NeonSplitterWorkload Class Reference

#include <NeonSplitterWorkload.hpp>

Inheritance diagram for NeonSplitterWorkload:

Public Member Functions
	NeonSplitterWorkload (const SplitterQueueDescriptor &descriptor, const WorkloadInfo &info)
void	Execute () const override
Public Member Functions inherited from BaseWorkload< SplitterQueueDescriptor >
	BaseWorkload (const SplitterQueueDescriptor &descriptor, const WorkloadInfo &info)
void	ExecuteAsync (WorkingMemDescriptor &workingMemDescriptor) override
void	PostAllocationConfigure () override
const SplitterQueueDescriptor &	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< SplitterQueueDescriptor >
SplitterQueueDescriptor	m_Data
const profiling::ProfilingGuid	m_Guid

Detailed Description

Definition at line 22 of file NeonSplitterWorkload.hpp.

Constructor & Destructor Documentation

NeonSplitterWorkload()

NeonSplitterWorkload	(	const SplitterQueueDescriptor &	descriptor,
		const WorkloadInfo &	info
	)

Definition at line 56 of file NeonSplitterWorkload.cpp.

References armnn::ComputeSplitAxis(), ViewsDescriptor::GetNumDimensions(), BaseWorkload< SplitterQueueDescriptor >::m_Data, QueueDescriptor::m_Inputs, QueueDescriptor::m_Outputs, and QueueDescriptorWithParameters< LayerDescriptor >::m_Parameters.

    : BaseWorkload<SplitterQueueDescriptor>(descriptor, info)
{
    bool allOutputsAreSubtensors = true;

    // Check that all outputs are sub-tensors
    for (auto output : m_Data.m_Outputs)
    {
        if (output && !output->GetParent())
        {
            // Non sub-tensor input found so we need to execute the split function
            allOutputsAreSubtensors = false;
            break;
        }
    }

    if (allOutputsAreSubtensors)
    {
        // Can skip configuring the split function since it's not executed
        return;
    }

    arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();

    std::vector<arm_compute::ITensor *> aclOutputs;
    for (auto output : m_Data.m_Outputs)
    {
        arm_compute::ITensor& aclOutput  = PolymorphicPointerDowncast<IAclTensorHandle>(output)->GetTensor();
        aclOutputs.emplace_back(&aclOutput);
    }

    // Create the layer function
    std::unique_ptr<arm_compute::NESplit> layer(new arm_compute::NESplit());

    // Configure input and output tensors
    std::set<unsigned int> splitAxis = ComputeSplitAxis(descriptor.m_Parameters, m_Data.m_Inputs[0]->GetShape());
    if (splitAxis.size() != 1)
    {
        throw InvalidArgumentException("Cannot derive split axis from SplitterDescriptor");
    }

    unsigned int aclAxis = CalcAclAxis(descriptor.m_Parameters.GetNumDimensions(), *splitAxis.begin());
    layer->configure(&input, aclOutputs, aclAxis);

    // Prepare
    layer->prepare();
    m_Layer.reset(layer.release());
}

Member Function Documentation

Execute()

void Execute ( ) const

overridevirtual

Implements IWorkload.

Definition at line 105 of file NeonSplitterWorkload.cpp.

References ARMNN_SCOPED_PROFILING_EVENT_NEON.

{
    if (m_Layer)
    {
        ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonSplitterWorkload_Execute");
        m_Layer->run();
    }
}

---

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

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