#include <NeonBackend.hpp>

Inheritance diagram for NeonBackend:

Public Member Functions
	NeonBackend ()=default

	~NeonBackend ()=default

const BackendId &	GetId () const override

IBackendInternal::IMemoryManagerUniquePtr	CreateMemoryManager () const override

IWorkloadFactoryPtr	CreateWorkloadFactory (const IBackendInternal::IMemoryManagerSharedPtr &memoryManager=nullptr) const override

IWorkloadFactoryPtr	CreateWorkloadFactory (class TensorHandleFactoryRegistry &tensorHandleFactoryRegistry) const override

IWorkloadFactoryPtr	CreateWorkloadFactory (const IMemoryManagerSharedPtr &memoryManager, const ModelOptions &modelOptions) const override

IWorkloadFactoryPtr	CreateWorkloadFactory (class TensorHandleFactoryRegistry &tensorHandleFactoryRegistry, const ModelOptions &modelOptions) const override

IBackendInternal::IBackendContextPtr	CreateBackendContext (const IRuntime::CreationOptions &) const override
	Create the runtime context of the backend. More...

IBackendInternal::IBackendProfilingContextPtr	CreateBackendProfilingContext (const IRuntime::CreationOptions &, IBackendProfilingPtr &backendProfiling) override
	Create context specifically used for profiling interaction from backends. More...

IBackendInternal::ILayerSupportSharedPtr	GetLayerSupport () const override

IBackendInternal::ILayerSupportSharedPtr	GetLayerSupport (const ModelOptions &modelOptions) const override

OptimizationViews	OptimizeSubgraphView (const SubgraphView &subgraph, const ModelOptions &modelOptions) const override

std::vector< ITensorHandleFactory::FactoryId >	GetHandleFactoryPreferences () const override
	(Optional) Returns a vector of supported TensorHandleFactory ids in preference order. More...

void	RegisterTensorHandleFactories (class TensorHandleFactoryRegistry &registry) override
	(Optional) Register TensorHandleFactories Either this method or CreateMemoryManager() and IWorkloadFactory::CreateTensor() IWorkloadFactory::CreateSubtensor() methods must be implemented. More...

IBackendInternal::IBackendSpecificModelContextPtr	CreateBackendSpecificModelContext (const ModelOptions &modelOptions) const override

BackendCapabilities	GetCapabilities () const override
	Returns a BackendCapability if the backend lists the capability The BackendCapability must then be inspected to check whether or not that BackendCapability is supported Otherwise returns an EmptyOptional if the BackendCapability is unlisted. More...

std::unique_ptr< ICustomAllocator >	GetDefaultAllocator () const override
	Returns the default memory allocator for the backend. More...

Public Member Functions inherited from IBackendInternal
	~IBackendInternal () override=default
	Allow backends created by the factory function to be destroyed through IBackendInternal. More...

virtual IWorkloadFactoryPtr	CreateWorkloadFactory (class TensorHandleFactoryRegistry &tensorHandleFactoryRegistry, const ModelOptions &modelOptions, MemorySourceFlags inputFlags, MemorySourceFlags outputFlags) const

virtual OptimizationViews	OptimizeSubgraphView (const SubgraphView &subgraph) const

bool	SupportsTensorAllocatorAPI () const

ITensorHandleFactory::FactoryId	GetBackwardCompatibleFavoriteHandleFactory ()

virtual void	RegisterTensorHandleFactories (class TensorHandleFactoryRegistry &registry, MemorySourceFlags inputFlags, MemorySourceFlags outputFlags)
	(Optional) Register TensorHandleFactories Either this method or CreateMemoryManager() and IWorkloadFactory::CreateTensor() IWorkloadFactory::CreateSubtensor() methods must be implemented. More...

virtual bool	UseCustomMemoryAllocator (std::shared_ptr< ICustomAllocator > allocator, armnn::Optional< std::string &> errMsg)
	Signals the backend to use a custom memory allocator provided by the user. More...

virtual unsigned int	GetNumberOfCacheFiles () const
	Returns the number of files cached if backend supports caching. More...

virtual ExecutionData	CreateExecutionData (WorkingMemDescriptor &workingMemDescriptor) const
	Returns ExecutionData for the backend. More...

virtual void	UpdateExecutionData (ExecutionData &executionData, WorkingMemDescriptor &workingMemDescriptor) const
	Update the ExecutionData for a layer. More...

