EgorBot for hamarb123 in #99596 #279

EgorBot · 2025-01-30T00:22:19Z

Processing dotnet/runtime#99596 (comment) command:

Command

-intel -amd

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

public unsafe class Benchmarks
{
    private byte* source = null;

    [Params(32, 64, 128, 256, 1024, 4096, 16384, 65536, 1 << 20)]
    public int N;

    [GlobalSetup]
    public void Setup()
    {
        source = (byte*)NativeMemory.AlignedAlloc((uint)N, 4096);
        new Random(42).NextBytes(new Span<byte>(source, N));
    }

    [GlobalCleanup]
    public void Teardown()
    {
        NativeMemory.AlignedFree(source);
        source = null;
    }

    [Benchmark]
    public byte TestWorkPipelineable() => WorkPipelineable(ref *source, (uint)N);

    [Benchmark]
    public float TestWorkDependencyChain() => WorkDependencyChain(ref *source, (uint)N);

    [Benchmark]
    public byte TestWorkPipelineableWithZeroing() => WorkPipelineableWithZeroing(ref *source, (uint)N);

    [Benchmark]
    public float TestWorkDependencyChainWithZeroing() => WorkDependencyChainWithZeroing(ref *source, (uint)N);

    //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 WorkPipelineable(ref byte b, nuint size)
    {
        Vector256<byte> result = Vector256<byte>.Zero;
        while (size >= (uint)Vector256<byte>.Count + 7u)
        {
            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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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 >= (uint)Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Create((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 WorkPipelineableWithZeroing(ref byte b, nuint size)
    {
        Vector256<byte> result = Vector256<byte>.Zero;
        while (size >= (uint)Vector256<byte>.Count + 7u)
        {
            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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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.Create((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 >= (uint)Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Create((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);
    }

    //this test should be very latency bound as it has a long dependency chain

    public static float WorkDependencyChain(ref byte b, nuint size)
    {
        Vector256<float> result = Vector256<float>.Zero;
        while (size >= (uint)Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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.AsSingle();
            b = ref Unsafe.Add(ref b, 1);
            size--;
        }
        return Vector256.Sum(result);
    }

    public static float WorkDependencyChainWithZeroing(ref byte b, nuint size)
    {
        Vector256<float> result = Vector256<float>.Zero;
        while (size >= (uint)Vector256<byte>.Count)
        {
            Vector256<byte> v0 = Vector256.LoadUnsafe(ref b);
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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));
            v0 = (v0.AsSingle() + Vector256.Create(1.0f)).AsByte();
            v0 = Vector256.Shuffle(v0, Vector256.Create((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.AsSingle();
            b = ref Unsafe.Add(ref b, 1);
            size--;
        }
        return Vector256.Sum(result);
    }
}

(EgorBot will reply in this issue)

EgorBot · 2025-01-30T01:14:46Z

Benchmark results on `linux_azure_genoa`

BenchmarkDotNet v0.14.0, Ubuntu 24.04.1 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 8 logical and 4 physical cores
  Job-QBCGOJ : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-FLOZDB : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

Method	Toolchain	N	Mean	Error	Ratio
TestWorkPipelineable	Main	32	1.621 ns	0.0318 ns	1.00
TestWorkPipelineable	PR	32	1.299 ns	0.0241 ns	0.80

TestWorkDependencyChain	Main	32	4.734 ns	0.0452 ns	1.00
TestWorkDependencyChain	PR	32	3.813 ns	0.0274 ns	0.81

TestWorkPipelineableWithZeroing	Main	32	1.231 ns	0.0274 ns	1.00
TestWorkPipelineableWithZeroing	PR	32	1.315 ns	0.0213 ns	1.07

TestWorkDependencyChainWithZeroing	Main	32	6.359 ns	0.0398 ns	1.00
TestWorkDependencyChainWithZeroing	PR	32	3.858 ns	0.0425 ns	0.61

TestWorkPipelineable	Main	64	9.140 ns	0.1290 ns	1.00
TestWorkPipelineable	PR	64	9.433 ns	0.0861 ns	1.03

TestWorkDependencyChain	Main	64	99.247 ns	0.6407 ns	1.00
TestWorkDependencyChain	PR	64	89.515 ns	0.5800 ns	0.90

TestWorkPipelineableWithZeroing	Main	64	11.335 ns	0.0852 ns	1.00
TestWorkPipelineableWithZeroing	PR	64	9.618 ns	0.0724 ns	0.85

TestWorkDependencyChainWithZeroing	Main	64	153.947 ns	1.3082 ns	1.00
TestWorkDependencyChainWithZeroing	PR	64	90.125 ns	0.5444 ns	0.59

TestWorkPipelineable	Main	128	26.055 ns	0.3297 ns	1.00
TestWorkPipelineable	PR	128	27.572 ns	0.1980 ns	1.06

