Fix size used for vectorization check in BitArray (#111558)

[AnsBalin]

This PR resolves #111558 by correcting the vector size used in the vectorization check within BitArray.

[EgorBo]

I think while you're on it, we can also replace various Sse2.And with just plain operators, e.g.

Vector128<byte> extractedLower = Sse2.And(shuffledLower, bitMask128);

to

Vector128<byte> extractedLower = shuffledLower & bitMask128;

but that's optional

[stephentoub]

Thanks

[jkotas]

Run a quick perf test that shows this is a perf improvement for all affected sizes?

[MihaZupan]

FWIW this would only affect pre-AVX2 X86 CPUs (or NAOT as I was informed we don't do Avx2 checks :/)

else if (Avx2.IsSupported && (uint)m_length >= Vector256<byte>.Count)
...
else if (Ssse3.IsSupported && ((uint)m_length >= Vector128<byte>.Count * 2u))
...
else if (AdvSimd.Arm64.IsSupported)

[AnsBalin]

Seems like the test failures are all unrelated to this PR. Is there anything else I should do? @stephentoub

[eiriktsarpalis]

Let's merge and retrigger to see what happens.

[AnsBalin]

Let's merge and retrigger to see what happens.

@eiriktsarpalis I still do not think the remaining failures are related as they seem to be timeouts. Are they expected to pass? Let me know what else I can do! Thanks

[jkotas]

@EgorBot -intel --envvars DOTNET_EnableAVX=0

using System.Collections;
using BenchmarkDotNet.Attributes;

public class Benchmarks
{
    bool[] buffer = new bool[1000];

    [Benchmark]
    [ArgumentsSource(nameof(TestData))]
    public void BitArrayCopyTo(BitArray bitArray) => bitArray.CopyTo(buffer, 0);

    public IEnumerable<BitArray> TestData()
    {
        yield return new BitArray(31);
        yield return new BitArray(32);
        yield return new BitArray(33);
        yield return new BitArray(64);
        yield return new BitArray(127);
        yield return new BitArray(128);
        yield return new BitArray(129);
        yield return new BitArray(256);
    }
}

[EgorBo]

@jkotas --envvars expects VAR:VALUE format (https://github.com/dotnet/BenchmarkDotNet/blob/master/docs/articles/guides/console-args.md#more) 🙂

[EgorBo]

@EgorBot -intel --envvars DOTNET_EnableAVX:0

using System.Collections;
using BenchmarkDotNet.Attributes;

public class Benchmarks
{
    bool[] buffer = new bool[1000];

    [Benchmark]
    [ArgumentsSource(nameof(TestData))]
    public void BitArrayCopyTo(BitArray bitArray) => bitArray.CopyTo(buffer, 0);

    public IEnumerable<BitArray> TestData()
    {
        yield return new BitArray(31);
        yield return new BitArray(32);
        yield return new BitArray(33);
        yield return new BitArray(64);
        yield return new BitArray(127);
        yield return new BitArray(128);
        yield return new BitArray(129);
        yield return new BitArray(256);
    }
}

[jkotas]

Run a quick perf test that shows this is a perf improvement for all affected sizes?

This change improves the perf 3x-10x for the affected sizes (on machines without AVX): EgorBot/runtime-utils#296 (comment)

[jkotas]

I still do not think the remaining failures are related

Right, the failures are unrelated.