Static Public Member Functions
static const BackendId &	GetIdStatic ()

Static Public Member Functions inherited from IBackendInternal
static constexpr BackendVersion	GetApiVersion ()
	Returns the version of the Backend API. More...

Additional Inherited Members
Public Types inherited from IBackendInternal
using	IWorkloadFactoryPtr = std::unique_ptr< IWorkloadFactory >

using	IBackendContextPtr = std::unique_ptr< IBackendContext >

using	IBackendProfilingContextPtr = std::shared_ptr< arm::pipe::IBackendProfilingContext >
	This is the bridge between backend and backend profiling we'll keep it in the backend namespace. More...

using	IBackendProfilingPtr = std::unique_ptr< arm::pipe::IBackendProfiling >

using	ILayerSupportSharedPtr = std::shared_ptr< ILayerSupport >

using	IBackendSpecificModelContextPtr = std::shared_ptr< IBackendModelContext >

using	IMemoryManagerUniquePtr = std::unique_ptr< IMemoryManager >

using	IMemoryManagerSharedPtr = std::shared_ptr< IMemoryManager >

Protected Member Functions inherited from IBackendInternal
	IBackendInternal ()=default
	Creation must be done through a specific backend interface. More...

Protected Member Functions inherited from IBackend
	IBackend ()

virtual	~IBackend ()

Detailed Description

Definition at line 26 of file NeonBackend.hpp.

Constructor & Destructor Documentation

◆ NeonBackend()

NeonBackend ( )

default

◆ ~NeonBackend()

~NeonBackend ( )

default

Member Function Documentation

◆ CreateBackendContext()

IBackendInternal::IBackendContextPtr CreateBackendContext ( const IRuntime::CreationOptions & ) const

overridevirtual

Create the runtime context of the backend.

Implementations may return a default-constructed IBackendContextPtr if no context is needed at runtime. Implementations must throw BackendUnavailableException if the backend cannot be used (for example, necessary accelerator hardware is not present). The default implementation always returns a default-constructed pointer.

Reimplemented from IBackendInternal.

Definition at line 107 of file NeonBackend.cpp.

Referenced by NeonBackend::GetId().

 {
     return IBackendContextPtr{};
 }

◆ CreateBackendProfilingContext()

IBackendInternal::IBackendProfilingContextPtr CreateBackendProfilingContext	(	const IRuntime::CreationOptions &	creationOptions,
		IBackendProfilingPtr &	backendProfiling
	)

overridevirtual

Create context specifically used for profiling interaction from backends.

Reimplemented from IBackendInternal.

Definition at line 112 of file NeonBackend.cpp.

Referenced by NeonBackend::GetId().

 {
     return IBackendProfilingContextPtr{};
 }

◆ CreateBackendSpecificModelContext()

IBackendInternal::IBackendSpecificModelContextPtr CreateBackendSpecificModelContext ( const ModelOptions & modelOptions ) const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 118 of file NeonBackend.cpp.

Referenced by NeonBackend::CreateWorkloadFactory(), NeonBackend::GetId(), and NeonBackend::GetLayerSupport().

 {
     return IBackendSpecificModelContextPtr{new NeonBackendModelContext{modelOptions}};
 }

◆ CreateMemoryManager()

IBackendInternal::IMemoryManagerUniquePtr CreateMemoryManager ( ) const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 50 of file NeonBackend.cpp.

References BaseMemoryManager::Offset.

Referenced by NeonBackend::GetId().

 {
     return std::make_unique<NeonMemoryManager>(std::make_unique<arm_compute::Allocator>(),
                                                BaseMemoryManager::MemoryAffinity::Offset);
 }

◆ CreateWorkloadFactory() [1/4]

IBackendInternal::IWorkloadFactoryPtr CreateWorkloadFactory ( const IBackendInternal::IMemoryManagerSharedPtr & memoryManager = nullptr ) const

overridevirtual

Implements IBackendInternal.

