`(x >> N) << N` should reduce to a mask

Description

(Assuming x is unsigned...)

A somewhat common pattern to mask off some least-significant bits is to do (x >> N) << N. However, CoreCLR lowers this exactly as written instead of a slightly more performant mask: x & ~((1 << N) - 1).

Assembly

https://godbolt.org/z/5zsWaes4q

internal static class Klass
{
    public static uint Test(uint x)
    {
        return (x >> 5) << 5;
    }
    public static uint Test2(uint x)
    {
        return (x / 32) * 32;
    }
    public static uint Test3(uint x)
    {
        return x & ~31u;
    }
}

Klass:Test(uint):uint (FullOpts):
       mov      eax, ecx
       shr      eax, 5
       shl      eax, 5
       ret

Klass:Test2(uint):uint (FullOpts):
       mov      eax, ecx
       shr      eax, 5
       shl      eax, 5
       ret

Klass:Test3(uint):uint (FullOpts):
       mov      eax, ecx
       and      eax, -32
       ret

Benchmark

public class Benchmarks
{
    static Benchmarks()
    {
        Random r = new(42);
        uint[] tests = new uint[100];
        r.GetItems(tests, 100);
        Tests = tests;
    }

    private static readonly uint[] Tests;

    [Benchmark]
    public uint TestShift()
    {
        uint[] tests = Tests;
        uint accum = 0;
        for (int i = 0; i < tests.Length; i++)
            accum += (tests[i] >> 5) << 5;
        return accum;
    }

    [Benchmark]
    public uint TestMask()
    {
        uint[] tests = Tests;
        uint accum = 0;
        for (int i = 0; i < tests.Length; i++)
            accum += tests[i] & ~31u;
        return accum;
    }
}

BenchmarkDotNet v0.15.2, Windows 11 (10.0.26100.4770/24H2/2024Update/HudsonValley)
Intel Core i9-10900X CPU 3.70GHz, 1 CPU, 20 logical and 10 physical cores
.NET SDK 9.0.304
  [Host]     : .NET 8.0.19 (8.0.1925.36514), X64 RyuJIT AVX-512F+CD+BW+DQ+VL
  DefaultJob : .NET 8.0.19 (8.0.1925.36514), X64 RyuJIT AVX-512F+CD+BW+DQ+VL


| Method    | Mean     | Error    | StdDev   | Code Size | Allocated |
|---------- |---------:|---------:|---------:|----------:|----------:|
| TestShift | 51.54 ns | 0.456 ns | 0.426 ns |      61 B |         - |
| TestMask  | 42.26 ns | 0.344 ns | 0.322 ns |      57 B |         - |