Use IL shuffle thunk on all platforms

[huoyaoyuan]

Open early to see the benchmark and tests.

[azure-pipelines]

Azure Pipelines:
Successfully started running 3 pipeline(s).
13 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[dotnet-policy-service]

Tagging subscribers to this area: @agocke
See info in area-owners.md if you want to be subscribed.

[huoyaoyuan]

@EgorBot -intel -amd -arm64

using System;
using BenchmarkDotNet.Attributes;

namespace BenchmarkGround
{
    public struct GCStruct
    {
        public object a, b, c, d, e, f, h, g;
    }

    public class Bench
    {
        private readonly object obj = new object();

        private readonly Guid guid = Guid.NewGuid();

        private readonly GCStruct gcStruct = new GCStruct
        {
            a = new object(),
            g = new object()
        };

        private Action<int> singleArg = SingleArg_Static;
        public static void SingleArg_Static(int a) { }

        private Func<int, double, Guid, GCStruct, object, double> manyArg_RetFPU = ManyArg_RetFPU_Static;
        public static double ManyArg_RetFPU_Static(int a, double b, Guid c, GCStruct d, object e) => 1.0;

        [Benchmark]
        public void SingleArg() => singleArg(0);

        [Benchmark]
        public double ManyArg_RetFPU() => manyArg_RetFPU(1, 2.0, guid, gcStruct, obj);
    }
}

[huoyaoyuan]

The initiative is getting such unbelievable benchmark result locally:

Method	Job	Toolchain	Mean	Error	StdDev	Ratio
SingleArg	Job-NHYPTB	\PR\corerun.exe	1.250 ns	0.0073 ns	0.0068 ns	1.00
SingleArg	Job-ZICBXN	\main\corerun.exe	1.248 ns	0.0120 ns	0.0112 ns	1.00
ManyArg_RetFPU	Job-NHYPTB	\PR\corerun.exe	1.857 ns	0.0143 ns	0.0134 ns	0.02
ManyArg_RetFPU	Job-ZICBXN	\main\corerun.exe	92.010 ns	0.5680 ns	0.5313 ns	1.00

The 90ns result also reproduces with 10.0.10.

It only reproduces on Windows. Even for 7 args on Linux, the time is still <3ns.

[huoyaoyuan]

@EgorBot -windows_x64 -linux_x64 -arm64

using System;
using BenchmarkDotNet.Attributes;

namespace BenchmarkGround
{
    public struct GCStruct
    {
        public object a, b, c, d, e, f, h, g;
    }

    public class Bench
    {
        private readonly object obj = new object();

        private readonly Guid guid = Guid.NewGuid();

        private readonly GCStruct gcStruct = new GCStruct
        {
            a = new object(),
            g = new object()
        };

        private Action<int> singleArg = SingleArg_Static;
        public static void SingleArg_Static(int a) { }

        private Func<int, double, Guid, GCStruct, object, double> manyArg_RetFPU = ManyArg_RetFPU_Static;
        public static double ManyArg_RetFPU_Static(int a, double b, Guid c, GCStruct d, object e) => 1.0;

        private Func<int, double, Guid, GCStruct, object, Int128, UInt128, double> moreArg_RetFPU = MoreArg_RetFPU_Static;
        public static double MoreArg_RetFPU_Static(int a, double b, Guid c, GCStruct d, object e, Int128 f, UInt128 g) => 1.0;

        [Benchmark]
        public void SingleArg() => singleArg(0);

        [Benchmark]
        public double ManyArg_RetFPU() => manyArg_RetFPU(1, 2.0, guid, gcStruct, obj);

        [Benchmark]
        public double MoreArg_RetFPU() => moreArg_RetFPU(1, 2.0, guid, gcStruct, obj, 0, 0);
    }
}

[huoyaoyuan]

Closing as problem in both tests and performance. Still curious why the 90ns case happens on local Windows.