Definition at line 56 of file NeonBackend.cpp.

Referenced by NeonBackend::GetId().

 {
     return std::make_unique<NeonWorkloadFactory>(
         PolymorphicPointerDowncast<NeonMemoryManager>(memoryManager));
 }

◆ CreateWorkloadFactory() [2/4]

IBackendInternal::IWorkloadFactoryPtr CreateWorkloadFactory ( class TensorHandleFactoryRegistry & tensorHandleFactoryRegistry ) const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 70 of file NeonBackend.cpp.

References BaseMemoryManager::Offset, TensorHandleFactoryRegistry::RegisterCopyAndImportFactoryPair(), TensorHandleFactoryRegistry::RegisterFactory(), and TensorHandleFactoryRegistry::RegisterMemoryManager().

 {
     auto memoryManager = std::make_shared<NeonMemoryManager>(std::make_unique<arm_compute::Allocator>(),
                                                              BaseMemoryManager::MemoryAffinity::Offset);
 
     tensorHandleFactoryRegistry.RegisterMemoryManager(memoryManager);
 
     auto factory = std::make_unique<NeonTensorHandleFactory>(memoryManager);
     // Register copy and import factory pair
     tensorHandleFactoryRegistry.RegisterCopyAndImportFactoryPair(factory->GetId(), factory->GetId());
     // Register the factory
     tensorHandleFactoryRegistry.RegisterFactory(std::move(factory));
 
 
     return std::make_unique<NeonWorkloadFactory>(
         PolymorphicPointerDowncast<NeonMemoryManager>(memoryManager));
 }

◆ CreateWorkloadFactory() [3/4]

IBackendInternal::IWorkloadFactoryPtr CreateWorkloadFactory	(	const IMemoryManagerSharedPtr &	memoryManager,
		const ModelOptions &	modelOptions
	)		const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 63 of file NeonBackend.cpp.

References NeonBackend::CreateBackendSpecificModelContext().

 {
     return std::make_unique<NeonWorkloadFactory>(
         PolymorphicPointerDowncast<NeonMemoryManager>(memoryManager), CreateBackendSpecificModelContext(modelOptions));
 }

◆ CreateWorkloadFactory() [4/4]

IBackendInternal::IWorkloadFactoryPtr CreateWorkloadFactory	(	class TensorHandleFactoryRegistry &	tensorHandleFactoryRegistry,
		const ModelOptions &	modelOptions
	)		const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 89 of file NeonBackend.cpp.

References NeonBackend::CreateBackendSpecificModelContext(), BaseMemoryManager::Offset, TensorHandleFactoryRegistry::RegisterCopyAndImportFactoryPair(), TensorHandleFactoryRegistry::RegisterFactory(), and TensorHandleFactoryRegistry::RegisterMemoryManager().

 {
     auto memoryManager = std::make_shared<NeonMemoryManager>(std::make_unique<arm_compute::Allocator>(),
                                                              BaseMemoryManager::MemoryAffinity::Offset);
 
     tensorHandleFactoryRegistry.RegisterMemoryManager(memoryManager);
 
     auto factory = std::make_unique<NeonTensorHandleFactory>(memoryManager);
     // Register copy and import factory pair
     tensorHandleFactoryRegistry.RegisterCopyAndImportFactoryPair(factory->GetId(), factory->GetId());
     // Register the factory
     tensorHandleFactoryRegistry.RegisterFactory(std::move(factory));
 
     return std::make_unique<NeonWorkloadFactory>(
         PolymorphicPointerDowncast<NeonMemoryManager>(memoryManager), CreateBackendSpecificModelContext(modelOptions));
 }

◆ GetCapabilities()

BackendCapabilities GetCapabilities ( ) const

inlineoverridevirtual

Returns a BackendCapability if the backend lists the capability The BackendCapability must then be inspected to check whether or not that BackendCapability is supported Otherwise returns an EmptyOptional if the BackendCapability is unlisted.

