Improve char.IsWhiteSpace for non-ASCII

[MihaZupan]

char.IsWhiteSpace looks something like

bool IsWhiteSpace(char c)
{
    if (c < 256) return Latin1LookupTable[c]:

    return Fallback(c);
}

static bool Fallback(char ch)
{
    nuint offset = GetCategoryCasingTableOffsetNoBoundsChecks(ch);
    return (sbyte)Unsafe.AddByteOffset(ref MemoryMarshal.GetReference(CategoriesValues), offset) < 0;
}

If you just run a microbenchmark over this with non-ASCII input, things will be fine as PGO will inline the whole thing together.

But if IsWhiteSpace ends up tiering up without seeing data super skewed for non-ASCII, the GetCategoryCasingTableOffsetNoBoundsChecks call in the fallback case won't be inlined, slowing down any future queries for non-ASCII chars.

public class NonAsciiTest
{
    private readonly string _text = string.Create(50, 0, (buffer, _) => new Random(42).NextBytes(MemoryMarshal.Cast<char, byte>(buffer)));

    public NonAsciiTest()
    {
        for (int i = 0; i < 20; i++)
        {
            for (int j = 0; j < 150; j++)
            {
                _ = char.IsWhiteSpace((char)j);
            }

            Thread.Sleep(16);
        }
    }

    [Benchmark]
    public int CountWhiteSpace()
    {
        int count = 0;
        foreach (char c in _text)
        {
            if (char.IsWhiteSpace(c))
            {
                count++;
            }
        }
        return count;
    }
}

Method	Toolchain	Mean	Error	Ratio
CountWhiteSpace	main	152.94 ns	3.041 ns	1.00
CountWhiteSpace	pr	95.31 ns	0.439 ns	0.62

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-globalization
See info in area-owners.md if you want to be subscribed.