JIT: Potential bad codegen with AlignRight

[jakobbotsch]

// Generated by Fuzzlyn v1.7 on 2024-07-25 11:40:55
// Run on X64 Windows
// Seed: 418121815423725559-vectort,vector128,vector256,x86aes,x86avx,x86avx2,x86bmi1,x86bmi1x64,x86bmi2,x86bmi2x64,x86fma,x86lzcnt,x86lzcntx64,x86pclmulqdq,x86popcnt,x86popcntx64,x86sse,x86ssex64,x86sse2,x86sse2x64,x86sse3,x86sse41,x86sse41x64,x86sse42,x86sse42x64,x86ssse3,x86x86base
// Reduced from 72.7 KiB to 1.0 KiB in 00:01:55
// Debug: Outputs <0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0>
// Release: Outputs <1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
using System;
using System.Runtime.CompilerServices;
using System.Numerics;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;

public struct S3
{
    public byte F0;
}

public class Program
{
    public static IRuntime s_rt;
    public static S3 s_3;
    public static void Main()
    {
        s_rt = new Runtime();
        var vr15 = (ushort)0;
        var vr16 = Vector256.CreateScalar(vr15);
        var vr17 = Vector256.Create<ushort>(1);
        var vr18 = (ushort)0;
        var vr19 = Vector256.CreateScalar(vr18);
        var vr20 = s_3.F0;
        var vr21 = Avx2.AlignRight(vr17, vr19, vr20);
        M6(vr16, vr21);
    }

    public static void M6(Vector256<ushort> arg1, Vector256<ushort> arg3)
    {
        s_rt.WriteLine(arg1);
    }
}

public interface IRuntime
{
    void WriteLine<T>(T value);
}

public class Runtime : IRuntime
{
    public void WriteLine<T>(T value) => System.Console.WriteLine(value);
}

[dotnet-policy-service]

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

[tannergooding]

This one appears to be related to IRuntime.WriteLine being a generic virtual:

This repros (modified to be in the format that the JIT tests expect):

// Generated by Fuzzlyn v1.7 on 2024-07-25 11:40:55
// Run on X64 Windows
// Seed: 418121815423725559-vectort,vector128,vector256,x86aes,x86avx,x86avx2,x86bmi1,x86bmi1x64,x86bmi2,x86bmi2x64,x86fma,x86lzcnt,x86lzcntx64,x86pclmulqdq,x86popcnt,x86popcntx64,x86sse,x86ssex64,x86sse2,x86sse2x64,x86sse3,x86sse41,x86sse41x64,x86sse42,x86sse42x64,x86ssse3,x86x86base
// Reduced from 72.7 KiB to 1.0 KiB in 00:01:55
// Debug: Outputs <0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0>
// Release: Outputs <1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
using System;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
using Xunit;

public struct S3
{
    public byte F0;
}

public class Runtime_105468
{
    public static IRuntime s_rt;
    public static S3 s_3;

    [Fact]
    public static void TestEntryPoint()
    {
        s_rt = new Runtime();
        var vr15 = (ushort)0;
        var vr16 = Vector256.CreateScalar(vr15);
        var vr17 = Vector256.Create<ushort>(1);
        var vr18 = (ushort)0;
        var vr19 = Vector256.CreateScalar(vr18);
        var vr20 = s_3.F0;
        var vr21 = Avx2.AlignRight(vr17, vr19, vr20);
        M6(vr16, vr21);
    }

    private static void M6(Vector256<ushort> arg1, Vector256<ushort> arg3)
    {
        s_rt.Validate<ushort>(arg1);
    }
}

public interface IRuntime
{
    void Validate<T>(Vector256<T> value);
}

public class Runtime : IRuntime
{
    public void Validate<T>(Vector256<T> value) => Assert.Equal(Vector256<T>.Zero, value);
}

But if we make Validate non-generic and just take it as Vector256<ushort> directly, then it no longer fails.

[tannergooding]

So not generic virtuals, but rather M6 not being inlined repros. The generic virtual just blocked M6 from being inlined