Reimplemented from IBackendInternal.

Definition at line 65 of file NeonBackend.hpp.

References armnn::cpuAccCapabilities, and NeonBackend::GetDefaultAllocator().

     {
         return cpuAccCapabilities;
     };

◆ GetDefaultAllocator()

std::unique_ptr< ICustomAllocator > GetDefaultAllocator ( ) const

overridevirtual

Returns the default memory allocator for the backend.

Returns: - Returns unique pointer to the Default Allocator of the Backend

Reimplemented from IBackendInternal.

Definition at line 455 of file NeonBackend.cpp.

Referenced by NeonBackend::GetCapabilities().

 {
     return std::make_unique<DefaultAllocator>();
 }

◆ GetHandleFactoryPreferences()

std::vector< ITensorHandleFactory::FactoryId > GetHandleFactoryPreferences ( ) const

overridevirtual

(Optional) Returns a vector of supported TensorHandleFactory ids in preference order.

Reimplemented from IBackendInternal.

Definition at line 436 of file NeonBackend.cpp.

References NeonTensorHandleFactory::GetIdStatic().

Referenced by NeonBackend::GetId().

 {
     return std::vector<ITensorHandleFactory::FactoryId>() = { NeonTensorHandleFactory::GetIdStatic() };
 }

◆ GetId()

const BackendId& GetId ( ) const

inlineoverridevirtual

Implements IBackend.

Definition at line 33 of file NeonBackend.hpp.

References NeonBackend::CreateBackendContext(), NeonBackend::CreateBackendProfilingContext(), NeonBackend::CreateBackendSpecificModelContext(), NeonBackend::CreateMemoryManager(), NeonBackend::CreateWorkloadFactory(), NeonBackend::GetHandleFactoryPreferences(), NeonBackend::GetIdStatic(), NeonBackend::GetLayerSupport(), NeonBackend::OptimizeSubgraphView(), and NeonBackend::RegisterTensorHandleFactories().

33 { return GetIdStatic(); }

armnn::NeonBackend::GetIdStatic

static const BackendId & GetIdStatic()

Definition: NeonBackend.cpp:44

◆ GetIdStatic()

const BackendId & GetIdStatic ( )

static

Definition at line 44 of file NeonBackend.cpp.

References armnn::NeonBackendId().

Referenced by NeonBackend::GetId().

 {
     static const BackendId s_Id{NeonBackendId()};
     return s_Id;
 }

◆ GetLayerSupport() [1/2]

IBackendInternal::ILayerSupportSharedPtr GetLayerSupport ( ) const

overridevirtual

Implements IBackendInternal.

Definition at line 124 of file NeonBackend.cpp.

Referenced by NeonBackend::GetId().

 {
     static ILayerSupportSharedPtr layerSupport
         {
             new NeonLayerSupport(IBackendInternal::IBackendSpecificModelContextPtr{})
         };
     return layerSupport;
 }

◆ GetLayerSupport() [2/2]

IBackendInternal::ILayerSupportSharedPtr GetLayerSupport ( const ModelOptions & modelOptions ) const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 133 of file NeonBackend.cpp.

References NeonBackend::CreateBackendSpecificModelContext().

 {
     static ILayerSupportSharedPtr layerSupport
         {
             new NeonLayerSupport(CreateBackendSpecificModelContext(modelOptions))
         };
     return layerSupport;
 }

◆ OptimizeSubgraphView()

OptimizationViews OptimizeSubgraphView	(	const SubgraphView &	subgraph,
		const ModelOptions &	modelOptions
	)		const

overridevirtual

Reimplemented from IBackendInternal.

Definition at line 142 of file NeonBackend.cpp.

