QLstmLayer Class Reference

This layer represents a QLstm operation. More...

#include <QLstmLayer.hpp>

Inheritance diagram for QLstmLayer:

Collaboration diagram for QLstmLayer:

Public Member Functions
virtual std::unique_ptr< IWorkload >	CreateWorkload (const IWorkloadFactory &factory) const override
	Makes a workload for the QLstm type. More...
QLstmLayer *	Clone (Graph &graph) const override
	Creates a dynamically-allocated copy of this layer. More...
void	ValidateTensorShapesFromInputs () override
	Check if the input tensor shape(s) will lead to a valid configuration of QLstmLayer. More...
std::vector< TensorShape >	InferOutputShapes (const std::vector< TensorShape > &inputShapes) const override
	By default returns inputShapes if the number of inputs are equal to number of outputs, otherwise infers the output shapes from given input shapes and layer properties. More...
void	ExecuteStrategy (IStrategy &strategy) const override
	Apply a visitor to this layer. More...
Public Member Functions inherited from LayerWithParameters< QLstmDescriptor >
const QLstmDescriptor &	GetParameters () const override
	If the layer has a descriptor return it. More...
void	SerializeLayerParameters (ParameterStringifyFunction &fn) const override
	Helper to serialize the layer parameters to string (currently used in DotSerializer and company). More...
Public Member Functions inherited from Layer
	Layer (unsigned int numInputSlots, unsigned int numOutputSlots, LayerType type, const char *name)
	Layer (unsigned int numInputSlots, unsigned int numOutputSlots, LayerType type, DataLayout layout, const char *name)
const std::string &	GetNameStr () const
const OutputHandler &	GetOutputHandler (unsigned int i=0) const
OutputHandler &	GetOutputHandler (unsigned int i=0)
ShapeInferenceMethod	GetShapeInferenceMethod () const
bool	GetAllowExpandedDims () const
const std::vector< InputSlot > &	GetInputSlots () const
const std::vector< OutputSlot > &	GetOutputSlots () const
std::vector< InputSlot >::iterator	BeginInputSlots ()
std::vector< InputSlot >::iterator	EndInputSlots ()
std::vector< OutputSlot >::iterator	BeginOutputSlots ()
std::vector< OutputSlot >::iterator	EndOutputSlots ()
bool	IsOutputUnconnected ()
void	ResetPriority () const
LayerPriority	GetPriority () const
LayerType	GetType () const override
	Returns the armnn::LayerType of this layer. More...
DataType	GetDataType () const
const BackendId &	GetBackendId () const
void	SetBackendId (const BackendId &id) override
	Set the backend of the IConnectableLayer. More...
virtual void	CreateTensorHandles (const TensorHandleFactoryRegistry &registry, const IWorkloadFactory &factory, const bool IsMemoryManaged=true)
void	VerifyLayerConnections (unsigned int expectedConnections, const CheckLocation &location) const
virtual void	ReleaseConstantData ()
template<typename Op >
void	OperateOnConstantTensors (Op op)
const char *	GetName () const override
	Returns the name of the layer. More...
unsigned int	GetNumInputSlots () const override
	Returns the number of connectable input slots. More...
unsigned int	GetNumOutputSlots () const override
	Returns the number of connectable output slots. More...
const InputSlot &	GetInputSlot (unsigned int index) const override
	Get a const input slot handle by slot index. More...
InputSlot &	GetInputSlot (unsigned int index) override
	Get the input slot handle by slot index. More...
const OutputSlot &	GetOutputSlot (unsigned int index=0) const override
	Get the const output slot handle by slot index. More...
OutputSlot &	GetOutputSlot (unsigned int index=0) override
	Get the output slot handle by slot index. More...
void	SetGuid (LayerGuid guid)
LayerGuid	GetGuid () const final
	Returns the unique id of the layer. More...
void	AddRelatedLayerName (const std::string layerName)
const std::list< std::string > &	GetRelatedLayerNames ()
virtual void	Reparent (Graph &dest, std::list< Layer * >::const_iterator iterator)=0
void	BackendSelectionHint (Optional< BackendId > backend) final
	Provide a hint for the optimizer as to which backend to prefer for this layer. More...
Optional< BackendId >	GetBackendHint () const
void	SetShapeInferenceMethod (ShapeInferenceMethod shapeInferenceMethod)
void	SetAllowExpandedDims (bool allowExpandedDims)
template<typename T >
std::shared_ptr< T >	GetAdditionalInformation () const
void	SetAdditionalInfoForObject (const AdditionalInfoObjectPtr &additionalInfo)
virtual const BaseDescriptor &	GetParameters () const override
	If the layer has a descriptor return it. More...

Public Attributes
QLstmBasicParameters	m_BasicParameters
QLstmOptCifgParameters	m_CifgParameters
QLstmOptProjectionParameters	m_ProjectionParameters
QLstmOptPeepholeParameters	m_PeepholeParameters
QLstmOptLayerNormParameters	m_LayerNormParameters

