[API Proposal]: Vector128/256 AddSaturate/RemoveSaturate

[EgorBo]

Background and motivation

I hit lack of these crossplatform helpers when I was doing "parse hex" work in #82521 - ended up with local functions. These APIs seem to present in SSE/AVX and AdvSimd and can be easily exposed as cross-platform helpers. Some places in BCL might benefit from this too, e.g.

runtime/src/libraries/System.Private.CoreLib/src/System/Text/Unicode/Utf16Utility.Validation.cs

Lines 133 to 142 in 220437e

	if (AdvSimd.IsSupported)
	{
	charIsThreeByteUtf8Encoded = AdvSimd.AddSaturate(utf16Data, vector7800);
	mask = GetNonAsciiBytes(AdvSimd.Or(charIsNonAscii, charIsThreeByteUtf8Encoded).AsByte(), bitMask128);
	}
	else
	{
	charIsThreeByteUtf8Encoded = Sse2.AddSaturate(utf16Data, vector7800);
	mask = (uint)Sse2.MoveMask(Sse2.Or(charIsNonAscii, charIsThreeByteUtf8Encoded).AsByte());
	}

API Proposal

namespace System.Numerics;

public static class Vector
{
+     public static Vector<T> AddSaturate<T>     (Vector<T> left, Vector<T> right) where T : struct;
+     public static Vector<T> SubtractSaturate<T>(Vector<T> left, Vector<T> right) where T : struct;
}

namespace System.Runtime.Intrinsics;

public static class Vector64
{
+     public static Vector64<T> AddSaturate<T>     (Vector64<T> left, Vector64<T> right) where T : struct;
+     public static Vector64<T> SubtractSaturate<T>(Vector64<T> left, Vector64<T> right) where T : struct;
}

public static class Vector128
{
+     public static Vector128<T> AddSaturate<T>     (Vector128<T> left, Vector128<T> right) where T : struct;
+     public static Vector128<T> SubtractSaturate<T>(Vector128<T> left, Vector128<T> right) where T : struct;
}

public static class Vector256
{
+     public static Vector256<T> AddSaturate<T>     (Vector256<T> left, Vector256<T> right) where T : struct;
+     public static Vector256<T> SubtractSaturate<T>(Vector256<T> left, Vector256<T> right) where T : struct;
}

public static class Vector512
{
+     public static Vector512<T> AddSaturate<T>     (Vector512<T> left, Vector512<T> right) where T : struct;
+     public static Vector512<T> SubtractSaturate<T>(Vector512<T> left, Vector512<T> right) where T : struct;
}

NOTE: SSE and AVX only accelerate byte, sbyte, short and ushort. AdvSimd supports more types.

[EgorBo]

cc @tannergooding

[ghost]

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

Issue Details

---

Background and motivation

I hit lack of these crossplatform helpers when I was doing "parse hex" work in #82521 - ended up with local functions. These APIs seem to present in SSE/AVX and AdvSimd and can be easily exposed as cross-platform helpers. Some places in BCL might benefit from this too, e.g.

runtime/src/libraries/System.Private.CoreLib/src/System/Text/Unicode/Utf16Utility.Validation.cs

Lines 133 to 142 in 220437e

	if (AdvSimd.IsSupported)
	{
	charIsThreeByteUtf8Encoded = AdvSimd.AddSaturate(utf16Data, vector7800);
	mask = GetNonAsciiBytes(AdvSimd.Or(charIsNonAscii, charIsThreeByteUtf8Encoded).AsByte(), bitMask128);
	}
	else
	{
	charIsThreeByteUtf8Encoded = Sse2.AddSaturate(utf16Data, vector7800);
	mask = (uint)Sse2.MoveMask(Sse2.Or(charIsNonAscii, charIsThreeByteUtf8Encoded).AsByte());
	}

API Proposal

namespace System.Runtime.Intrinsics;

public static class Vector128
{
+   public static Vector128<sbyte> AddSaturate(Vector128<sbyte> left, Vector128<sbyte> right);
+   public static Vector128<byte> AddSaturate(Vector128<byte> left, Vector128<byte> right);
+   public static Vector128<short> AddSaturate(Vector128<short> left, Vector128<short> right);
+   public static Vector128<ushort> AddSaturate(Vector128<ushort> left, Vector128<ushort> right);

+   public static Vector128<sbyte> SubtractSaturate(Vector128<sbyte> left, Vector128<sbyte> right);
+   public static Vector128<byte> SubtractSaturate(Vector128<byte> left, Vector128<byte> right);
+   public static Vector128<short> SubtractSaturate(Vector128<short> left, Vector128<short> right);
+   public static Vector128<ushort> SubtractSaturate(Vector128<ushort> left, Vector128<ushort> right);
}

public static class Vector256
{
+   public static Vector256<sbyte> AddSaturate(Vector256<sbyte> left, Vector256<sbyte> right);
+   public static Vector256<byte> AddSaturate(Vector256<byte> left, Vector256<byte> right);
+   public static Vector256<short> AddSaturate(Vector256<short> left, Vector256<short> right);
+   public static Vector256<ushort> AddSaturate(Vector256<ushort> left, Vector256<ushort> right);

+   public static Vector256<sbyte> SubtractSaturate(Vector256<sbyte> left, Vector256<sbyte> right);
+   public static Vector256<byte> SubtractSaturate(Vector256<byte> left, Vector256<byte> right);
+   public static Vector256<short> SubtractSaturate(Vector256<short> left, Vector256<short> right);
+   public static Vector256<ushort> SubtractSaturate(Vector256<ushort> left, Vector256<ushort> right);
}

NOTE: AdvSimd supports more types, so these are just common denominators

Author:	EgorBo
Assignees:	-
Labels:	area-System.Runtime.Intrinsics, api-ready-for-review
Milestone:	-

[bartonjs]

Video

• The AddSaturate name is consistent with the intrinsics
• We squared the circle to also include Vector, Vector64, and Vector512

namespace System.Runtime.Intrinsics;

public static partial class Vector
{
      public static Vector<T>   AddSaturate(Vector<T>   left, Vector<T>   right);
      public static Vector<T>   SubtractSaturate(Vector<T>   left, Vector<T>   right);
}

public static partial class Vector64
{
      public static Vector64<T>   AddSaturate(Vector64<T>   left, Vector64<T>   right);
      public static Vector64<T>   SubtractSaturate(Vector64<T>   left, Vector64<T>   right);
}

public static partial class Vector128
{
      public static Vector128<T>   AddSaturate(Vector128<T>   left, Vector128<T>   right);
      public static Vector128<T>   SubtractSaturate(Vector128<T>   left, Vector128<T>   right);
}

public static partial class Vector256
{
      public static Vector256<T>   AddSaturate(Vector256<T>   left, Vector256<T>   right);
      public static Vector256<T>   SubtractSaturate(Vector256<T>   left, Vector256<T>   right);
}

public static partial class Vector512
{
      public static Vector512<T>   AddSaturate(Vector512<T>   left, Vector512<T>   right);
      public static Vector512<T>   SubtractSaturate(Vector512<T>   left, Vector512<T>   right);
}

[tannergooding]

Not going to land this for .NET 8; we can take a PR anytime after the repo opens for .NET 9 changes