Interface IWorker

Public interface for Workers. A worker is able to schedule models execution for a given backend. Use WorkerFactory to instantiate a worker.

Namespace: Barracuda
Syntax
public interface IWorker : IDisposable

Methods

Execute()

Non-blocking API that schedules network execution in one go Remark: This API will only be non-blocking for GPU inference.

Declaration
void Execute()

Execute(Tensor)

Non-blocking API that schedules network execution in one go, using the provider tensor as input. Remark: This API will only be non-blocking for GPU inference. Useful when network have only one input as input name is not needed.

Declaration
void Execute(Tensor input)
Parameters
Type Name Description
Tensor input

Execute(IDictionary<String, Tensor>)

Non-blocking API that schedules network execution in one go, using the provider tensor dictionary for inputs. Remark: This API will only be non-blocking for GPU inference.

Declaration
void Execute(IDictionary<string, Tensor> inputs)
Parameters
Type Name Description
IDictionary<String, Tensor> inputs

ExecuteAsync()

Non-blocking API that schedules network execution one layer at the time. Remark: This API will only be non-blocking for GPU inference. Check GetAsyncProgress() for progress.

Declaration
IEnumerator ExecuteAsync()
Returns
Type Description
IEnumerator

ExecuteAsync(Tensor)

Non-blocking API that schedules network execution one layer at the time, using the provider tensor as input. Remark: This API will only be non-blocking for GPU inference. Useful when network have only one input as input name is not needed. Check GetAsyncProgress() for progress.

Declaration
IEnumerator ExecuteAsync(Tensor input)
Parameters
Type Name Description
Tensor input
Returns
Type Description
IEnumerator

ExecuteAsync(IDictionary<String, Tensor>)

Non-blocking API that schedules network execution one layer at the time, using the provider tensor dictionary for inputs. Remark: This API will only be non-blocking for GPU inference. Check GetAsyncProgress() for progress.

Declaration
IEnumerator ExecuteAsync(IDictionary<string, Tensor> inputs)
Parameters
Type Name Description
IDictionary<String, Tensor> inputs
Returns
Type Description
IEnumerator

GetAsyncProgress()

Progress of the scheduling, 0.0 = 0%, 1.0 = 100%

Declaration
float GetAsyncProgress()
Returns
Type Description
Single

PeekOutput()

Returns a reference to tensor from the last layer of the network Useful when network has only one output. IMPORTANT: follow with TakeOwnership() call, if you want tensor to outlive worker or make tensor copy with DeepCopy() see also WorkerExtensions.FetchAndTakeOwnership()

Declaration
Tensor PeekOutput()
Returns
Type Description
Tensor

PeekOutput(String)

Returns a reference to tensor by name. IMPORTANT: follow with TakeOwnership() call, if you want tensor to outlive worker or make tensor copy with DeepCopy() see also WorkerExtensions.FetchAndTakeOwnership()

Declaration
Tensor PeekOutput(string name)
Parameters
Type Name Description
String name
Returns
Type Description
Tensor

PrepareForInput(IDictionary<String, TensorShape>)

Optional method to prepare network for particular input dimensions

Declaration
void PrepareForInput(IDictionary<string, TensorShape> inputShapes)
Parameters
Type Name Description
IDictionary<String, TensorShape> inputShapes

SetInput(Tensor)

Specify single tensor value as the input for the network. Useful when network has only one input and caller does not need to know input's name.

Declaration
void SetInput(Tensor x)
Parameters
Type Name Description
Tensor x

SetInput(String, Tensor)

Specify tensor value for the named input of the network.

Declaration
void SetInput(string name, Tensor x)
Parameters
Type Name Description
String name
Tensor x

Summary()

Returns a string summary after execution.

Declaration
string Summary()
Returns
Type Description
String

WaitForCompletion()

Wait for completion of part of the network that was scheduled via ExecuteAsync()

Declaration
void WaitForCompletion()

Extension Methods