Class Tensor

Represents data in a multidimensional array-like structure.

Ownership and lifetime:

• Disposed needs to be called on the main thread.
• Ownership is always to the owner of the object.

Data Representation:

• TensorShape represents the data layout of the tensor
• Data is held by a tensorData (ITensorData) which can be on a given backend
• Data is stored in a flattened row major format
• Data can be pending (ie computation is being done in parallel)
  • call CompleteAllPendingOperations for a blocking call to finish computing the tensor's data
  • call CompleteOperationsAndDownloadAsync/ReadbackRequest for a non blocking call to finish computing the tensor's data Data can be in a non readable type (GPU/NPU)
  • Call CompleteOperationsAndDownload to make the tensor readable (this will fetch the data on the backend and convert it to a readable format)
  • CompleteOperationsAndDownload is a blocking call if called on its own. for a non blocking call, make sure that CompleteOperationsAndDownloadAsync/ReadbackRequest have been called previously you can check that with IsReadbackRequestDone

Data manipulation

• ToReadOnlyArray returns a copy of the tensor's data
• dataOnBackend can be manipulated directly to avoid a unnecessary copy see ComputeTensorData/BurstTensorData for info

Inheritance

object

Tensor

TensorByte

TensorFloat

TensorInt

TensorShort

Implements

IDisposable

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

Namespace: Unity.Sentis

Assembly: Unity.Sentis.dll

Syntax

public abstract class Tensor : IDisposable

Properties

backendType

The backend type where the tensor data is currently stored.

Declaration

public BackendType backendType { get; }

Property Value

Type	Description
BackendType

count

The length of the tensor.

Declaration

public abstract int count { get; }

Property Value

Type	Description
int

dataOnBackend

The device-specific internal representation of the tensor data.

Declaration

public ITensorData dataOnBackend { get; }

Property Value

Type	Description
ITensorData

dataType

The data type of the elements of the tensor.

Declaration

public abstract DataType dataType { get; }

Property Value

Type	Description
DataType

shape

The shape of the tensor, as a TensorShape.

Declaration

public TensorShape shape { get; }

Property Value

Type	Description
TensorShape

Methods

AttachToDevice(ITensorData)

Associates a tensor with the block of data on a device. Sentis downloads from source on first access.

Make sure source contains initialized and valid data that represents tensor values.

Declaration

public void AttachToDevice(ITensorData source)

Parameters

Type	Name	Description
ITensorData	source	The data on device to associate to the tensor.

CompleteAllPendingOperations()

Completes all scheduled tensor operations on device.

Declaration

public void CompleteAllPendingOperations()

CompleteOperationsAndDownload()

Blocking call to make tensor data read/write.

Issues a blocking download of the internal data. And converts tensorData to BurstTensorData

Declaration

public void CompleteOperationsAndDownload()

CompleteOperationsAndDownloadAsync()

Non blocking call to make tensor data read/write.

Issues a non blocking download of the internal data. And converts tensorData to BurstTensorData

Declaration

public Task<bool> CompleteOperationsAndDownloadAsync()

Returns

Type	Description
Task<bool>	The async task.

DetachFromDevice(bool)

Synchronizes the tensor data with the data on the device, then remove the tensor from the device.

Declaration

public ITensorData DetachFromDevice(bool disposeDeviceData = true)

Parameters

Type	Name	Description
bool	disposeDeviceData	Whether to free the space on device after detaching.

Returns

Type	Description
ITensorData	The detached tensor data.

Dispose()

Disposes of the tensor and any associated memory.

Declaration

public void Dispose()

~Tensor()

Dispose of the tensor and any associated memory.

Declaration

protected ~Tensor()

IsReadbackRequestDone()

Checks if asynchronous readback request it done.

Returns true if async readback is successful.

Declaration

public bool IsReadbackRequestDone()

Returns

Type	Description
bool	Whether the async readback request is successful.

ReadbackRequest(Action<bool>)

Schedules asynchronous download of the internal data.

Declaration

public void ReadbackRequest(Action<bool> callback = null)

Parameters

Type	Name	Description
Action<bool>	callback	Callback invoked when async readback is finished. Return value indicates if async readback is successful.

ReadbackRequestAsync()

Schedules asynchronous download task of the internal data.

Boolean indicates if async readback is successful.

Declaration

public Task<bool> ReadbackRequestAsync()

Returns

Type	Description
Task<bool>	awaitable task for when the readback request is successful.

Reshape(TensorShape)

Change the shape of a tensor without changing the backing data.

The new shape must be the same length as the current shape, and the data cannot be on the GPUPixel backend.

Declaration

public void Reshape(TensorShape shape)

Parameters

Type	Name	Description
TensorShape	shape	The new shape for the tensor.

ToString()

Returns a string that represents the Tensor.

Declaration

public override string ToString()

Returns

Type	Description
string	String representation of tensor.

Overrides

object.ToString()

UploadToDevice(ITensorData)

Uploads the tensor data to the destination data location on device.

Declaration

public abstract void UploadToDevice(ITensorData destination)

Parameters

Type	Name	Description
ITensorData	destination	The data on device to upload the tensor data to.

Implements

IDisposable

Extension Methods

TensorExtensions.PrintDataPart(Tensor, int, string)