[API Proposal]: Add BitwiseAndNot To TensorPrimitives

[LEI-Hongfaan]

Background and motivation

Certain architectures have specialized instructions to compute the AND NOT operation efficiently. By using these instructions, we can perform bitwise operations faster.

API Proposal

 namespace System.Numerics.Tensors {
   public static class TensorPrimitives {
+    public static void BitwiseAndNot<T>(ReadOnlySpan<T> x, ReadOnlySpan<T> y, Span<T> destination) where T : System.Numerics.IBitwiseOperators<T, T, T>;
+    public static void BitwiseAndNot<T>(ReadOnlySpan<T> x, T y, Span<T> destination) where T : System.Numerics.IBitwiseOperators<T, T, T>;
+    public static void BitwiseAndNot<T>(T x, ReadOnlySpan<T> y, Span<T> destination) where T : System.Numerics.IBitwiseOperators<T, T, T>;
   }
 }

 namespace System.Numerics {
   public interface IBitwiseOperators<TSelf,TOther,TResult> where TSelf : IBitwiseOperators<TSelf, TOther, TResult> {
+    public static abstract TResult BitwiseAndNot(TSelf left, TOther right);
   }
 }

API Usage

For example, when performing a bitwise AND operation on BigInteger values, if x is nonnegative and y is negative, then x & y = x & ~(abs(y) - 1). In code, this would look like:

TensorPrimitives.BitwiseAndNot(xRemainingUpperBits, yRemainingUpperBits, resultRemainingUpperBits);

Alternative Designs

1. Use AndNot instead of BitwiseAndNot
2. 1. Put the base case BitwiseAndNot in another interface.
   2. Do not add additional constraints on T.

Risks

No response

[dotnet-policy-service[bot]]

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

[huoyaoyuan]

public static abstract TResult BitwiseAndNot(TSelf left, TOther right);

Adding abstract method is a breaking change and unacceptable. Instead, this should be a static virtual method with fallback implementation.