Protected Member Functions
	QLstmLayer (const QLstmDescriptor &param, const char *name)
	Constructor to create a QLstmLayer. More...
	~QLstmLayer ()=default
	Default destructor. More...
Layer::ImmutableConstantTensors	GetConstantTensorsByRef () const override
	Retrieve the handles to the constant values stored by the layer. More...
Protected Member Functions inherited from LayerWithParameters< QLstmDescriptor >
	LayerWithParameters (unsigned int numInputSlots, unsigned int numOutputSlots, LayerType type, const QLstmDescriptor &param, const char *name)
	~LayerWithParameters ()=default
WorkloadInfo	PrepInfoAndDesc (QueueDescriptor &descriptor) const
	Helper function to reduce duplication in *Layer::CreateWorkload. More...
void	ExecuteStrategy (IStrategy &strategy) const override
	Apply a visitor to this layer. More...
Layer::ImmutableConstantTensors	GetConnectedConstantAsInputTensors () const
Protected Member Functions inherited from Layer
virtual	~Layer ()=default
template<typename QueueDescriptor >
void	CollectQueueDescriptorInputs (QueueDescriptor &descriptor, WorkloadInfo &info) const
template<typename QueueDescriptor >
void	CollectQueueDescriptorOutputs (QueueDescriptor &descriptor, WorkloadInfo &info) const
void	ValidateAndCopyShape (const TensorShape &outputShape, const TensorShape &inferredShape, const ShapeInferenceMethod shapeInferenceMethod, const std::string &layerName, const unsigned int outputSlotIndex=0)
void	VerifyShapeInferenceType (const TensorShape &outputShape, ShapeInferenceMethod shapeInferenceMethod)
template<typename QueueDescriptor >
WorkloadInfo	PrepInfoAndDesc (QueueDescriptor &descriptor) const
	Helper function to reduce duplication in *Layer::CreateWorkload. More...
template<typename LayerType , typename ... Params>
LayerType *	CloneBase (Graph &graph, Params &&... params) const
virtual ConstantTensors	GetConstantTensorsByRef () override final
void	SetAdditionalInfo (QueueDescriptor &descriptor) const
Protected Member Functions inherited from IConnectableLayer
	~IConnectableLayer ()
	Objects are not deletable via the handle. More...

Additional Inherited Members
Public Types inherited from LayerWithParameters< QLstmDescriptor >
using	DescriptorType = QLstmDescriptor
Public Types inherited from IConnectableLayer
using	ConstantTensors = std::vector< std::reference_wrapper< std::shared_ptr< ConstTensorHandle > >>
using	ImmutableConstantTensors = std::vector< std::reference_wrapper< const std::shared_ptr< ConstTensorHandle > >>
Protected Attributes inherited from LayerWithParameters< QLstmDescriptor >
QLstmDescriptor	m_Param
	The parameters for the layer (not including tensor-valued weights etc.). More...
Protected Attributes inherited from Layer
AdditionalInfoObjectPtr	m_AdditionalInfoObject
std::vector< OutputHandler >	m_OutputHandlers
ShapeInferenceMethod	m_ShapeInferenceMethod

Detailed Description

This layer represents a QLstm operation.

Definition at line 79 of file QLstmLayer.hpp.

Constructor & Destructor Documentation

QLstmLayer()

QLstmLayer ( const QLstmDescriptor &  param, const char *  name  )

protected

Constructor to create a QLstmLayer.

Parameters

[in]

name

Optional name for the layer.

Definition at line 17 of file QLstmLayer.cpp.

        : LayerWithParameters(3, 3, LayerType::QLstm, param, name)
{
}

References armnn::QLstm.

~QLstmLayer()

~QLstmLayer ( )

protecteddefault

Default destructor.

Member Function Documentation

Clone()

QLstmLayer * Clone ( Graph &  graph ) const

overridevirtual

Creates a dynamically-allocated copy of this layer.

Parameters

[in]

graph

The graph into which this layer is being cloned.

Implements Layer.

Definition at line 81 of file QLstmLayer.cpp.

{
    auto layer = CloneBase<QLstmLayer>(graph, m_Param, GetName());
 
    layer->m_BasicParameters.m_InputToForgetWeights = m_BasicParameters.m_InputToForgetWeights ?
            m_BasicParameters.m_InputToForgetWeights : nullptr;
    layer->m_BasicParameters.m_InputToCellWeights = m_BasicParameters.m_InputToCellWeights ?
            m_BasicParameters.m_InputToCellWeights : nullptr;
    layer->m_BasicParameters.m_InputToOutputWeights = m_BasicParameters.m_InputToOutputWeights ?
            m_BasicParameters.m_InputToOutputWeights : nullptr;
    layer->m_BasicParameters.m_RecurrentToForgetWeights = m_BasicParameters.m_RecurrentToForgetWeights ?
            m_BasicParameters.m_RecurrentToForgetWeights : nullptr;
    layer->m_BasicParameters.m_RecurrentToCellWeights = m_BasicParameters.m_RecurrentToCellWeights ?
            m_BasicParameters.m_RecurrentToCellWeights : nullptr;
    layer->m_BasicParameters.m_RecurrentToOutputWeights = m_BasicParameters.m_RecurrentToOutputWeights ?
            m_BasicParameters.m_RecurrentToOutputWeights : nullptr;
    layer->m_BasicParameters.m_ForgetGateBias = m_BasicParameters.m_ForgetGateBias ?
            m_BasicParameters.m_ForgetGateBias : nullptr;
    layer->m_BasicParameters.m_CellBias = m_BasicParameters.m_CellBias ?
            m_BasicParameters.m_CellBias : nullptr;
    layer->m_BasicParameters.m_OutputGateBias = m_BasicParameters.m_OutputGateBias ?
            m_BasicParameters.m_OutputGateBias : nullptr;
 
    if (!m_Param.m_CifgEnabled)
    {
        layer->m_CifgParameters.m_InputToInputWeights = m_CifgParameters.m_InputToInputWeights ?
                m_CifgParameters.m_InputToInputWeights : nullptr;
        layer->m_CifgParameters.m_RecurrentToInputWeights = m_CifgParameters.m_RecurrentToInputWeights ?
                m_CifgParameters.m_RecurrentToInputWeights : nullptr;
        layer->m_CifgParameters.m_InputGateBias = m_CifgParameters.m_InputGateBias ?
                m_CifgParameters.m_InputGateBias : nullptr;
    }
 
