Class WorkerFactory

Factory to create worker that executes specified model on a particular device (GPU, CPU, etc) using particular backend. See IWorker for usage of the worker itself.

Inheritance

Object

WorkerFactory

BarracudaWorkerFactory

Namespace: Barracuda

Syntax

public class WorkerFactory

Methods

CreateComputeWorker(Model, Boolean)

Create a worker using the precompiled GPU backend for the given model. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateComputeWorker(Model model, bool verbose = false)

Parameters

Type	Name	Description
Model	model
Boolean	verbose

Returns

Type	Description
IWorker

CreateReferenceComputeWorker(Model, Boolean)

Create a worker using the reference GPU backend for the given model. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateReferenceComputeWorker(Model model, bool verbose = false)

Parameters

Type	Name	Description
Model	model
Boolean	verbose

Returns

Type	Description
IWorker

CreateReferenceCPUWorker(Model, Boolean)

Create a worker using the reference CPU backend for the given model. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateReferenceCPUWorker(Model model, bool verbose = false)

Parameters

Type	Name	Description
Model	model
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(Model, WorkerFactory.Device, Boolean)

Create a worker that will execute model using the best backend that is available for a given device type. model is the associated model. See ModelLoader.cs. device is the preferred device for execution. For example WorkerFactory.Device.GPU specifies the fast GPU path. verbose will log scheduling of layers execution to the console.

Declaration

public static IWorker CreateWorker(Model model, WorkerFactory.Device device, bool verbose = false)

Parameters

Type	Name	Description
Model	model
WorkerFactory.Device	device
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(Model, Boolean)

Create a worker that will execute model using the best backend that is available for a given device type. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console.

Declaration

public static IWorker CreateWorker(Model model, bool verbose = false)

Parameters

Type	Name	Description
Model	model
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(Model, String[], WorkerFactory.Device, Boolean)

Create a worker that will execute model using the best backend that is available for a given device type. model is the associated model. See ModelLoader.cs. additionalOutputs are the additional outputs to track but not directly specified by the model. device is the device type to run worker on. For example WorkerFactory.Device.GPU specifies the fast GPU path. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateWorker(Model model, string[] additionalOutputs, WorkerFactory.Device device = WorkerFactory.Device.Auto, bool verbose = false)

Parameters

Type	Name	Description
Model	model
String[]	additionalOutputs
WorkerFactory.Device	device
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(Model, String[], String[], WorkerFactory.Device, Boolean)

Create a worker that will execute model using the best backend that is available for a given device type. model is the associated model. See ModelLoader.cs. additionalOutputs are the additional outputs to track but not directly specified by the model. trimOutputs are the outputs not discard even if they are specified by the model. device is the device type to run worker on. For example WorkerFactory.Device.GPU specifies the fast GPU path. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateWorker(Model model, string[] additionalOutputs, string[] trimOutputs, WorkerFactory.Device device = WorkerFactory.Device.Auto, bool verbose = false)

Parameters

Type	Name	Description
Model	model
String[]	additionalOutputs
String[]	trimOutputs
WorkerFactory.Device	device
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(WorkerFactory.Type, Model, Boolean)

Create a worker with explicitly specified backend type to execute the given model. type is backend type to use. For example WorkerFactory.Type.Compute specifies the fast GPU path. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console.

Declaration

public static IWorker CreateWorker(WorkerFactory.Type type, Model model, bool verbose)

Parameters

Type	Name	Description
WorkerFactory.Type	type
Model	model
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(WorkerFactory.Type, Model, Boolean, WorkerFactory.Type, Boolean)

Create a worker with explicitly specified backend type to execute the given model. type is backend type to use. For example WorkerFactory.Type.Compute specifies the fast GPU path. model is the associated model. See ModelLoader.cs. verbose will log scheduling of layers execution to the console. compareAgainstType if different than type model will be run on those two backend and the result of every layer will be compared, checking for divergence. Great for debugging, but very slow because of the sync needed. differenceAsError if compareAgainstType is used difference will be reported as error is this is true or warning otherwise.

