View on GitHub
Install in Dash
SerranoResourceManager Class Reference
SerranoResourceManager
public class SerranoResourceManagerThis is a framework level class
-
Table tracking allocated Tensor and its corresponding MTLBuffer
Declaration
Swift
public var tensorBufferTable: [Tensor: MTLBuffer] -
The dictionary tracking the usage status of tensor
Declaration
Swift
public var tensorStatusTable: [Tensor: SerranoTensorStatus] -
The operation queue when operate on
tensorBufferTableandtensorStatusDeclaration
Swift
public var operationQueue: DispatchQueue -
Readable label
Declaration
Swift
public var label: String -
Description
Declaration
Swift
public var description: String -
Undocumented
Declaration
Swift
public static let globalManager = SerranoResourceManager(label: "gloabal_resource_manager")
-
Undocumented
Declaration
Swift
public init(label: String = "Resource Manager")
-
Allocate unmanaged tensors for shapes.
Unmanagedmeans that the alloated tensors will not be hold via strong refrence by manager.Declaration
Swift
public func allocateUnamangedTensors(_ shapes: [TensorShape]) -> [Tensor]Parameters
shapesshapes description
Return Value
return value description
-
Allocate unmanaged tensors for shapes.
Unmanagedmeans that the alloated tensors will not be hold via strong refrence by manager.Declaration
Swift
public func allocateUnamangedTensor(_ shape: TensorShape) -> TensorParameters
shapesshapes description
Return Value
return value description
-
Allocate unamanged
MTLBuffers.Warning
If engine has no available GPU device,
fatalErrorraised.Declaration
Swift
public func allocateUnmanagedMTLBuffers(_ tensors: [Tensor]) -> [MTLBuffer]Parameters
tensorstarget tensors
Return Value
Array of
MTLBuffer -
Allocate unamanged
MTLBuffer.Warning
If engine has no available GPU device,
fatalErrorraised.Declaration
Swift
public func allocateUnmanagedMTLBuffe(_ tensor: Tensor) -> MTLBufferParameters
tensortarget tensor
Return Value
MTLBuffer
-
Request managed tensors for target shapes.
Note
All tensors requested by this method should be return manually by calling
returnTensors(_ tensors: [Tensor])orreturnTensor(_ tensor: Tensor)Warning
This method is desgined for framework internal usage. Usually user should not call this method.
Declaration
Swift
public func allocateTensors(_ shapes: [TensorShape]) -> [Tensor]Parameters
shapestarget shapes
Return Value
tensors
-
Allocate single tensor object. This function actually call
allocateTensors(forShapes shapes: [TensorShape]).Note
All tensors requested by this method should be return manually by calling
returnTensors(_ tensors: [Tensor])orreturnTensor(_ tensor: Tensor)Warning
This method is desgined for framework internal usage. Usually user should not call this method.
Declaration
Swift
public func allocateTensor(_ tensorShape: TensorShape) -> TensorParameters
tensorShapeshape
Return Value
tensor
-
Return managed tensors.
Note
Tensor not managed by this manager or is a slice tensor will be ignored.
Declaration
Swift
public func returnTensors(_ tensors: [Tensor])Parameters
tensorsreturned tensors
-
Return single tensor to resource manager
Declaration
Swift
public func returnTensor(_ tensor: Tensor)Parameters
tensortensor
-
Release target tensors. Actually clear corresponding entries in
tensorStatusTableandtensorBufferTable.Note
Tensor not managed by this manager or is a slice tensor will be ignored.
Declaration
Swift
public func releaseTensors(_ tensors: [Tensor])Parameters
tensorstarget tensors
-
Check if a managed tensor is availabel for reuse If the passed in tensor is a sliced tensor, we check the status of its root tensor.
Note
falsewill be returned iftensoris not managed by this manager.Declaration
Swift
public func isTensorAvailable(_ tensor: Tensor) -> Bool
-
Request
MTLBufferResourcefor tensors.Note
If the passed in tensors are not mamaged by this resource manager, it will just call
allocateUnmanagedMTLBuffers(_ tensors: [Tensor]).Declaration
Swift
public func allocateMTLBufferResources(_ tensors: [Tensor]) -> [MTLBufferResource]Parameters
tensorstarget tensors
Return Value
Array of
MTLBufferResource -
Request
MTLBufferResourcefor a tensor.Note
If the passed in tensor is not mamaged by this resource manager, it will just call
allocateUnmanagedMTLBuffer(_ tensor: Tensor).Declaration
Swift
public func allocateMTLBufferResource(_ tensor: Tensor) -> MTLBufferResourceParameters
tensorstarget tensor
Return Value
-
Release all managed resources
Warning
This function does not guranteen that all managed tensors and buffers would be released in memory, since if tensor or buffer objects are still used in other places, the ARC will nor clear it. Before calling this function, the caller must be clear that all managed resources are in idle states.Declaration
Swift
public func releaseAllResources()
-
Check if tensor is managed by this manager. If a tensor is managed by this manager, all its sliced tensors are managed by this manager.
Note
If the tensor is a sliced tensor, we will check if managing its root tensor.
Declaration
Swift
public func isManagingTensor(_ tensor: Tensor) -> BoolParameters
tensortensor description
Return Value
return value description
-
Check if a MTLBuffer is managed by this manager
Declaration
Swift
public func isManagingBufferr(_ buffer: MTLBuffer) -> BoolParameters
bufferbuffer description
Return Value
return value description