    if (m_Param.m_ProjectionEnabled)
    {
        layer->m_ProjectionParameters.m_ProjectionWeights = m_ProjectionParameters.m_ProjectionWeights ?
                m_ProjectionParameters.m_ProjectionWeights : nullptr;
        layer->m_ProjectionParameters.m_ProjectionBias = m_ProjectionParameters.m_ProjectionBias ?
                m_ProjectionParameters.m_ProjectionBias : nullptr;
    }
 
    if (m_Param.m_PeepholeEnabled)
    {
        if (!m_Param.m_CifgEnabled) {
            layer->m_PeepholeParameters.m_CellToInputWeights = m_PeepholeParameters.m_CellToInputWeights ?
                    m_PeepholeParameters.m_CellToInputWeights : nullptr;
        }
 
        layer->m_PeepholeParameters.m_CellToForgetWeights = m_PeepholeParameters.m_CellToForgetWeights ?
                m_PeepholeParameters.m_CellToForgetWeights : nullptr;
        layer->m_PeepholeParameters.m_CellToOutputWeights = m_PeepholeParameters.m_CellToOutputWeights ?
                m_PeepholeParameters.m_CellToOutputWeights : nullptr;
    }
 
    if (m_Param.m_LayerNormEnabled)
    {
        if (!m_Param.m_CifgEnabled) {
            layer->m_LayerNormParameters.m_InputLayerNormWeights = m_LayerNormParameters.m_InputLayerNormWeights ?
                    m_LayerNormParameters.m_InputLayerNormWeights : nullptr;
        }
 
        layer->m_LayerNormParameters.m_ForgetLayerNormWeights = m_LayerNormParameters.m_ForgetLayerNormWeights ?
                m_LayerNormParameters.m_ForgetLayerNormWeights : nullptr;
        layer->m_LayerNormParameters.m_CellLayerNormWeights = m_LayerNormParameters.m_CellLayerNormWeights ?
                m_LayerNormParameters.m_CellLayerNormWeights : nullptr;
        layer->m_LayerNormParameters.m_OutputLayerNormWeights = m_LayerNormParameters.m_OutputLayerNormWeights ?
                m_LayerNormParameters.m_OutputLayerNormWeights : nullptr;
    }
 
    return std::move(layer);
}

References Layer::GetName(), QLstmLayer::m_BasicParameters, QLstmBasicParameters::m_CellBias, QLstmOptLayerNormParameters::m_CellLayerNormWeights, QLstmOptPeepholeParameters::m_CellToForgetWeights, QLstmOptPeepholeParameters::m_CellToInputWeights, QLstmOptPeepholeParameters::m_CellToOutputWeights, QLstmDescriptor::m_CifgEnabled, QLstmLayer::m_CifgParameters, QLstmBasicParameters::m_ForgetGateBias, QLstmOptLayerNormParameters::m_ForgetLayerNormWeights, QLstmOptCifgParameters::m_InputGateBias, QLstmOptLayerNormParameters::m_InputLayerNormWeights, QLstmBasicParameters::m_InputToCellWeights, QLstmBasicParameters::m_InputToForgetWeights, QLstmOptCifgParameters::m_InputToInputWeights, QLstmBasicParameters::m_InputToOutputWeights, QLstmDescriptor::m_LayerNormEnabled, QLstmLayer::m_LayerNormParameters, QLstmBasicParameters::m_OutputGateBias, QLstmOptLayerNormParameters::m_OutputLayerNormWeights, LayerWithParameters< QLstmDescriptor >::m_Param, QLstmDescriptor::m_PeepholeEnabled, QLstmLayer::m_PeepholeParameters, QLstmOptProjectionParameters::m_ProjectionBias, QLstmDescriptor::m_ProjectionEnabled, QLstmLayer::m_ProjectionParameters, QLstmOptProjectionParameters::m_ProjectionWeights, QLstmBasicParameters::m_RecurrentToCellWeights, QLstmBasicParameters::m_RecurrentToForgetWeights, QLstmOptCifgParameters::m_RecurrentToInputWeights, and QLstmBasicParameters::m_RecurrentToOutputWeights.

CreateWorkload()

std::unique_ptr< IWorkload > CreateWorkload ( const IWorkloadFactory &  factory ) const

overridevirtual

Makes a workload for the QLstm type.

Parameters

[in]	graph	The graph where this layer can be found.
[in]	factory	The workload factory which will create the workload.

Returns

A pointer to the created workload, or nullptr if not created.

Implements Layer.

Definition at line 22 of file QLstmLayer.cpp.

{
    QLstmQueueDescriptor descriptor;
 
    // Basic parameters
    descriptor.m_InputToForgetWeights = m_BasicParameters.m_InputToForgetWeights.get();
    descriptor.m_InputToCellWeights = m_BasicParameters.m_InputToCellWeights.get();
    descriptor.m_InputToOutputWeights = m_BasicParameters.m_InputToOutputWeights.get();
    descriptor.m_RecurrentToForgetWeights = m_BasicParameters.m_RecurrentToForgetWeights.get();
    descriptor.m_RecurrentToCellWeights = m_BasicParameters.m_RecurrentToCellWeights.get();
    descriptor.m_RecurrentToOutputWeights = m_BasicParameters.m_RecurrentToOutputWeights.get();
    descriptor.m_ForgetGateBias = m_BasicParameters.m_ForgetGateBias.get();
    descriptor.m_CellBias = m_BasicParameters.m_CellBias.get();
    descriptor.m_OutputGateBias = m_BasicParameters.m_OutputGateBias.get();
 
