Simplify ManagedWebSocket.ApplyMask using TensorPrimitives.Xor

[Copilot]

Description

Replaces unsafe code and manual SIMD vectorization in ManagedWebSocket.ApplyMask with TensorPrimitives.Xor.

Changes

• Added System.Numerics.Tensors project reference to System.Net.WebSockets.csproj
• Removed unsafe keyword and all pointer operations from ApplyMask
• Replaced manual Vector<byte> loop with TensorPrimitives.Xor<int>(toMaskInts, rolledMask, toMaskInts)
• Simplified tail byte processing using foreach ref instead of pointer arithmetic

Before

private static unsafe int ApplyMask(Span<byte> toMask, int mask, int maskIndex)
{
    fixed (byte* toMaskBeg = &MemoryMarshal.GetReference(toMask))
    {
        // Manual Vector<byte> loop + pointer arithmetic (~45 lines)
    }
}

After

private static int ApplyMask(Span<byte> toMask, int mask, int maskIndex)
{
    Span<int> toMaskInts = MemoryMarshal.Cast<byte, int>(toMask);
    TensorPrimitives.Xor(toMaskInts, rolledMask, toMaskInts);
    // Tail bytes (~30 lines total)
}

Original prompt

ManagedWebSocket.cs has a ApplyMask(Span<byte> toMask, int mask, int maskIndex) method that uses unsafe code and manual vectorization. Try to simplify the implementation by calling the TensorPrimitives.Xor helper instead. Likely Xor<T>(ReadOnlySpan<T>, T, Span<T>)

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[dotnet-policy-service]

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

[jkotas]

System.Net.WebSockets is in the shared framework. This reference would require including System.Numerics.Tensors in shared framework as well. Is the extra baseline code bloat really worth it?

[MihaZupan]

From just the WebSockets perspective I don't think we'd care, we can live with the current setup. I was curious what copilot would do in this case.

But if there are more places where we're duplicating such logic it might be worth considering? Linq's Min/Average/Sum implementations come to mind where we have some SIMD sprinkled around (I haven't looked if they could actually be replaced).

cc: @tannergooding @stephentoub In case you've already put any thought into whether it could make sense to use Tensor helpers in other libraries.