Declaration

public static IWorker CreateWorker(WorkerFactory.Type type, Model model, bool verbose, WorkerFactory.Type compareAgainstType, bool differenceAsError = false)

Parameters

Type	Name	Description
WorkerFactory.Type	type
Model	model
Boolean	verbose
WorkerFactory.Type	compareAgainstType
Boolean	differenceAsError

Returns

Type	Description
IWorker

CreateWorker(WorkerFactory.Type, Model, String[], Boolean)

Create a worker with explicitly specified backend type to execute the given model. type is backend type to use. For example WorkerFactory.Type.Compute specifies the fast GPU path. model is the associated model. See ModelLoader.cs. additionalOutputs are the additional outputs to track but not directly specified by the model. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateWorker(WorkerFactory.Type type, Model model, string[] additionalOutputs, bool verbose = false)

Parameters

Type	Name	Description
WorkerFactory.Type	type
Model	model
String[]	additionalOutputs
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(WorkerFactory.Type, Model, String[], String[], Boolean)

Create a worker with explicitly specified backend type to execute the given model. type is backend type to use. For example WorkerFactory.Type.Compute specifies the fast GPU path. model is the associated model. See ModelLoader.cs. additionalOutputs are the additional outputs to track but not directly specified by the model. trimOutputs are the outputs not discard even if they are specified by the model. verbose will log scheduling of layers execution to the console (default == false)

Declaration

public static IWorker CreateWorker(WorkerFactory.Type type, Model model, string[] additionalOutputs = null, string[] trimOutputs = null, bool verbose = false)

Parameters

Type	Name	Description
WorkerFactory.Type	type
Model	model
String[]	additionalOutputs
String[]	trimOutputs
Boolean	verbose

Returns

Type	Description
IWorker

CreateWorker(WorkerFactory.Type, Model, String[], String[], Boolean, WorkerFactory.Type, Boolean)

Create a worker with explicitly specified backend type to execute the given model. type is backend type to use. For example WorkerFactory.Type.Compute specifies the fast GPU path. model is the associated model. See ModelLoader.cs. additionalOutputs are the additional outputs to track but not directly specified by the model. trimOutputs are the outputs not discard even if they are specified by the model. verbose will log scheduling of layers execution to the console. compareAgainstType if different than type model will be run on those two backend and the result of every layer will be compared, checking for divergence. Great for debugging, but very slow because of the sync needed. differenceAsError if compareAgainstType is used difference will be reported as error is this is true or warning otherwise.

Declaration

public static IWorker CreateWorker(WorkerFactory.Type type, Model model, string[] additionalOutputs, string[] trimOutputs, bool verbose, WorkerFactory.Type compareAgainstType, bool differenceAsError = false)

Parameters

Type	Name	Description
WorkerFactory.Type	type
Model	model
String[]	additionalOutputs
String[]	trimOutputs
Boolean	verbose
WorkerFactory.Type	compareAgainstType
Boolean	differenceAsError

Returns

Type	Description
IWorker

GetBestTypeForDevice(WorkerFactory.Device)

Returns the best backend type that can run on a device given the model.

Declaration

public static WorkerFactory.Type GetBestTypeForDevice(WorkerFactory.Device device)

Parameters

Type	Name	Description
WorkerFactory.Device	device

Returns

Type	Description
WorkerFactory.Type

IsType(WorkerFactory.Type, WorkerFactory.Device)

Check if a backend is of a given type. For example: IsType(Type.CSharpRef, Device.GPU) == true

Declaration

public static bool IsType(WorkerFactory.Type type, WorkerFactory.Device device)

Parameters

Type	Name	Description
WorkerFactory.Type	type
WorkerFactory.Device	device

Returns

Type	Description
Boolean

ValidateType(WorkerFactory.Type)

Validate if a backend of type is supported, otherwise return a fallback type.

Declaration

public static WorkerFactory.Type ValidateType(WorkerFactory.Type type)

Parameters

Type	Name	Description
WorkerFactory.Type	type

Returns

Type	Description
WorkerFactory.Type