    // CIFG parameters
    if (!m_Param.m_CifgEnabled)
    {
        descriptor.m_InputToInputWeights     = m_CifgParameters.m_InputToInputWeights.get();
        descriptor.m_RecurrentToInputWeights = m_CifgParameters.m_RecurrentToInputWeights.get();
        descriptor.m_InputGateBias           = m_CifgParameters.m_InputGateBias.get();
    }
 
    // Projection parameters
    if (m_Param.m_ProjectionEnabled)
    {
        descriptor.m_ProjectionWeights = m_ProjectionParameters.m_ProjectionWeights.get();
        descriptor.m_ProjectionBias    = m_ProjectionParameters.m_ProjectionBias.get();
    }
 
    // Peephole parameters
    if (m_Param.m_PeepholeEnabled)
    {
        if (!m_Param.m_CifgEnabled)
        {
            descriptor.m_CellToInputWeights = m_PeepholeParameters.m_CellToInputWeights.get();
        }
 
        descriptor.m_CellToForgetWeights = m_PeepholeParameters.m_CellToForgetWeights.get();
        descriptor.m_CellToOutputWeights = m_PeepholeParameters.m_CellToOutputWeights.get();
    }
 
    // Layer normalisation parameters
    if(m_Param.m_LayerNormEnabled)
    {
        if (!m_Param.m_CifgEnabled)
        {
            descriptor.m_InputLayerNormWeights = m_LayerNormParameters.m_InputLayerNormWeights.get();
        }
        descriptor.m_ForgetLayerNormWeights = m_LayerNormParameters.m_ForgetLayerNormWeights.get();
        descriptor.m_CellLayerNormWeights   = m_LayerNormParameters.m_CellLayerNormWeights.get();
        descriptor.m_OutputLayerNormWeights = m_LayerNormParameters.m_OutputLayerNormWeights.get();
    }
 
    SetAdditionalInfo(descriptor);
 
    return factory.CreateWorkload(LayerType::QLstm, descriptor, PrepInfoAndDesc(descriptor));
}

References IWorkloadFactory::CreateWorkload(), QLstmLayer::m_BasicParameters, QLstmBasicParameters::m_CellBias, QLstmQueueDescriptor::m_CellBias, QLstmOptLayerNormParameters::m_CellLayerNormWeights, QLstmQueueDescriptor::m_CellLayerNormWeights, QLstmOptPeepholeParameters::m_CellToForgetWeights, QLstmQueueDescriptor::m_CellToForgetWeights, QLstmOptPeepholeParameters::m_CellToInputWeights, QLstmQueueDescriptor::m_CellToInputWeights, QLstmOptPeepholeParameters::m_CellToOutputWeights, QLstmQueueDescriptor::m_CellToOutputWeights, QLstmDescriptor::m_CifgEnabled, QLstmLayer::m_CifgParameters, QLstmBasicParameters::m_ForgetGateBias, QLstmQueueDescriptor::m_ForgetGateBias, QLstmOptLayerNormParameters::m_ForgetLayerNormWeights, QLstmQueueDescriptor::m_ForgetLayerNormWeights, QLstmOptCifgParameters::m_InputGateBias, QLstmQueueDescriptor::m_InputGateBias, QLstmOptLayerNormParameters::m_InputLayerNormWeights, QLstmQueueDescriptor::m_InputLayerNormWeights, QLstmBasicParameters::m_InputToCellWeights, QLstmQueueDescriptor::m_InputToCellWeights, QLstmBasicParameters::m_InputToForgetWeights, QLstmQueueDescriptor::m_InputToForgetWeights, QLstmOptCifgParameters::m_InputToInputWeights, QLstmQueueDescriptor::m_InputToInputWeights, QLstmBasicParameters::m_InputToOutputWeights, QLstmQueueDescriptor::m_InputToOutputWeights, QLstmDescriptor::m_LayerNormEnabled, QLstmLayer::m_LayerNormParameters, QLstmBasicParameters::m_OutputGateBias, QLstmQueueDescriptor::m_OutputGateBias, QLstmOptLayerNormParameters::m_OutputLayerNormWeights, QLstmQueueDescriptor::m_OutputLayerNormWeights, LayerWithParameters< QLstmDescriptor >::m_Param, QLstmDescriptor::m_PeepholeEnabled, QLstmLayer::m_PeepholeParameters, QLstmOptProjectionParameters::m_ProjectionBias, QLstmQueueDescriptor::m_ProjectionBias, QLstmDescriptor::m_ProjectionEnabled, QLstmLayer::m_ProjectionParameters, QLstmOptProjectionParameters::m_ProjectionWeights, QLstmQueueDescriptor::m_ProjectionWeights, QLstmBasicParameters::m_RecurrentToCellWeights, QLstmQueueDescriptor::m_RecurrentToCellWeights, QLstmBasicParameters::m_RecurrentToForgetWeights, QLstmQueueDescriptor::m_RecurrentToForgetWeights, QLstmOptCifgParameters::m_RecurrentToInputWeights, QLstmQueueDescriptor::m_RecurrentToInputWeights, QLstmBasicParameters::m_RecurrentToOutputWeights, QLstmQueueDescriptor::m_RecurrentToOutputWeights, LayerWithParameters< QLstmDescriptor >::PrepInfoAndDesc(), armnn::QLstm, and Layer::SetAdditionalInfo().

ExecuteStrategy()

void ExecuteStrategy ( IStrategy &  strategy ) const

overridevirtual

Apply a visitor to this layer.

Reimplemented from Layer.

Definition at line 307 of file QLstmLayer.cpp.

