Edit

Share via


TensorSpan<T> Struct

Definition

Represents a contiguous region of arbitrary memory. Unlike arrays, it can point to either managed or native memory, or to memory allocated on the stack. It is type-safe and memory-safe.

generic <typename T>
public value class TensorSpan
public readonly ref struct TensorSpan<T>
type TensorSpan<'T> = struct
Public Structure TensorSpan(Of T)

Type Parameters

T
Inheritance
TensorSpan<T>

Constructors

TensorSpan<T>(Array)

Creates a new TensorSpan<T> over the provided Array. The new TensorSpan<T> will have a rank of 1 and a length equal to the length of the provided Array.

TensorSpan<T>(Array, ReadOnlySpan<Int32>, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new TensorSpan<T> over the provided Array using the specified start offsets, lengths, and strides. If the strides are not provided, they will be automatically calculated.

TensorSpan<T>(Array, ReadOnlySpan<NIndex>, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new TensorSpan<T> over the provided Array using the specified start offsets, lengths, and strides. If the strides are not provided, they will be automatically calculated.

TensorSpan<T>(Span<T>)

Creates a new TensorSpan<T> over the provided Span<T>. The new TensorSpan<T> will have a rank of 1 and a length equal to the length of the provided Span<T>.

TensorSpan<T>(Span<T>, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new TensorSpan<T> over the provided Span<T> using the specified lengths and strides. If the strides are not provided, they will be automatically calculated.

TensorSpan<T>(T*, IntPtr)

Creates a new span over the target unmanaged buffer.

TensorSpan<T>(T*, IntPtr, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new span over the target unmanaged buffer. Clearly this is quite dangerous, because the length is not checked. But if this creation is correct, then all subsequent uses are correct.

TensorSpan<T>(T[])

Creates a new span over the entirety of the target array.

TensorSpan<T>(T[], Index, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new span over the portion of the target array beginning at 'start' index and ending at 'end' index (exclusive).

TensorSpan<T>(T[], Int32, ReadOnlySpan<IntPtr>, ReadOnlySpan<IntPtr>)

Creates a new span over the portion of the target array beginning at 'start' index and ending at 'end' index (exclusive).

Properties

Empty

Gets an empty TensorSpan<T>.

FlattenedLength

Gets the number of items in the span.

IsEmpty

Gets a value indicating whether this TensorSpan<T> is empty.

Item[ReadOnlySpan<IntPtr>]

Returns a reference to specified element of the TensorSpan.

Item[ReadOnlySpan<NIndex>]

Returns a reference to specified element of the TensorSpan.

Item[ReadOnlySpan<NRange>]

Returns a slice of the TensorSpan.

Lengths

Gets the length of each dimension in this TensorSpan<T>.

Rank

Gets the rank, or number of dimensions, of this TensorSpan<T>.

Strides

Gets the strides of this TensorSpan<T>.

Methods

Clear()

Clears the contents of this span.

CopyTo(TensorSpan<T>)

Copies the contents of this span into destination span. If the source and destinations overlap, this method behaves as if the original values in a temporary location before the destination is overwritten.

Equals(Object)
Obsolete.

This method is not supported as spans cannot be boxed. To compare two spans, use operator ==.

Fill(T)

Fills the contents of this span with the given value.

FlattenTo(Span<T>)

Flattens the contents of this span into the provided Span<T>.

GetEnumerator()

Gets an enumerator for this span.

GetHashCode()
Obsolete.

This method is not supported as spans cannot be boxed.

GetPinnableReference()

Returns a reference to the 0th element of the TensorSpan. If the TensorSpan is empty, returns null reference. It can be used for pinning and is required to support the use of span within a fixed statement.

Slice(ReadOnlySpan<NIndex>)

Returns a reference to specified element of the TensorSpan.

Slice(ReadOnlySpan<NRange>)

Forms a slice out of the given span

ToString()

For Span<T>, returns a new instance of string that represents the characters pointed to by the span. Otherwise, returns a String with the name of the type and the number of elements.

TryCopyTo(TensorSpan<T>)

Copies the contents of this span into destination span. If the source and destinations overlap, this method behaves as if the original values in a temporary location before the destination is overwritten.

TryFlattenTo(Span<T>)

Flattens the contents of this span into the provided Span<T>.

Operators

Equality(TensorSpan<T>, TensorSpan<T>)

Compares two System.Numerics.Tensors.TensorSpan<T> structs for equality.

Implicit(T[] to TensorSpan<T>)
Implicit(TensorSpan<T> to ReadOnlyTensorSpan<T>)

Converts a TensorSpan<T> to a ReadOnlyTensorSpan<T>.

Inequality(TensorSpan<T>, TensorSpan<T>)

Compares two System.Numerics.Tensors.TensorSpan<T> structs for inequality.

Extension Methods

BroadcastTo<T>(TensorSpan<T>, TensorSpan<T>)

Broadcasts the data from source to destination.

FilteredUpdate<T>(TensorSpan<T>, ReadOnlyTensorSpan<Boolean>, T)

Updates the tensor tensor with the value where the filter is true.

FilteredUpdate<T>(TensorSpan<T>, ReadOnlyTensorSpan<Boolean>, ReadOnlyTensorSpan<T>)

Updates the tensor tensor with the values where the filter is true. If dimensions are not the same an exception is thrown.

Reshape<T>(TensorSpan<T>, ReadOnlySpan<IntPtr>)

Reshapes the tensor tensor to the specified lengths. If one of the lengths is -1, it will be calculated automatically. Does not change the length of the underlying memory nor does it allocate new memory. If the new shape is not compatible with the old shape, an exception is thrown.

SetSlice<T>(TensorSpan<T>, ReadOnlyTensorSpan<T>, ReadOnlySpan<NRange>)

Sets a slice of the given tensor with the provided values for the given ranges.

Squeeze<T>(TensorSpan<T>, Int32)
ToString<T>(TensorSpan<T>, ReadOnlySpan<IntPtr>)

Creates a String representation of the TensorSpan<T>.

TryBroadcastTo<T>(TensorSpan<T>, TensorSpan<T>)

Broadcasts the data from tensor to the smallest broadcastable shape compatible with destination and stores it in destination.

Unsqueeze<T>(TensorSpan<T>, Int32)

Inserts a new dimension of length 1 that will appear at the dimension position.

Applies to