13 #include <arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h> 17 using namespace armcomputetensorutils;
25 const arm_compute::TensorInfo aclInput = BuildArmComputeTensorInfo(input);
26 const arm_compute::TensorInfo aclOutput = BuildArmComputeTensorInfo(output);
27 const arm_compute::TensorInfo aclWeights = BuildArmComputeTensorInfo(weights);
29 arm_compute::TensorInfo aclBiases;
30 arm_compute::TensorInfo *optionalAclBiases =
nullptr;
33 aclBiases = BuildArmComputeTensorInfo(biases);
34 optionalAclBiases = &aclBiases;
37 const arm_compute::FullyConnectedLayerInfo fullyConnectedLayerInfo =
41 return arm_compute::NEFullyConnectedLayer::validate(&aclInput,
45 fullyConnectedLayerInfo);
49 const WorkloadInfo&
info, std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager)
54 arm_compute::ITensor& input = boost::polymorphic_downcast<IAclTensorHandle*>(
m_Data.
m_Inputs[0])->GetTensor();
55 arm_compute::ITensor& output = boost::polymorphic_downcast<IAclTensorHandle*>(
m_Data.
m_Outputs[0])->GetTensor();
57 m_WeightsTensor = std::make_unique<arm_compute::Tensor>();
62 m_BiasesTensor = std::make_unique<arm_compute::Tensor>();
67 arm_compute::FullyConnectedLayerInfo fc_info;
70 auto layer = std::make_unique<arm_compute::NEFullyConnectedLayer>(memoryManager);
71 layer->configure(&input, m_WeightsTensor.get(), m_BiasesTensor.get(), &output, fc_info);
72 m_FullyConnectedLayer.reset(layer.release());
98 m_FullyConnectedLayer->prepare();
105 m_FullyConnectedLayer->run();
108 void NeonFullyConnectedWorkload::FreeUnusedTensors()
110 FreeTensorIfUnused(m_WeightsTensor);
111 FreeTensorIfUnused(m_BiasesTensor);
const TensorInfo & GetTensorInfo() const
arm_compute::Status NeonFullyConnectedWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TensorInfo &weights, const TensorInfo &biases, const FullyConnectedDescriptor &descriptor)
#define ARMNN_SCOPED_PROFILING_EVENT_NEON(name)
const ConstCpuTensorHandle * m_Weight
arm_compute::FullyConnectedLayerInfo ConvertFullyConnectedDescriptorToAclFullyConnectedLayerInfo(const FullyConnectedDescriptor &fullyConnectedDesc)
LayerDescriptor m_Parameters
const ConstCpuTensorHandle * m_Bias
bool m_BiasEnabled
Enable/disable bias.
virtual void Execute() const override
const FullyConnectedQueueDescriptor m_Data
A FullyConnectedDescriptor for the FullyConnectedLayer.
bool m_TransposeWeightMatrix
Enable/disable transpose weight matrix.
void ValidateInputsOutputs(const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const
DataType GetDataType() const
std::vector< ITensorHandle * > m_Outputs
std::vector< ITensorHandle * > m_Inputs
void InitializeArmComputeTensorData(arm_compute::Tensor &tensor, const ConstCpuTensorHandle *handle)
NeonFullyConnectedWorkload(const FullyConnectedQueueDescriptor &descriptor, const WorkloadInfo &info, std::shared_ptr< arm_compute::MemoryManagerOnDemand > &memoryManager)