{
    std::vector<ConstTensor> constTensors;
    ManagedConstTensorHandle managedInputToForgetWeights(m_BasicParameters.m_InputToForgetWeights);
    ManagedConstTensorHandle managedInputToCellWeights(m_BasicParameters.m_InputToCellWeights);
    ManagedConstTensorHandle managedInputToOutputWeights(m_BasicParameters.m_InputToOutputWeights);
    ManagedConstTensorHandle managedRecurrentToForgetWeights(m_BasicParameters.m_RecurrentToForgetWeights);
    ManagedConstTensorHandle managedRecurrentToCellWeights(m_BasicParameters.m_RecurrentToCellWeights);
    ManagedConstTensorHandle managedRecurrentToOutputWeights(m_BasicParameters.m_RecurrentToOutputWeights);
    ManagedConstTensorHandle managedForgetGateBias(m_BasicParameters.m_ForgetGateBias);
    ManagedConstTensorHandle managedCellBias(m_BasicParameters.m_CellBias);
    ManagedConstTensorHandle managedOutputGateBias(m_BasicParameters.m_OutputGateBias);
 
    // Cifg parameters
    ManagedConstTensorHandle managedInputToInputWeights(m_CifgParameters.m_InputToInputWeights);
    ManagedConstTensorHandle managedRecurrentToInputWeights(m_CifgParameters.m_RecurrentToInputWeights);
    ManagedConstTensorHandle managedInputGateBias(m_CifgParameters.m_InputGateBias);
 
    // Projection parameters
    ManagedConstTensorHandle managedProjectionWeights(m_ProjectionParameters.m_ProjectionWeights);
    ManagedConstTensorHandle managedProjectionBias(m_ProjectionParameters.m_ProjectionBias);
 
    // Peephole parameters
    ManagedConstTensorHandle managedCellToInputWeights(m_PeepholeParameters.m_CellToInputWeights);
    ManagedConstTensorHandle managedCellToForgetWeights(m_PeepholeParameters.m_CellToForgetWeights);
    ManagedConstTensorHandle managedCellToOutputWeights(m_PeepholeParameters.m_CellToOutputWeights);
 
    // Layer normalisation parameters
    ManagedConstTensorHandle managedInputLayerNormWeights(m_LayerNormParameters.m_InputLayerNormWeights);
    ManagedConstTensorHandle managedForgetLayerNormWeights(m_LayerNormParameters.m_ForgetLayerNormWeights);
    ManagedConstTensorHandle managedCellLayerNormWeights(m_LayerNormParameters.m_CellLayerNormWeights);
    ManagedConstTensorHandle managedOutputLayerNormWeights(m_LayerNormParameters.m_OutputLayerNormWeights);
 
    // First add mandatory/basic parameters
    if (m_BasicParameters.m_InputToForgetWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputToForgetWeights.GetTensorInfo(),
                                              managedInputToForgetWeights.Map()));
    }
    if (m_BasicParameters.m_InputToCellWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputToCellWeights.GetTensorInfo(),
                                              managedInputToCellWeights.Map()));
    }
    if (m_BasicParameters.m_InputToOutputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputToOutputWeights.GetTensorInfo(),
                                              managedInputToOutputWeights.Map()));
    }
    if (m_BasicParameters.m_RecurrentToForgetWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(
                managedRecurrentToForgetWeights.GetTensorInfo(),
                managedRecurrentToForgetWeights.Map()));
    }
    if (m_BasicParameters.m_RecurrentToCellWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(
                managedRecurrentToCellWeights.GetTensorInfo(),
                managedRecurrentToCellWeights.Map()));
    }
    if (m_BasicParameters.m_RecurrentToOutputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(
                managedRecurrentToOutputWeights.GetTensorInfo(),
                managedRecurrentToOutputWeights.Map()));
    }
    if (m_BasicParameters.m_ForgetGateBias != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedForgetGateBias.GetTensorInfo(),
                                              managedForgetGateBias.Map()));
    }
    if (m_BasicParameters.m_CellBias != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedCellBias.GetTensorInfo(),
                                              managedCellBias.Map()));
    }
    if (m_BasicParameters.m_OutputGateBias != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedOutputGateBias.GetTensorInfo(),
                                              managedOutputGateBias.Map()));
    }
 
    // Add cifig parameters
    if (m_CifgParameters.m_InputToInputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputToInputWeights.GetTensorInfo(),
                                              managedInputToInputWeights.Map()));
    }
    if (m_CifgParameters.m_RecurrentToInputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(
                managedRecurrentToInputWeights.GetTensorInfo(),
                managedRecurrentToInputWeights.Map()));
    }
    if (m_CifgParameters.m_InputGateBias != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputGateBias.GetTensorInfo(),
                                              managedInputGateBias.Map()));
    }
 
    // Add peephole parameters
    if (m_PeepholeParameters.m_CellToInputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedCellToInputWeights.GetTensorInfo(),
                                              managedCellToInputWeights.Map()));
    }
    if (m_PeepholeParameters.m_CellToForgetWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedCellToForgetWeights.GetTensorInfo(),
                                              managedCellToForgetWeights.Map()));
    }
    if (m_PeepholeParameters.m_CellToOutputWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedCellToOutputWeights.GetTensorInfo(),
                                              managedCellToOutputWeights.Map()));
    }
 
    // Add projection parameters
    if (m_ProjectionParameters.m_ProjectionWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedProjectionWeights.GetTensorInfo(),
                                              managedProjectionWeights.Map()));
    }
    if (m_ProjectionParameters.m_ProjectionBias != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedProjectionBias.GetTensorInfo(),
                                              managedProjectionBias.Map()));
    }
 
