NeonConvertFp32ToBf16Workload Class Reference

#include <NeonConvertFp32ToBf16Workload.hpp>

Inheritance diagram for NeonConvertFp32ToBf16Workload:

Public Member Functions
	NeonConvertFp32ToBf16Workload (const ConvertFp32ToBf16QueueDescriptor &descriptor, const WorkloadInfo &info)
virtual void	Execute () const override
Public Member Functions inherited from MultiTypedWorkload< QueueDescriptor, InputDataType, OutputDataType >
	MultiTypedWorkload (const QueueDescriptor &descriptor, const WorkloadInfo &info)
Public Member Functions inherited from BaseWorkload< QueueDescriptor >
	BaseWorkload (const QueueDescriptor &descriptor, const WorkloadInfo &info)
void	PostAllocationConfigure () override
const QueueDescriptor &	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< QueueDescriptor >
const QueueDescriptor	m_Data
const profiling::ProfilingGuid	m_Guid

Detailed Description

Definition at line 15 of file NeonConvertFp32ToBf16Workload.hpp.

Constructor & Destructor Documentation

NeonConvertFp32ToBf16Workload()

NeonConvertFp32ToBf16Workload	(	const ConvertFp32ToBf16QueueDescriptor &	descriptor,
		const WorkloadInfo &	info
	)

Definition at line 18 of file NeonConvertFp32ToBf16Workload.cpp.

References armnn::GatherTensorHandlePairs(), BaseWorkload< QueueDescriptor >::m_Data, and QueueDescriptor::ValidateInputsOutputs().

    : Float32ToBFloat16Workload<ConvertFp32ToBf16QueueDescriptor>(descriptor, info)
{
    this->m_Data.ValidateInputsOutputs("NeonConvertFp32ToBf16Workload", 1, 1);
    GatherTensorHandlePairs(descriptor, m_TensorHandlePairs);
}

Member Function Documentation

Execute()

void Execute ( ) const

overridevirtual

Implements IWorkload.

Definition at line 26 of file NeonConvertFp32ToBf16Workload.cpp.

References ARMNN_SCOPED_PROFILING_EVENT_NEON, FloatingPointConverter::ConvertFloat32ToBFloat16(), and armnn::CopyTensorContentsGeneric().

{
    ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonConvertFp32ToBf16Workload_Execute");

    auto convertFunc = [](uint8_t* dst, const uint8_t* src, size_t size)
        {
            auto input = reinterpret_cast<const float*>(src);
            auto output = reinterpret_cast<BFloat16*>(dst);
            size_t numElements = size/2; // 2 bytes per bf16
            armnnUtils::FloatingPointConverter::ConvertFloat32ToBFloat16(input, numElements, output);
        };

    for (const auto& pair : m_TensorHandlePairs)
    {
        CopyTensorContentsGeneric(pair.first, pair.second, convertFunc);
    }
}

---

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

• src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp
• src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.cpp