References armnn::Activation, armnn::Addition, OptimizationViews::AddUntouchedSubgraph(), armnn::BatchNormalization, SubgraphView::beginIConnectable(), Layer::BeginOutputSlots(), armnn::Convolution2d, armnn::DepthwiseConvolution2d, armnn::Division, SubgraphView::endIConnectable(), Layer::EndOutputSlots(), armnn::FullyConnected, Layer::GetAdditionalInformation(), InputSlot::GetConnectedOutputSlot(), Layer::GetGuid(), Layer::GetInputSlot(), Layer::GetName(), LayerWithParameters< Parameters >::GetParameters(), OptimizationViews::GetSubstitutions(), OutputSlot::GetTensorInfo(), Layer::GetType(), BatchNormalizationLayer::m_Beta, Convolution2dDescriptor::m_BiasEnabled, DepthwiseConvolution2dDescriptor::m_BiasEnabled, BatchNormalizationLayer::m_Gamma, BatchNormalizationLayer::m_Mean, BatchNormalizationLayer::m_Variance, armnn::Multiplication, armnn::NeonAdditionWorkloadValidate(), armnn::NeonBatchNormalizationValidate(), armnn::NeonConvolution2dWorkloadValidate(), armnn::NeonDepthwiseConvolutionWorkloadValidate(), armnn::NeonDivisionWorkloadValidate(), armnn::NeonFullyConnectedWorkloadValidate(), armnn::NeonMultiplicationWorkloadValidate(), armnn::NeonSubtractionWorkloadValidate(), armnn::Reduce, armnn::ReportUntouchedLayers(), and armnn::Subtraction.

Referenced by NeonBackend::GetId().

 {
     OptimizationViews optimizationViews(modelOptions);
 
     auto it = subgraph.endIConnectable();
     std::map<LayerGuid, Layer*> untouched;
 
     while (it != subgraph.beginIConnectable())
     {
         --it;
         Layer& base = *(PolymorphicDowncast<Layer*>(*it));
         untouched.insert({base.GetGuid(), &base});
     }
 
     it = subgraph.endIConnectable();
     while (it != subgraph.beginIConnectable())
     {
         --it;
         Layer& base = *(PolymorphicDowncast<Layer*>(*it));
 
         // Fuse activation into previous layer if supported by backend
         if ((base.GetType() == LayerType::DepthwiseConvolution2d || base.GetType() == LayerType::Convolution2d
              || base.GetType() == LayerType::BatchNormalization || base.GetType() == LayerType::FullyConnected
              || base.GetType() == LayerType::Addition || base.GetType() == LayerType::Multiplication
              || base.GetType() == LayerType::Subtraction || base.GetType() == LayerType::Division)
             && (base.GetAdditionalInformation<ActivationDescriptor>() == nullptr))
         {
             for (auto output = base.BeginOutputSlots(); output != base.EndOutputSlots(); ++output)
             {
                 if (output->GetNumConnections() == 1)
                 {
                     for (auto&& childInput : output->GetConnections())
                     {
                         if ((childInput->GetOwningLayer().GetType() == LayerType::Activation) &&
                             (checkDataTypeInputandOutput(childInput->GetOwningLayer())))
                         {
                             Layer& child = childInput->GetOwningLayer();
 
                             auto* activationLayer = PolymorphicDowncast<ActivationLayer*>(&child);
 
                             const std::string name = std::string("fused-") + child.GetName() + std::string("-into-") +
                                                      base.GetName();
 
                             // Get params from activation layer
                             ActivationDescriptor activationDesc = activationLayer->GetParameters();
 
                             if (base.GetType() == LayerType::Convolution2d)
                             {
                                 Convolution2dLayer* baseLayer = PolymorphicDowncast<Convolution2dLayer*>(&base);
 
                                 Optional<TensorInfo> biases;
 
                                 if (baseLayer->GetParameters().m_BiasEnabled)
                                 {
                                     biases = baseLayer->GetInputSlot(2).GetConnectedOutputSlot()->GetTensorInfo();
                                 }
 
                                 arm_compute::Status status = NeonConvolution2dWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetParameters(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         biases,
                                         false,
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseConvolution2dLayer<Convolution2dLayer>(optimizationViews,
                                                                                baseLayer,
                                                                                activationLayer,
                                                                                activationDesc,
                                                                                name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::DepthwiseConvolution2d)
                             {
                                 DepthwiseConvolution2dLayer* baseLayer =
                                         PolymorphicDowncast<DepthwiseConvolution2dLayer*>(&base);
 