    // Add norm parameters
    if (m_LayerNormParameters.m_InputLayerNormWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedInputLayerNormWeights.GetTensorInfo(),
                                              managedInputLayerNormWeights.Map()));
    }
    if (m_LayerNormParameters.m_ForgetLayerNormWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedForgetLayerNormWeights.GetTensorInfo(),
                                              managedForgetLayerNormWeights.Map()));
    }
    if (m_LayerNormParameters.m_CellLayerNormWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedCellLayerNormWeights.GetTensorInfo(),
                                              managedCellLayerNormWeights.Map()));
    }
    if (m_LayerNormParameters.m_OutputLayerNormWeights != nullptr)
    {
        constTensors.emplace_back(ConstTensor(managedOutputLayerNormWeights.GetTensorInfo(),
                                              managedOutputLayerNormWeights.Map()));
    }
    strategy.ExecuteStrategy(this, GetParameters(), constTensors, GetName());
}

References IStrategy::ExecuteStrategy(), Layer::GetName(), LayerWithParameters< QLstmDescriptor >::GetParameters(), ManagedConstTensorHandle::GetTensorInfo(), QLstmLayer::m_BasicParameters, QLstmBasicParameters::m_CellBias, QLstmOptLayerNormParameters::m_CellLayerNormWeights, QLstmOptPeepholeParameters::m_CellToForgetWeights, QLstmOptPeepholeParameters::m_CellToInputWeights, QLstmOptPeepholeParameters::m_CellToOutputWeights, QLstmLayer::m_CifgParameters, QLstmBasicParameters::m_ForgetGateBias, QLstmOptLayerNormParameters::m_ForgetLayerNormWeights, QLstmOptCifgParameters::m_InputGateBias, QLstmOptLayerNormParameters::m_InputLayerNormWeights, QLstmBasicParameters::m_InputToCellWeights, QLstmBasicParameters::m_InputToForgetWeights, QLstmOptCifgParameters::m_InputToInputWeights, QLstmBasicParameters::m_InputToOutputWeights, QLstmLayer::m_LayerNormParameters, QLstmBasicParameters::m_OutputGateBias, QLstmOptLayerNormParameters::m_OutputLayerNormWeights, QLstmLayer::m_PeepholeParameters, QLstmOptProjectionParameters::m_ProjectionBias, QLstmLayer::m_ProjectionParameters, QLstmOptProjectionParameters::m_ProjectionWeights, QLstmBasicParameters::m_RecurrentToCellWeights, QLstmBasicParameters::m_RecurrentToForgetWeights, QLstmOptCifgParameters::m_RecurrentToInputWeights, QLstmBasicParameters::m_RecurrentToOutputWeights, and ManagedConstTensorHandle::Map().

GetConstantTensorsByRef()

Layer::ImmutableConstantTensors GetConstantTensorsByRef ( ) const

overrideprotectedvirtual

Retrieve the handles to the constant values stored by the layer.

Returns

A vector of the constant tensors stored by this layer.

Reimplemented from Layer.

Definition at line 272 of file QLstmLayer.cpp.

{
    // For API stability DO NOT ALTER order and add new members to the end of vector
    return {m_BasicParameters.m_InputToForgetWeights,
            m_BasicParameters.m_InputToCellWeights,
            m_BasicParameters.m_InputToOutputWeights,
            m_BasicParameters.m_RecurrentToForgetWeights,
            m_BasicParameters.m_RecurrentToCellWeights,
            m_BasicParameters.m_RecurrentToOutputWeights,
            m_BasicParameters.m_ForgetGateBias,
            m_BasicParameters.m_CellBias,
            m_BasicParameters.m_OutputGateBias,
 
            // Cifg parameters
            m_CifgParameters.m_InputToInputWeights,
            m_CifgParameters.m_RecurrentToInputWeights,
            m_CifgParameters.m_InputGateBias,
 
            // Projection parameters
            m_ProjectionParameters.m_ProjectionWeights,
            m_ProjectionParameters.m_ProjectionBias,
 
            // Peephole parameters
            m_PeepholeParameters.m_CellToInputWeights,
            m_PeepholeParameters.m_CellToForgetWeights,
            m_PeepholeParameters.m_CellToOutputWeights,
 
            // Layer normalisation parameters
            m_LayerNormParameters.m_InputLayerNormWeights,
            m_LayerNormParameters.m_ForgetLayerNormWeights,
            m_LayerNormParameters.m_CellLayerNormWeights,
            m_LayerNormParameters.m_OutputLayerNormWeights};
}

References QLstmLayer::m_BasicParameters, QLstmBasicParameters::m_CellBias, QLstmOptLayerNormParameters::m_CellLayerNormWeights, QLstmOptPeepholeParameters::m_CellToForgetWeights, QLstmOptPeepholeParameters::m_CellToInputWeights, QLstmOptPeepholeParameters::m_CellToOutputWeights, QLstmLayer::m_CifgParameters, QLstmBasicParameters::m_ForgetGateBias, QLstmOptLayerNormParameters::m_ForgetLayerNormWeights, QLstmOptCifgParameters::m_InputGateBias, QLstmOptLayerNormParameters::m_InputLayerNormWeights, QLstmBasicParameters::m_InputToCellWeights, QLstmBasicParameters::m_InputToForgetWeights, QLstmOptCifgParameters::m_InputToInputWeights, QLstmBasicParameters::m_InputToOutputWeights, QLstmLayer::m_LayerNormParameters, QLstmBasicParameters::m_OutputGateBias, QLstmOptLayerNormParameters::m_OutputLayerNormWeights, QLstmLayer::m_PeepholeParameters, QLstmOptProjectionParameters::m_ProjectionBias, QLstmLayer::m_ProjectionParameters, QLstmOptProjectionParameters::m_ProjectionWeights, QLstmBasicParameters::m_RecurrentToCellWeights, QLstmBasicParameters::m_RecurrentToForgetWeights, QLstmOptCifgParameters::m_RecurrentToInputWeights, and QLstmBasicParameters::m_RecurrentToOutputWeights.

