29 auto convertFunc = [](uint8_t* dst,
const uint8_t* src,
size_t size)
31 auto input =
reinterpret_cast<const BFloat16*
>(src);
32 auto output =
reinterpret_cast<float*
>(dst);
33 size_t numElements = size/2;
37 for (
const auto& pair : m_TensorHandlePairs)
76 void NeonConvertBf16ToFp32Workload::Reconfigure()
virtual void Execute() const override
static void ConvertBFloat16ToFloat32(const void *srcBFloat16Buffer, size_t numElements, float *dstFloat32Buffer)
NeonConvertBf16ToFp32Workload(const ConvertBf16ToFp32QueueDescriptor &descriptor, const WorkloadInfo &info)
arm::pipe::ProfilingGuid GetGuid() const final
void ValidateInputsOutputs(const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const
Copyright (c) 2021 ARM Limited and Contributors.
void ReplaceOutputTensorHandle(ITensorHandle *tensorHandle, unsigned int slot) override
void ReplaceInputTensorHandle(ITensorHandle *tensorHandle, unsigned int slot) override
Contains information about TensorInfos of a layer.
void CopyTensorContentsGeneric(const ITensorHandle *srcTensor, ITensorHandle *dstTensor, CopyFunc copy)
std::vector< ITensorHandle * > m_Inputs
void GatherTensorHandlePairs(const DescriptorType &descriptor, std::vector< std::pair< SrcTensorHandleType *, DstTensorHandleType *>> &tensorHandlePairs)
#define ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID(name, guid)