                                 Optional<TensorInfo> biases;
 
                                 if (baseLayer->GetParameters().m_BiasEnabled)
                                 {
                                     biases = baseLayer->GetInputSlot(2).GetConnectedOutputSlot()->GetTensorInfo();
                                 }
 
                                 arm_compute::Status status = NeonDepthwiseConvolutionWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetParameters(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         biases,
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseDepthwiseConvolution2dLayer<DepthwiseConvolution2dLayer>(optimizationViews,
                                                                                                  baseLayer,
                                                                                                  activationLayer,
                                                                                                  activationDesc,
                                                                                                  name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::FullyConnected)
                             {
                                 FullyConnectedLayer* baseLayer = PolymorphicDowncast<FullyConnectedLayer*>(&base);
                                 FullyConnectedDescriptor descriptor = baseLayer->GetParameters();
 
                                 // As bias is optional only try to get TensorInfo from input if bias is enabled.
                                 Optional<TensorInfo> biases;
                                 if (descriptor.m_BiasEnabled)
                                 {
                                     biases = baseLayer->GetInputSlot(2).GetConnectedOutputSlot()->GetTensorInfo();
                                 }
 
                                 arm_compute::Status status = NeonFullyConnectedWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         biases,
                                         baseLayer->GetParameters(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseFullyConnectedLayer<FullyConnectedLayer>(optimizationViews,
                                                                                  baseLayer,
                                                                                  activationLayer,
                                                                                  activationDesc,
                                                                                  name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::BatchNormalization)
                             {
                                 BatchNormalizationLayer* baseLayer =
                                         PolymorphicDowncast<BatchNormalizationLayer*>(&base);
 
                                 arm_compute::Status status = NeonBatchNormalizationValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->m_Mean->GetTensorInfo(),
                                         baseLayer->m_Variance->GetTensorInfo(),
                                         baseLayer->m_Beta->GetTensorInfo(),
                                         baseLayer->m_Gamma->GetTensorInfo(),
                                         baseLayer->GetParameters(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     BatchNormalizationLayer* replacementLayer =
                                         FuseBatchNormalizationLayer<BatchNormalizationLayer>(optimizationViews,
                                                                                              baseLayer,
                                                                                              activationLayer,
                                                                                              activationDesc,
                                                                                              name);
 
                                     replacementLayer->m_Beta     = std::move(baseLayer->m_Beta);
                                     replacementLayer->m_Gamma    = std::move(baseLayer->m_Gamma);
                                     replacementLayer->m_Mean     = std::move(baseLayer->m_Mean);
                                     replacementLayer->m_Variance = std::move(baseLayer->m_Variance);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::Addition)
                             {
                                 AdditionLayer* baseLayer = PolymorphicDowncast<AdditionLayer*>(&base);
 
                                 arm_compute::Status status = NeonAdditionWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseAdditionLayer<AdditionLayer>(optimizationViews,
                                                                      baseLayer,
                                                                      activationLayer,
                                                                      activationDesc,
                                                                      name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::Division)
                             {
                                 DivisionLayer* baseLayer = PolymorphicDowncast<DivisionLayer*>(&base);
 
                                 arm_compute::Status status = NeonDivisionWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseDivisionLayer<DivisionLayer>(optimizationViews,
                                                                      baseLayer,
                                                                      activationLayer,
                                                                      activationDesc,
                                                                      name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::Multiplication)
                             {
                                 MultiplicationLayer* baseLayer = PolymorphicDowncast<MultiplicationLayer*>(&base);
 
                                 arm_compute::Status status = NeonMultiplicationWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseMultiplicationLayer<MultiplicationLayer>(optimizationViews,
                                                                                  baseLayer,
                                                                                  activationLayer,
                                                                                  activationDesc,
                                                                                  name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                             else if (base.GetType() == LayerType::Subtraction)
                             {
                                 SubtractionLayer* baseLayer = PolymorphicDowncast<SubtractionLayer*>(&base);
 