InferOutputShapes()

std::vector< TensorShape > InferOutputShapes ( const std::vector< TensorShape > &  inputShapes ) const

overridevirtual

By default returns inputShapes if the number of inputs are equal to number of outputs, otherwise infers the output shapes from given input shapes and layer properties.

Parameters

[in]

inputShapes

The input shapes layer has.

Returns

A vector to the inferred output shape.

Reimplemented from Layer.

Definition at line 153 of file QLstmLayer.cpp.

{
    ARMNN_ASSERT(inputShapes.size() == 3);
 
    // Get input values for validation
    unsigned int batchSize = inputShapes[0][0];
    unsigned int outputSize = inputShapes[1][1];
    unsigned int numUnits = inputShapes[2][1];
 
    std::vector<TensorShape> outShapes;
    outShapes.push_back(TensorShape({ batchSize, outputSize })); // outputStateOut
    outShapes.push_back(TensorShape({ batchSize, numUnits })); // cellStateOut
    outShapes.push_back(TensorShape({ batchSize, outputSize })); // output
 
    return outShapes;
}

References ARMNN_ASSERT.

Referenced by QLstmLayer::ValidateTensorShapesFromInputs().

ValidateTensorShapesFromInputs()

void ValidateTensorShapesFromInputs ( )

overridevirtual

Check if the input tensor shape(s) will lead to a valid configuration of QLstmLayer.

Parameters

[in]

shapeInferenceMethod

Indicates if output shape shall be overwritten or just validated.

Implements Layer.

Definition at line 170 of file QLstmLayer.cpp.

{
    VerifyLayerConnections(3, CHECK_LOCATION());
 
    const TensorShape& outputShape = GetOutputSlot(0).GetTensorInfo().GetShape();
 
    VerifyShapeInferenceType(outputShape, m_ShapeInferenceMethod);
 
    auto inferredShapes = InferOutputShapes(
    {
        GetInputSlot(0).GetTensorInfo().GetShape(), // input
        GetInputSlot(1).GetTensorInfo().GetShape(), // previousOutputIn
        GetInputSlot(2).GetTensorInfo().GetShape()  // previousCellStateIn
    });
 
    ARMNN_ASSERT(inferredShapes.size() == 3);
 
    // Check if the weights are nullptr for basic params
    ARMNN_ASSERT_MSG(m_BasicParameters.m_InputToForgetWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_InputToForgetWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_InputToCellWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_InputToCellWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_InputToOutputWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_InputToOutputWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_RecurrentToForgetWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_RecurrentToForgetWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_RecurrentToCellWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_RecurrentToCellWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_RecurrentToOutputWeights != nullptr,
            "QLstmLayer: m_BasicParameters.m_RecurrentToOutputWeights should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_ForgetGateBias != nullptr,
            "QLstmLayer: m_BasicParameters.m_ForgetGateBias should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_CellBias != nullptr,
            "QLstmLayer: m_BasicParameters.m_CellBias should not be null.");
    ARMNN_ASSERT_MSG(m_BasicParameters.m_OutputGateBias != nullptr,
            "QLstmLayer: m_BasicParameters.m_OutputGateBias should not be null.");
 
    if (!m_Param.m_CifgEnabled)
    {
        ARMNN_ASSERT_MSG(m_CifgParameters.m_InputToInputWeights != nullptr,
                "QLstmLayer: m_CifgParameters.m_InputToInputWeights should not be null.");
        ARMNN_ASSERT_MSG(m_CifgParameters.m_RecurrentToInputWeights != nullptr,
                "QLstmLayer: m_CifgParameters.m_RecurrentToInputWeights should not be null.");
        ARMNN_ASSERT_MSG(m_CifgParameters.m_InputGateBias != nullptr,
                "QLstmLayer: m_CifgParameters.m_InputGateBias should not be null.");
 
        ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "QLstmLayer");
    }
    else
    {
        ARMNN_ASSERT_MSG(m_CifgParameters.m_InputToInputWeights == nullptr,
                "QLstmLayer: m_CifgParameters.m_InputToInputWeights should not have a value when CIFG is enabled.");
        ARMNN_ASSERT_MSG(m_CifgParameters.m_RecurrentToInputWeights == nullptr,
                "QLstmLayer: m_CifgParameters.m_RecurrentToInputWeights should "
                             "not have a value when CIFG is enabled.");
        ARMNN_ASSERT_MSG(m_CifgParameters.m_InputGateBias == nullptr,
                "QLstmLayer: m_CifgParameters.m_InputGateBias should not have a value when CIFG is enabled.");
 
        ValidateAndCopyShape(outputShape, inferredShapes[0], m_ShapeInferenceMethod, "QLstmLayer");
    }
 
    if (m_Param.m_ProjectionEnabled)
    {
        ARMNN_ASSERT_MSG(m_ProjectionParameters.m_ProjectionWeights != nullptr,
                         "QLstmLayer: m_ProjectionParameters.m_ProjectionWeights should not be null.");
    }
 
    if (m_Param.m_PeepholeEnabled)
    {
        if (!m_Param.m_CifgEnabled) {
            ARMNN_ASSERT_MSG(m_PeepholeParameters.m_CellToInputWeights != nullptr,
                    "QLstmLayer: m_PeepholeParameters.m_CellToInputWeights should not be null "
                    "when Peephole is enabled and CIFG is disabled.");
        }
 