TestWorkDependencyChain	Main	128	289.060 ns	1.8196 ns	1.00
TestWorkDependencyChain	PR	128	259.310 ns	1.7688 ns	0.90

TestWorkPipelineableWithZeroing	Main	128	33.047 ns	0.2265 ns	1.00
TestWorkPipelineableWithZeroing	PR	128	27.831 ns	0.2607 ns	0.84

TestWorkDependencyChainWithZeroing	Main	128	451.250 ns	2.7357 ns	1.00
TestWorkDependencyChainWithZeroing	PR	128	260.039 ns	1.4276 ns	0.58

TestWorkPipelineable	Main	256	60.098 ns	0.5447 ns	1.00
TestWorkPipelineable	PR	256	60.034 ns	0.8126 ns	1.00

TestWorkDependencyChain	Main	256	670.485 ns	5.3941 ns	1.00
TestWorkDependencyChain	PR	256	605.254 ns	3.8565 ns	0.90

TestWorkPipelineableWithZeroing	Main	256	72.381 ns	0.8625 ns	1.00
TestWorkPipelineableWithZeroing	PR	256	60.506 ns	0.6167 ns	0.84

TestWorkDependencyChainWithZeroing	Main	256	1,036.934 ns	4.4639 ns	1.00
TestWorkDependencyChainWithZeroing	PR	256	610.374 ns	3.3020 ns	0.59

TestWorkPipelineable	Main	1024	268.519 ns	5.1901 ns	1.00
TestWorkPipelineable	PR	1024	260.932 ns	5.2617 ns	0.97

TestWorkDependencyChain	Main	1024	2,948.048 ns	27.4073 ns	1.00
TestWorkDependencyChain	PR	1024	2,652.525 ns	9.7212 ns	0.90

TestWorkPipelineableWithZeroing	Main	1024	305.643 ns	3.1595 ns	1.00
TestWorkPipelineableWithZeroing	PR	1024	261.943 ns	4.3686 ns	0.86

TestWorkDependencyChainWithZeroing	Main	1024	4,587.634 ns	10.1879 ns	1.00
TestWorkDependencyChainWithZeroing	PR	1024	2,678.700 ns	22.9431 ns	0.58

TestWorkPipelineable	Main	4096	1,101.384 ns	19.5415 ns	1.00
TestWorkPipelineable	PR	4096	1,078.131 ns	10.5645 ns	0.98

TestWorkDependencyChain	Main	4096	12,134.039 ns	51.5571 ns	1.00
TestWorkDependencyChain	PR	4096	10,982.100 ns	50.7779 ns	0.91

TestWorkPipelineableWithZeroing	Main	4096	1,241.756 ns	5.0880 ns	1.00
TestWorkPipelineableWithZeroing	PR	4096	1,070.295 ns	6.2527 ns	0.86

TestWorkDependencyChainWithZeroing	Main	4096	19,118.332 ns	208.6301 ns	1.00
TestWorkDependencyChainWithZeroing	PR	4096	10,948.115 ns	95.9672 ns	0.57

TestWorkPipelineable	Main	16384	4,429.837 ns	48.8727 ns	1.00
TestWorkPipelineable	PR	16384	4,232.977 ns	26.0471 ns	0.96

TestWorkDependencyChain	Main	16384	48,351.725 ns	248.0566 ns	1.00
TestWorkDependencyChain	PR	16384	44,271.369 ns	260.4794 ns	0.92

TestWorkPipelineableWithZeroing	Main	16384	5,017.464 ns	43.0286 ns	1.00
TestWorkPipelineableWithZeroing	PR	16384	4,290.491 ns	45.4591 ns	0.86

TestWorkDependencyChainWithZeroing	Main	16384	76,137.381 ns	561.5843 ns	1.00
TestWorkDependencyChainWithZeroing	PR	16384	43,992.386 ns	355.6288 ns	0.58

TestWorkPipelineable	Main	65536	17,712.365 ns	115.5184 ns	1.00
TestWorkPipelineable	PR	65536	17,300.111 ns	76.8237 ns	0.98

TestWorkDependencyChain	Main	65536	194,939.787 ns	555.3184 ns	1.00
TestWorkDependencyChain	PR	65536	177,258.114 ns	995.2090 ns	0.91

TestWorkPipelineableWithZeroing	Main	65536	19,923.054 ns	159.5374 ns	1.00
TestWorkPipelineableWithZeroing	PR	65536	17,355.804 ns	107.9378 ns	0.87

TestWorkDependencyChainWithZeroing	Main	65536	303,664.000 ns	1,977.6564 ns	1.00
TestWorkDependencyChainWithZeroing	PR	65536	176,106.015 ns	760.8893 ns	0.58