                                 arm_compute::Status status = NeonSubtractionWorkloadValidate(
                                         baseLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         baseLayer->GetInputSlot(1).GetConnectedOutputSlot()->GetTensorInfo(),
                                         activationLayer->GetInputSlot(0).GetConnectedOutputSlot()->GetTensorInfo(),
                                         &activationDesc);
 
                                 if (status)
                                 {
                                     FuseSubtractionLayer<SubtractionLayer>(optimizationViews,
                                                                            baseLayer,
                                                                            activationLayer,
                                                                            activationDesc,
                                                                            name);
                                     untouched.erase(baseLayer->GetGuid());
                                     untouched.erase(activationLayer->GetGuid());
                                 }
                             }
                         }
                     }
                 }
             }
         }
 
         // Separate reduce layer with multiple axes into multiple reduce layers with 1 axis.
         if (base.GetType() == LayerType::Reduce)
         {
             ReduceLayer* baseLayer            = PolymorphicDowncast<ReduceLayer*>(&base);
             ReduceDescriptor reduceDescriptor = baseLayer->GetParameters();
 
             if (!reduceDescriptor.m_vAxis.empty() && reduceDescriptor.m_vAxis.size() > 1)
             {
                 // Add new layers to the graph and connect them.
                 std::vector<IConnectableLayer*> layers = ChainReduceLayers<ReduceLayer>(optimizationViews,
                                                                                         baseLayer,
                                                                                         reduceDescriptor);
 
                 // Replace existing baselayer with new subgraph.
                 ReplaceLayers<ReduceLayer>(optimizationViews, baseLayer, layers);
                 untouched.erase(baseLayer->GetGuid());
             }
         }
     }
 
     if (optimizationViews.GetSubstitutions().empty())
     {
         optimizationViews.AddUntouchedSubgraph(SubgraphView(subgraph));
     }
     else
     {
         ReportUntouchedLayers(optimizationViews, untouched);
     }
 
     return optimizationViews;
 }

◆ RegisterTensorHandleFactories()

void RegisterTensorHandleFactories ( class TensorHandleFactoryRegistry & )

overridevirtual

(Optional) Register TensorHandleFactories Either this method or CreateMemoryManager() and IWorkloadFactory::CreateTensor() IWorkloadFactory::CreateSubtensor() methods must be implemented.

Reimplemented from IBackendInternal.

Definition at line 441 of file NeonBackend.cpp.

References BaseMemoryManager::Offset, TensorHandleFactoryRegistry::RegisterCopyAndImportFactoryPair(), TensorHandleFactoryRegistry::RegisterFactory(), and TensorHandleFactoryRegistry::RegisterMemoryManager().

Referenced by NeonBackend::GetId().

 {
     auto memoryManager = std::make_shared<NeonMemoryManager>(std::make_unique<arm_compute::Allocator>(),
                                                              BaseMemoryManager::MemoryAffinity::Offset);
 
     registry.RegisterMemoryManager(memoryManager);
 
     auto factory = std::make_unique<NeonTensorHandleFactory>(memoryManager);
     // Register copy and import factory pair
     registry.RegisterCopyAndImportFactoryPair(factory->GetId(), factory->GetId());
     // Register the factory
     registry.RegisterFactory(std::move(factory));
 }

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

src/backends/neon/NeonBackend.hpp
src/backends/neon/NeonBackend.cpp

Public Member Functions

Static Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ NeonBackend()

◆ ~NeonBackend()

Member Function Documentation

◆ CreateBackendContext()

◆ CreateBackendProfilingContext()

◆ CreateBackendSpecificModelContext()

◆ CreateMemoryManager()

◆ CreateWorkloadFactory() [1/4]

◆ CreateWorkloadFactory() [2/4]

◆ CreateWorkloadFactory() [3/4]

◆ CreateWorkloadFactory() [4/4]

◆ GetCapabilities()

◆ GetDefaultAllocator()

◆ GetHandleFactoryPreferences()

◆ GetId()

◆ GetIdStatic()

◆ GetLayerSupport() [1/2]

◆ GetLayerSupport() [2/2]

◆ OptimizeSubgraphView()

◆ RegisterTensorHandleFactories()