarkDotNet.Attributes;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;

public class Benchmarks
{
    private byte[] source = null!;

    [Params(64, 128, 256, 1024, 4096, 16384, 65536, 1 << 20)]
    public int N;

    [GlobalSetup]
    public void Setup()
    {
        source = new byte[N];
        new Random(42).NextBytes(source);
    }

    [Benchmark]
    public byte TestWork() => Work(ref MemoryMarshal.GetArrayDataReference(source), (uint)source.Length);

    [Benchmark]
    public byte TestWorkWithZeroing() => WorkWithZeroing(ref MemoryMarshal.GetArrayDataReference(source), (uint)source.Length);

    //test to see if the new codegen using shuf over perm is meaningfully faster
    //this should be more pipelineable than the other test I ran

    public static byte Work(ref byte b, nuint size)
    {
        Vector256<byte> result = Vector256<byte>.Zero;
        while (size >= Vector256<byte>.Count + 7)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            Vector256<byte> v1 = Vector256.LoadUnsafe(ref b, 1);
            Vector256<byte> v2 = Vector256.LoadUnsafe(ref b, 2);
            Vector256<byte> v3 = Vector256.LoadUnsafe(ref b, 3);
            Vector256<byte> v4 = Vector256.LoadUnsafe(ref b, 4);
            Vector256<byte> v5 = Vector256.LoadUnsafe(ref b, 5);
            Vector256<byte> v6 = Vector256.LoadUnsafe(ref b, 6);
            Vector256<byte> v7 = Vector256.LoadUnsafe(ref b, 7);
            v0 = Vector256.Shuffle(v0, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v1 = Vector256.Shuffle(v1, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v2 = Vector256.Shuffle(v2, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v3 = Vector256.Shuffle(v3, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v4 = Vector256.Shuffle(v4, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v5 = Vector256.Shuffle(v5, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v6 = Vector256.Shuffle(v6, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v7 = Vector256.Shuffle(v7, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            result += ((v0 ^ v1) - (v2 + v3)) ^ ((v4 + ~v5) + (v6 + v7));
            b = ref Unsafe.Add(ref b, 8);
            size -= 8;
        }
        while (size >= Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Shuffle((byte)15, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            result += v0;
            b = ref Unsafe.Add(ref b, 1);
            size--;
        }
        return Vector256.Sum(result);
    }

    public static byte WorkWithZeroing(ref byte b, nuint size)
    {
        Vector256<byte> result = Vector256<byte>.Zero;
        while (size >= Vector256<byte>.Count + 7)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            Vector256<byte> v1 = Vector256.LoadUnsafe(ref b, 1);
            Vector256<byte> v2 = Vector256.LoadUnsafe(ref b, 2);
            Vector256<byte> v3 = Vector256.LoadUnsafe(ref b, 3);
            Vector256<byte> v4 = Vector256.LoadUnsafe(ref b, 4);
            Vector256<byte> v5 = Vector256.LoadUnsafe(ref b, 5);
            Vector256<byte> v6 = Vector256.LoadUnsafe(ref b, 6);
            Vector256<byte> v7 = Vector256.LoadUnsafe(ref b, 7);
            v0 = Vector256.Shuffle(v0, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v1 = Vector256.Shuffle(v1, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v2 = Vector256.Shuffle(v2, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v3 = Vector256.Shuffle(v3, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v4 = Vector256.Shuffle(v4, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v5 = Vector256.Shuffle(v5, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v6 = Vector256.Shuffle(v6, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            v7 = Vector256.Shuffle(v7, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            result += ((v0 ^ v1) - (v2 + v3)) ^ ((v4 + ~v5) + (v6 + v7));
            b = ref Unsafe.Add(ref b, 8);
            size -= 8;
        }
        while (size >= Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Shuffle((byte)99, 0, 14, 1, 13, 2, 12, 3, 11, 4, 10, 5, 9, 6, 8, 7, 31, 16, 30, 17, 29, 18, 28, 19, 27, 20, 26, 21, 25, 22, 24, 23));
            result += v0;
            b = ref Unsafe.Add(ref b, 1);
            size--;
        }
        return Vector256.Sum(result);
    }
}

(EgorBot will reply in this issue)

[EgorBot]

❌ Failed on AzureCascadeLake: Job failed, see logs.

cc @hamarb123 (logs). EgorBot manual: link.

[EgorBot]

❌ Failed on AzureGenoa: Job failed, see logs.

cc @hamarb123 (logs). EgorBot manual: link.