NeonConvolution3dWorkload Class Reference

#include <NeonConvolution3dWorkload.hpp>

Inheritance diagram for NeonConvolution3dWorkload:

Collaboration diagram for NeonConvolution3dWorkload:

Public Member Functions
	NeonConvolution3dWorkload (const Convolution3dQueueDescriptor &descriptor, const WorkloadInfo &info, std::shared_ptr< arm_compute::MemoryManagerOnDemand > &memoryManager, const bool isFastMathENabled=false)
void	Execute () const override
Public Member Functions inherited from NeonBaseWorkload< Convolution3dQueueDescriptor >
	NeonBaseWorkload (const Convolution3dQueueDescriptor &descriptor, const WorkloadInfo &info)
void	ReplaceInputTensorHandle (ITensorHandle *tensorHandle, unsigned int slot) override
void	ReplaceOutputTensorHandle (ITensorHandle *tensorHandle, unsigned int slot) override
Public Member Functions inherited from BaseWorkload< Convolution3dQueueDescriptor >
	BaseWorkload (const Convolution3dQueueDescriptor &descriptor, const WorkloadInfo &info)
virtual const std::string &	GetName () const override
void	ExecuteAsync (ExecutionData &executionData) override
void	PostAllocationConfigure () override
const Convolution3dQueueDescriptor &	GetData () const
arm::pipe::ProfilingGuid	GetGuid () const final
virtual bool	SupportsTensorHandleReplacement () const override
Public Member Functions inherited from IWorkload
virtual	~IWorkload ()
virtual arm::pipe::ProfilingGuid	GetGuid () const =0
virtual bool	SupportsTensorHandleReplacement () const =0
virtual const std::string &	GetName () const =0
virtual void	RegisterDebugCallback (const DebugCallbackFunction &)
virtual armnn::Optional< armnn::MemoryRequirements >	GetMemoryRequirements ()

Additional Inherited Members
Protected Member Functions inherited from NeonBaseWorkload< Convolution3dQueueDescriptor >
virtual void	Reconfigure ()
Protected Attributes inherited from BaseWorkload< Convolution3dQueueDescriptor >
Convolution3dQueueDescriptor	m_Data
const arm::pipe::ProfilingGuid	m_Guid
const std::string	m_Name

Detailed Description

Definition at line 27 of file NeonConvolution3dWorkload.hpp.

Constructor & Destructor Documentation

NeonConvolution3dWorkload()

NeonConvolution3dWorkload	(	const Convolution3dQueueDescriptor &	descriptor,
		const WorkloadInfo &	info,
		std::shared_ptr< arm_compute::MemoryManagerOnDemand > &	memoryManager,
		const bool	isFastMathENabled = false
	)

Definition at line 60 of file NeonConvolution3dWorkload.cpp.

    : NeonBaseWorkload<Convolution3dQueueDescriptor>(descriptor, info)
{
    IgnoreUnused(memoryManager);
 
    using arm_compute::NEConv3D;
    uint32_t numInputs = m_Data.m_Parameters.m_BiasEnabled ? 3: 2;
    m_Data.ValidateInputsOutputs("NeonConvolution3dWorkload", numInputs, 1);
 
    arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
    arm_compute::ITensor& weights = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
    arm_compute::ITensor* biasesPtr = nullptr;
    if (m_Data.m_Parameters.m_BiasEnabled)
    {
        biasesPtr = &PolymorphicDowncast<IAclTensorHandle *>(m_Data.m_Inputs[2])->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);
    weights.info()->set_data_layout(aclDataLayout);
    output.info()->set_data_layout(aclDataLayout);
 
    const arm_compute::Conv3dInfo aclConv3DInfo = ComputeConv3DInfo(descriptor, isFastMathEnabled);
 
    auto convolutionLayer = std::make_unique<arm_compute::NEConv3D>();
    convolutionLayer->configure(&input,
                                &weights,
                                biasesPtr,
                                &output,
                                aclConv3DInfo);
 
    // Add details for profiling output
    WorkloadInfo detailsInfo;
 
    detailsInfo.m_InputTensorInfos = info.m_InputTensorInfos;
    detailsInfo.m_OutputTensorInfos = info.m_OutputTensorInfos;
 
    // Report Profiling Details
    ARMNN_REPORT_PROFILING_WORKLOAD_DESC("NeonConvolution3dWorkload_Construct",
                                         descriptor.m_Parameters,
                                         detailsInfo,
                                         this->GetGuid());
 
    m_ConvolutionLayer.reset(convolutionLayer.release());
 
    ARMNN_ASSERT(m_ConvolutionLayer);
 
    m_ConvolutionLayer->prepare();
}

References armnn::IgnoreUnused(), armnn::info, Convolution3dDescriptor::m_BiasEnabled, BaseWorkload< Convolution3dQueueDescriptor >::m_Data, QueueDescriptor::m_Inputs, QueueDescriptor::m_Outputs, QueueDescriptorWithParameters< LayerDescriptor >::m_Parameters, and QueueDescriptor::ValidateInputsOutputs().

Member Function Documentation

Execute()

void Execute ( ) const

overridevirtual

Implements IWorkload.

Definition at line 114 of file NeonConvolution3dWorkload.cpp.

{
    ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID("NeonConvolution3dWorkload_Execute");
    m_ConvolutionLayer->run();
}

References ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID.

---

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

• src/backends/neon/workloads/NeonConvolution3dWorkload.hpp
• src/backends/neon/workloads/NeonConvolution3dWorkload.cpp