[API Proposal]: Extend System.Numerics.Tensors.TensorPrimitives with primitive types other than float

[asmirnov82]

Background and motivation

With the implementation of #89639 a new non-generic static API for the main arithmetic and conversion operators was added into the System.Numerics.Tensors namespace. This API provides hardware accelerated math on Span<float> type of data. While float is very often used in AI, other primitive types are also widely used in many areas (for example for data wrangling). Microsoft.Data.Analysis is one of the packages that will benefit from the hardware accelerated math for other primitive types.

API Proposal

1. Extend basic element-wise Arithmetic from existing API to support double, sbyte , byte , short , ushort , int , uint , long and ulong types.

Methods affected:

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Abs(ReadOnlySpan<float> x, Span<float> destination);

    public static void Add(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Add(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Divide(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Divide(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static float Max(ReadOnlySpan<float> value);
    public static float Min(ReadOnlySpan<float> value);

    public static void Multiply(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Multiply(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Negate(ReadOnlySpan<float> values, Span<float> destination);

    public static void Subtract(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Subtract(ReadOnlySpan<float> x, float y, Span<float> destination);
}

2. Add reverse method for non commutative operations where one of the operands is scalar.

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Divide(float x, ReadOnlySpan<float> y, Span<float> destination);

     public static void Subtract(float x, ReadOnlySpan<float> y, Span<float> destination);
}

3. Add new operations for hardware accelerated comparison:

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Equals(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static bool EqualsAll(ReadOnlySpan<float> x, ReadOnlySpan<float> y);
    public static bool EqualsAny(ReadOnlySpan<float> x, ReadOnlySpan<float> y);

    public static void GreaterThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void GreaterThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static void LessThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void LessThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
}

API Usage

The same as for the existing API proposed in #89639

Alternative Designs

No response

Risks

No response

[ghost]

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

Issue Details

---

Background and motivation

With the implementation of #89639 a new non-generic static API for the main arithmetic and conversion operators was added into the System.Numerics.Tensors namespace. This API provides hardware accelerated math on Span<float> type of data. While float is very often used in AI, other primitive types are also widely used in many areas (for example for data wrangling). Microsoft.Data.Analysis is one of the packages that will benefit from the hardware accelerated math for other primitive types.

API Proposal

1. Extend basic element-wise Arithmetic from existing API to support double, sbyte , byte , short , ushort , int , uint , long and ulong types.

Methods affected:

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Abs(ReadOnlySpan<float> x, Span<float> destination);

    public static void Add(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Add(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Divide(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Divide(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static float Max(ReadOnlySpan<float> value);
    public static float Min(ReadOnlySpan<float> value);

    public static void Multiply(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Multiply(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Negate(ReadOnlySpan<float> values, Span<float> destination);

    public static void Subtract(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Subtract(ReadOnlySpan<float> x, float y, Span<float> destination);
}

2. Add reverse method for non commutative operation where one of the operand is scalar.

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Divide(float x, ReadOnlySpan<float> y, Span<float> destination);

     public static void Subtract(float x, ReadOnlySpan<float> y, Span<float> destination);
}

3. Add new operations for hardware accelerated comparison:

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Equals(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static bool EqualsAll(ReadOnlySpan<float> x, ReadOnlySpan<float> y);
    public static bool EqualsAny(ReadOnlySpan<float> x, ReadOnlySpan<float> y);

    public static void GreaterThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void GreaterThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static void LessThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void LessThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
}

API Usage

The same as for the existing API proposed in #89639

Alternative Designs

No response

Risks

No response

Author:	asmirnov82
Assignees:	-
Labels:	api-suggestion, area-System.Numerics
Milestone:	-

[ghost]

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

Issue Details

---

Background and motivation

With the implementation of #89639 a new non-generic static API for the main arithmetic and conversion operators was added into the System.Numerics.Tensors namespace. This API provides hardware accelerated math on Span<float> type of data. While float is very often used in AI, other primitive types are also widely used in many areas (for example for data wrangling). Microsoft.Data.Analysis is one of the packages that will benefit from the hardware accelerated math for other primitive types.

API Proposal

1. Extend basic element-wise Arithmetic from existing API to support double, sbyte , byte , short , ushort , int , uint , long and ulong types.

Methods affected:

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Abs(ReadOnlySpan<float> x, Span<float> destination);

    public static void Add(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Add(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Divide(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Divide(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static float Max(ReadOnlySpan<float> value);
    public static float Min(ReadOnlySpan<float> value);

    public static void Multiply(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Multiply(ReadOnlySpan<float> x, float y, Span<float> destination);

    public static void Negate(ReadOnlySpan<float> values, Span<float> destination);

    public static void Subtract(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<float> destination);
    public static void Subtract(ReadOnlySpan<float> x, float y, Span<float> destination);
}

2. Add reverse method for non commutative operation where one of the operand is scalar.

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Divide(float x, ReadOnlySpan<float> y, Span<float> destination);

     public static void Subtract(float x, ReadOnlySpan<float> y, Span<float> destination);
}

3. Add new operations for hardware accelerated comparison:

New methods (for all primitive types):

namespace System.Numerics.Tensors;

public static partial class TensorPrimitives
{
    public static void Equals(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static bool EqualsAll(ReadOnlySpan<float> x, ReadOnlySpan<float> y);
    public static bool EqualsAny(ReadOnlySpan<float> x, ReadOnlySpan<float> y);

    public static void GreaterThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void GreaterThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);

    public static void LessThan(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
    public static void LessThanOrEqual(ReadOnlySpan<float> x, ReadOnlySpan<float> y, Span<bool> destination);
}

API Usage

The same as for the existing API proposed in #89639

Alternative Designs

No response

Risks

No response

Author:	asmirnov82
Assignees:	-
Labels:	api-suggestion, area-System.Numerics.Tensors, untriaged
Milestone:	-

[stephentoub]

See #93286.

[stephentoub]

(1) is handled via the added generic overloads.

(2) is implemented via #96451.

@tannergooding, do we want (3)? (generic rather than float-specific)