TestWorkPipelineable	Main	1048576	285,622.234 ns	1,619.0143 ns	1.00
TestWorkPipelineable	PR	1048576	276,786.718 ns	2,967.4042 ns	0.97

TestWorkDependencyChain	Main	1048576	3,157,656.744 ns	15,055.0844 ns	1.00
TestWorkDependencyChain	PR	1048576	2,839,843.859 ns	23,074.2656 ns	0.90

TestWorkPipelineableWithZeroing	Main	1048576	322,385.050 ns	2,905.5243 ns	1.00
TestWorkPipelineableWithZeroing	PR	1048576	274,143.051 ns	824.2073 ns	0.85

TestWorkDependencyChainWithZeroing	Main	1048576	4,907,468.092 ns	33,081.6507 ns	1.00
TestWorkDependencyChainWithZeroing	PR	1048576	2,806,245.509 ns	14,689.7401 ns	0.57

BDN_Artifacts.zip

cc @hamarb123 (agent_logs.txt). EgorBot manual: link.

EgorBot · 2025-01-30T01:15:51Z

Benchmark results on `linux_azure_cascadelake`

BenchmarkDotNet v0.14.0, Ubuntu 24.04.1 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 8 logical and 4 physical cores
  Job-IYBOSE : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-HAUWTU : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

Method	Toolchain	N	Mean	Error	Ratio
TestWorkPipelineable	Main	32	2.015 ns	0.0008 ns	1.00
TestWorkPipelineable	PR	32	1.591 ns	0.0565 ns	0.79

TestWorkDependencyChain	Main	32	4.129 ns	0.0013 ns	1.00
TestWorkDependencyChain	PR	32	3.308 ns	0.0007 ns	0.80

TestWorkPipelineableWithZeroing	Main	32	1.438 ns	0.0004 ns	1.00
TestWorkPipelineableWithZeroing	PR	32	1.546 ns	0.0069 ns	1.07

TestWorkDependencyChainWithZeroing	Main	32	6.221 ns	0.0010 ns	1.00
TestWorkDependencyChainWithZeroing	PR	32	3.309 ns	0.0005 ns	0.53

TestWorkPipelineable	Main	64	9.347 ns	0.0017 ns	1.00
TestWorkPipelineable	PR	64	9.299 ns	0.0062 ns	0.99

TestWorkDependencyChain	Main	64	81.480 ns	0.0096 ns	1.00
TestWorkDependencyChain	PR	64	71.044 ns	0.0065 ns	0.87

TestWorkPipelineableWithZeroing	Main	64	12.317 ns	0.0018 ns	1.00
TestWorkPipelineableWithZeroing	PR	64	9.306 ns	0.0140 ns	0.76

TestWorkDependencyChainWithZeroing	Main	64	137.336 ns	0.0807 ns	1.00
TestWorkDependencyChainWithZeroing	PR	64	69.753 ns	0.0124 ns	0.51

TestWorkPipelineable	Main	128	27.244 ns	0.0051 ns	1.00
TestWorkPipelineable	PR	128	25.090 ns	0.0063 ns	0.92

TestWorkDependencyChain	Main	128	242.621 ns	0.0292 ns	1.00
TestWorkDependencyChain	PR	128	208.483 ns	0.0272 ns	0.86

TestWorkPipelineableWithZeroing	Main	128	33.718 ns	0.0251 ns	1.00
TestWorkPipelineableWithZeroing	PR	128	25.081 ns	0.0061 ns	0.74

TestWorkDependencyChainWithZeroing	Main	128	409.063 ns	0.0501 ns	1.00
TestWorkDependencyChainWithZeroing	PR	128	208.404 ns	0.0326 ns	0.51

TestWorkPipelineable	Main	256	63.669 ns	0.0034 ns	1.00
TestWorkPipelineable	PR	256	57.261 ns	0.0100 ns	0.90

TestWorkDependencyChain	Main	256	568.083 ns	0.0742 ns	1.00
TestWorkDependencyChain	PR	256	489.799 ns	0.0676 ns	0.86

TestWorkPipelineableWithZeroing	Main	256	76.111 ns	0.0287 ns	1.00
TestWorkPipelineableWithZeroing	PR	256	57.279 ns	0.0084 ns	0.75

TestWorkDependencyChainWithZeroing	Main	256	954.488 ns	0.0915 ns	1.00
TestWorkDependencyChainWithZeroing	PR	256	489.585 ns	0.0868 ns	0.51

TestWorkPipelineable	Main	1024	284.233 ns	0.0333 ns	1.00
TestWorkPipelineable	PR	1024	249.602 ns	0.0688 ns	0.88

