BFloat16ToFloat32 Struct Reference

#include <ConvertConstants.hpp>

Static Public Member Functions
static void	Func (std::unique_ptr< ScopedCpuTensorHandle > &handle)

Detailed Description

Definition at line 24 of file ConvertConstants.hpp.

Member Function Documentation

Func()

static void Func ( std::unique_ptr< ScopedCpuTensorHandle > &  handle )

inlinestatic

Definition at line 26 of file ConvertConstants.hpp.

References armnn::BFloat16, FloatingPointConverter::ConvertBFloat16ToFloat32(), armnn::Float32, TensorInfo::GetDataType(), TensorInfo::GetNumElements(), TensorInfo::GetShape(), and armnn::info.

    {
        const TensorInfo& info = handle->GetTensorInfo();

        if (info.GetDataType() == DataType::BFloat16)
        {
            std::vector<float> newValues(info.GetNumElements());

            armnnUtils::FloatingPointConverter::ConvertBFloat16ToFloat32(handle->GetTensor<BFloat16>(),
                                                                         info.GetNumElements(),
                                                                         newValues.data());

            TensorInfo newInfo(info.GetShape(), DataType::Float32);
            ConstTensor newInput(newInfo, newValues);
            handle.reset(new ScopedCpuTensorHandle(newInput));
        }
    }

---

The documentation for this struct was generated from the following file:

• src/armnn/optimizations/ConvertConstants.hpp