        ARMNN_ASSERT_MSG(m_PeepholeParameters.m_CellToForgetWeights != nullptr,
                         "QLstmLayer: m_PeepholeParameters.m_CellToForgetWeights should not be null.");
        ARMNN_ASSERT_MSG(m_PeepholeParameters.m_CellToOutputWeights != nullptr,
                         "QLstmLayer: m_PeepholeParameters.m_CellToOutputWeights should not be null.");
    }
 
    ValidateAndCopyShape(
            GetOutputSlot(1).GetTensorInfo().GetShape(), inferredShapes[1], m_ShapeInferenceMethod, "QLstmLayer", 1);
    ValidateAndCopyShape(
            GetOutputSlot(2).GetTensorInfo().GetShape(), inferredShapes[2], m_ShapeInferenceMethod, "QLstmLayer", 2);
 
    if (m_Param.m_LayerNormEnabled)
    {
        if(!m_Param.m_CifgEnabled)
        {
            ARMNN_ASSERT_MSG(m_LayerNormParameters.m_InputLayerNormWeights != nullptr,
                             "QLstmLayer: m_LayerNormParameters.m_InputLayerNormWeights should not be null.");
        }
        ARMNN_ASSERT_MSG(m_LayerNormParameters.m_ForgetLayerNormWeights != nullptr,
                         "QLstmLayer: m_LayerNormParameters.m_ForgetLayerNormWeights should not be null.");
        ARMNN_ASSERT_MSG(m_LayerNormParameters.m_CellLayerNormWeights != nullptr,
                         "QLstmLayer: m_LayerNormParameters.m_CellLayerNormWeights should not be null.");
        ARMNN_ASSERT_MSG(m_LayerNormParameters.m_OutputLayerNormWeights != nullptr,
                         "QLstmLayer: m_LayerNormParameters.m_UutputLayerNormWeights should not be null.");
    }
}

References ARMNN_ASSERT, ARMNN_ASSERT_MSG, CHECK_LOCATION, Layer::GetInputSlot(), Layer::GetOutputSlot(), TensorInfo::GetShape(), armnn::GetTensorInfo(), InputSlot::GetTensorInfo(), OutputSlot::GetTensorInfo(), QLstmLayer::InferOutputShapes(), QLstmLayer::m_BasicParameters, QLstmBasicParameters::m_CellBias, QLstmOptPeepholeParameters::m_CellToForgetWeights, QLstmOptPeepholeParameters::m_CellToInputWeights, QLstmOptPeepholeParameters::m_CellToOutputWeights, QLstmDescriptor::m_CifgEnabled, QLstmLayer::m_CifgParameters, QLstmBasicParameters::m_ForgetGateBias, QLstmOptCifgParameters::m_InputGateBias, QLstmBasicParameters::m_InputToCellWeights, QLstmBasicParameters::m_InputToForgetWeights, QLstmOptCifgParameters::m_InputToInputWeights, QLstmBasicParameters::m_InputToOutputWeights, QLstmBasicParameters::m_OutputGateBias, LayerWithParameters< QLstmDescriptor >::m_Param, QLstmDescriptor::m_PeepholeEnabled, QLstmLayer::m_PeepholeParameters, QLstmDescriptor::m_ProjectionEnabled, QLstmLayer::m_ProjectionParameters, QLstmOptProjectionParameters::m_ProjectionWeights, QLstmBasicParameters::m_RecurrentToCellWeights, QLstmBasicParameters::m_RecurrentToForgetWeights, QLstmOptCifgParameters::m_RecurrentToInputWeights, QLstmBasicParameters::m_RecurrentToOutputWeights, Layer::m_ShapeInferenceMethod, Layer::ValidateAndCopyShape(), Layer::VerifyLayerConnections(), and Layer::VerifyShapeInferenceType().

Member Data Documentation

m_BasicParameters

QLstmBasicParameters m_BasicParameters

Definition at line 83 of file QLstmLayer.hpp.

Referenced by NetworkImpl::AddQLstmLayer(), QLstmLayer::Clone(), QLstmLayer::CreateWorkload(), QLstmLayer::ExecuteStrategy(), QLstmLayer::GetConstantTensorsByRef(), and QLstmLayer::ValidateTensorShapesFromInputs().

m_CifgParameters

QLstmOptCifgParameters m_CifgParameters

Definition at line 84 of file QLstmLayer.hpp.

Referenced by QLstmLayer::Clone(), QLstmLayer::CreateWorkload(), QLstmLayer::ExecuteStrategy(), QLstmLayer::GetConstantTensorsByRef(), and QLstmLayer::ValidateTensorShapesFromInputs().

m_LayerNormParameters

QLstmOptLayerNormParameters m_LayerNormParameters

Definition at line 87 of file QLstmLayer.hpp.

Referenced by QLstmLayer::Clone(), QLstmLayer::CreateWorkload(), QLstmLayer::ExecuteStrategy(), and QLstmLayer::GetConstantTensorsByRef().

m_PeepholeParameters

QLstmOptPeepholeParameters m_PeepholeParameters

Definition at line 86 of file QLstmLayer.hpp.

Referenced by QLstmLayer::Clone(), QLstmLayer::CreateWorkload(), QLstmLayer::ExecuteStrategy(), QLstmLayer::GetConstantTensorsByRef(), and QLstmLayer::ValidateTensorShapesFromInputs().

m_ProjectionParameters

QLstmOptProjectionParameters m_ProjectionParameters

Definition at line 85 of file QLstmLayer.hpp.

Referenced by QLstmLayer::Clone(), QLstmLayer::CreateWorkload(), QLstmLayer::ExecuteStrategy(), QLstmLayer::GetConstantTensorsByRef(), and QLstmLayer::ValidateTensorShapesFromInputs().

---

The documentation for this class was generated from the following files:

• src/armnn/layers/QLstmLayer.hpp
• src/armnn/layers/QLstmLayer.cpp