TestWorkDependencyChain	Main	1024	2,502.301 ns	0.2558 ns	1.00
TestWorkDependencyChain	PR	1024	2,154.098 ns	0.1747 ns	0.86

TestWorkPipelineableWithZeroing	Main	1024	332.632 ns	0.2007 ns	1.00
TestWorkPipelineableWithZeroing	PR	1024	249.624 ns	0.0426 ns	0.75

TestWorkDependencyChainWithZeroing	Main	1024	4,214.613 ns	0.3169 ns	1.00
TestWorkDependencyChainWithZeroing	PR	1024	2,154.157 ns	0.1785 ns	0.51

TestWorkPipelineable	Main	4096	1,173.618 ns	0.2041 ns	1.00
TestWorkPipelineable	PR	4096	1,027.609 ns	0.6465 ns	0.88

TestWorkDependencyChain	Main	4096	10,231.266 ns	1.5051 ns	1.00
TestWorkDependencyChain	PR	4096	8,823.374 ns	2.7114 ns	0.86

TestWorkPipelineableWithZeroing	Main	4096	1,362.735 ns	0.6094 ns	1.00
TestWorkPipelineableWithZeroing	PR	4096	1,027.476 ns	0.6673 ns	0.75

TestWorkDependencyChainWithZeroing	Main	4096	17,255.498 ns	1.5601 ns	1.00
TestWorkDependencyChainWithZeroing	PR	4096	8,818.680 ns	3.4113 ns	0.51

TestWorkPipelineable	Main	16384	4,747.632 ns	0.5336 ns	1.00
TestWorkPipelineable	PR	16384	3,993.880 ns	56.8574 ns	0.84

TestWorkDependencyChain	Main	16384	41,216.636 ns	17.4360 ns	1.00
TestWorkDependencyChain	PR	16384	35,451.041 ns	5.9434 ns	0.86

TestWorkPipelineableWithZeroing	Main	16384	5,505.710 ns	0.6226 ns	1.00
TestWorkPipelineableWithZeroing	PR	16384	3,944.768 ns	6.8011 ns	0.72

TestWorkDependencyChainWithZeroing	Main	16384	69,444.435 ns	9.2554 ns	1.00
TestWorkDependencyChainWithZeroing	PR	16384	35,446.611 ns	4.0584 ns	0.51

TestWorkPipelineable	Main	65536	19,051.137 ns	2.6371 ns	1.00
TestWorkPipelineable	PR	65536	15,802.064 ns	42.4753 ns	0.83

TestWorkDependencyChain	Main	65536	165,448.616 ns	145.9864 ns	1.00
TestWorkDependencyChain	PR	65536	142,138.702 ns	112.8088 ns	0.86

TestWorkPipelineableWithZeroing	Main	65536	22,090.972 ns	1.8095 ns	1.00
TestWorkPipelineableWithZeroing	PR	65536	15,786.253 ns	32.6605 ns	0.71

TestWorkDependencyChainWithZeroing	Main	65536	278,195.406 ns	29.7947 ns	1.00
TestWorkDependencyChainWithZeroing	PR	65536	142,244.863 ns	202.4602 ns	0.51

TestWorkPipelineable	Main	1048576	305,169.314 ns	37.8255 ns	1.00
TestWorkPipelineable	PR	1048576	252,216.886 ns	86.1144 ns	0.83

TestWorkDependencyChain	Main	1048576	2,694,406.521 ns	24,977.6341 ns	1.00
TestWorkDependencyChain	PR	1048576	2,330,393.989 ns	22,686.3393 ns	0.86

TestWorkPipelineableWithZeroing	Main	1048576	353,888.669 ns	28.2779 ns	1.00
TestWorkPipelineableWithZeroing	PR	1048576	252,309.066 ns	57.2610 ns	0.71

TestWorkDependencyChainWithZeroing	Main	1048576	4,454,224.889 ns	1,140.0706 ns	1.00
TestWorkDependencyChainWithZeroing	PR	1048576	2,309,306.266 ns	5,442.9339 ns	0.52

BDN_Artifacts.zip

cc @hamarb123 (agent_logs.txt). EgorBot manual: link.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

EgorBot for hamarb123 in #99596 #279

EgorBot for hamarb123 in #99596 #279

EgorBot commented Jan 30, 2025

EgorBot commented Jan 30, 2025

EgorBot commented Jan 30, 2025

EgorBot for hamarb123 in #99596 #279

EgorBot for hamarb123 in #99596 #279

Comments

EgorBot commented Jan 30, 2025

EgorBot commented Jan 30, 2025

Benchmark results on linux_azure_genoa

EgorBot commented Jan 30, 2025

Benchmark results on linux_azure_cascadelake

Benchmark results on `linux_azure_genoa`

Benchmark results on `linux_azure_cascadelake`