#include <NeonConvolution2dWorkload.hpp>

Inheritance diagram for NeonConvolution2dWorkload:

Public Member Functions
	NeonConvolution2dWorkload (const Convolution2dQueueDescriptor &descriptor, const WorkloadInfo &info, std::shared_ptr< arm_compute::MemoryManagerOnDemand > &memoryManager, const bool isFastMathENabled=false)

void	Execute () const override

arm_compute::ConvolutionMethod	GetConvolutionMethod () const

Public Member Functions inherited from BaseWorkload< Convolution2dQueueDescriptor >
	BaseWorkload (const Convolution2dQueueDescriptor &descriptor, const WorkloadInfo &info)

void	PostAllocationConfigure () override

const Convolution2dQueueDescriptor &	GetData () const

profiling::ProfilingGuid	GetGuid () const final

Public Member Functions inherited from IWorkload
virtual	~IWorkload ()

virtual void	RegisterDebugCallback (const DebugCallbackFunction &)

Additional Inherited Members
Protected Attributes inherited from BaseWorkload< Convolution2dQueueDescriptor >
const Convolution2dQueueDescriptor	m_Data

const profiling::ProfilingGuid	m_Guid

Detailed Description

Definition at line 27 of file NeonConvolution2dWorkload.hpp.

Constructor & Destructor Documentation

◆ NeonConvolution2dWorkload()

NeonConvolution2dWorkload	(	const Convolution2dQueueDescriptor &	descriptor,
		const WorkloadInfo &	info,
		std::shared_ptr< arm_compute::MemoryManagerOnDemand > &	memoryManager,
		const bool	isFastMathENabled = `false`
	)

Definition at line 66 of file NeonConvolution2dWorkload.cpp.

References BaseWorkload< Convolution2dQueueDescriptor >::m_Data, QueueDescriptor::m_Inputs, QueueDescriptor::m_Outputs, and QueueDescriptor::ValidateInputsOutputs().

     : BaseWorkload<Convolution2dQueueDescriptor>(descriptor, info)
 {
     using arm_compute::NEConvolutionLayer;
 
     m_Data.ValidateInputsOutputs("NeonConvolution2dWorkload", 1, 1);
 
     // todo: check tensor shapes match.
 
     arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
     arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
 
     arm_compute::DataLayout aclDataLayout = ConvertDataLayout(m_Data.m_Parameters.m_DataLayout);
     input.info()->set_data_layout(aclDataLayout);
     output.info()->set_data_layout(aclDataLayout);
 
     m_KernelTensor = std::make_unique<arm_compute::Tensor>();
     BuildArmComputeTensor(*m_KernelTensor, m_Data.m_Weight->GetTensorInfo(), m_Data.m_Parameters.m_DataLayout);
 
     if (m_Data.m_Parameters.m_BiasEnabled)
     {
         m_BiasTensor = std::make_unique<arm_compute::Tensor>();
         BuildArmComputeTensor(*m_BiasTensor, m_Data.m_Bias->GetTensorInfo(), m_Data.m_Parameters.m_DataLayout);
     }
 
     arm_compute::PadStrideInfo padStrideInfo = BuildArmComputePadStrideInfo(m_Data.m_Parameters);
 
     const arm_compute::Size2D aclDilationInfo = BuildArmComputeSize2D(m_Data.m_Parameters.m_DilationX,
                                                                       m_Data.m_Parameters.m_DilationY);
 
     const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
 
     auto convolutionLayer = std::make_unique<arm_compute::NEConvolutionLayer>(memoryManager);
     convolutionLayer->configure(&input,
                                 m_KernelTensor.get(),
                                 m_BiasTensor.get(),
                                 &output,
                                 padStrideInfo,
                                 arm_compute::WeightsInfo(),
                                 aclDilationInfo,
                                 activationInfo,
                                 isFastMathEnabled);
 
     m_ConvolutionMethod =
         convolutionLayer->get_convolution_method(input.info(),
                                                  m_KernelTensor->info(),
                                                  output.info(),
                                                  padStrideInfo,
                                                  arm_compute::WeightsInfo(),
                                                  aclDilationInfo,
                                                  activationInfo,
                                                  isFastMathEnabled);
 
     m_ConvolutionLayer.reset(convolutionLayer.release());
 
     ARMNN_ASSERT(m_ConvolutionLayer);
 
     InitializeArmComputeTensorData(*m_KernelTensor, m_Data.m_Weight);
 
     if (m_Data.m_Parameters.m_BiasEnabled)
     {
         InitializeArmComputeTensorData(*m_BiasTensor, m_Data.m_Bias);
     }
 
     m_ConvolutionLayer->prepare();
     FreeUnusedTensors();
 }

Member Function Documentation

◆ Execute()

void Execute ( ) const

overridevirtual

Implements IWorkload.

Definition at line 138 of file NeonConvolution2dWorkload.cpp.

References ARMNN_SCOPED_PROFILING_EVENT_NEON.

 {
     ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonConvolution2dWorkload_Execute");
     m_ConvolutionLayer->run();
 }

◆ GetConvolutionMethod()

arm_compute::ConvolutionMethod GetConvolutionMethod ( ) const

Definition at line 144 of file NeonConvolution2dWorkload.cpp.

 {
     return m_ConvolutionMethod;
 }

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

src/backends/neon/workloads/NeonConvolution2dWorkload.hpp
src/backends/neon/workloads/NeonConvolution2dWorkload.cpp