[Safe code] Clean up Guid.TryFormatX

[EgorBo]

Contributes to #94941

This PR removes unsafe code from Guid.TryFormatX + small clean up in TryParseExact.

New version is 10-15% faster.

[EgorBo]

@EgorBot -amd -arm -profiler

using System.Runtime.CompilerServices;
using BenchmarkDotNet.Attributes;

public class Benchmark
{
    static Guid guid = Guid.Parse("{D420C46B-04DC-4ABD-BF32-5132E82EE7F7}");

    [Benchmark]
    public bool FormatX()
    {
        Span<char> buffer = stackalloc char[100];
        bool result = guid.TryFormat(buffer, out int charsWritten, "X");
        Consume(buffer.Slice(0, charsWritten));
        return result;
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    static void Consume(Span<char> _){}
}

[dotnet-policy-service]

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

[MihaZupan]

New version is 10-15% faster.

I'm guessing that's from HexsToCharsHexOutput not being inlined before?

[EgorBo]

I'm guessing that's from HexsToCharsHexOutput not being inlined before?

Yep, but the goal was to remove the unsafe code & maintain same perf as before 🙂

[tannergooding]

Is the cast to char actually necessary here? It should work without it and that would make smaller IL and avoid an extra zero extension that the JIT has to otherwise remove here

[EgorBo]

@tannergooding hm. I've removed the cast but the codegen hasn't changed. But I guess it is still makes sense to remove it as useless

[tannergooding]

I'd guess the JIT is recognizing it as unnecessary and removing it already; just something small that caught my eye as likely unneeded

[EgorBo]

/ba-g "browser-wasm linux Release LibraryTests is stuck"