14 #include <arm_compute/runtime/NEON/functions/NEConv3D.h> 22 using namespace armcomputetensorutils;
29 bool isFastMathEnabled,
32 const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input, descriptor.
m_DataLayout);
33 const arm_compute::TensorInfo aclWeightsInfo = BuildArmComputeTensorInfo(weights, descriptor.
m_DataLayout);
34 arm_compute::TensorInfo aclBiasesInfo;
35 arm_compute::TensorInfo *optionalAclBiasesInfo =
nullptr;
40 aclBiasesInfo = BuildArmComputeTensorInfo(biases.
value(), descriptor.
m_DataLayout);
41 optionalAclBiasesInfo = &aclBiasesInfo;
43 const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output, descriptor.
m_DataLayout);
47 activationDescriptor);
49 return arm_compute::NEConv3D::validate(&aclInputInfo,
51 optionalAclBiasesInfo,
58 std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager,
59 const bool isFastMathEnabled)
64 using arm_compute::NEConv3D;
68 arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(
m_Data.
m_Inputs[0])->GetTensor();
69 arm_compute::ITensor& weights = PolymorphicDowncast<IAclTensorHandle*>(
m_Data.
m_Inputs[1])->GetTensor();
70 arm_compute::ITensor* biasesPtr =
nullptr;
73 biasesPtr = &PolymorphicDowncast<IAclTensorHandle *>(
m_Data.
m_Inputs[2])->GetTensor();
75 arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(
m_Data.
m_Outputs[0])->GetTensor();
78 input.info()->set_data_layout(aclDataLayout);
79 weights.info()->set_data_layout(aclDataLayout);
80 output.info()->set_data_layout(aclDataLayout);
82 const arm_compute::Conv3dInfo aclConv3DInfo =
ComputeConv3DInfo(descriptor, isFastMathEnabled);
84 auto convolutionLayer = std::make_unique<arm_compute::NEConv3D>();
85 convolutionLayer->configure(&input,
103 m_ConvolutionLayer.reset(convolutionLayer.release());
107 m_ConvolutionLayer->prepare();
113 m_ConvolutionLayer->run();
NeonConvolution3dWorkload(const Convolution3dQueueDescriptor &descriptor, const WorkloadInfo &info, std::shared_ptr< arm_compute::MemoryManagerOnDemand > &memoryManager, const bool isFastMathENabled=false)
arm_compute::Conv3dInfo ComputeConv3DInfo(const armnn::Convolution3dDescriptor descriptor, bool isFastMathEnabled, const ActivationDescriptor *activationDescriptor)
Utility function used to setup an arm_compute::Conv3dInfo object from convolution3d descriptor...
void Execute() const override
bool m_BiasEnabled
Enable/disable bias.
void ValidateInputsOutputs(const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const
Copyright (c) 2021 ARM Limited and Contributors.
void IgnoreUnused(Ts &&...)
LayerDescriptor m_Parameters
std::vector< TensorInfo > m_InputTensorInfos
Convolution3dQueueDescriptor m_Data
bool has_value() const noexcept
#define ARMNN_ASSERT(COND)
std::vector< TensorInfo > m_OutputTensorInfos
An ActivationDescriptor for the ActivationLayer.
A Convolution3dDescriptor for the Convolution3dLayer.
profiling::ProfilingGuid GetGuid() const final
arm_compute::Status NeonConvolution3dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const Convolution3dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, bool isFastMathEnabled, const ActivationDescriptor *activationDescriptor)
DataLayout m_DataLayout
The data layout to be used (NDHWC, NCDHW).
std::vector< ITensorHandle * > m_Outputs
#define ARMNN_REPORT_PROFILING_WORKLOAD_DESC(name, desc, infos, guid)
Contains information about TensorInfos of a layer.
std::vector< ITensorHandle * > m_Inputs
#define ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID(name, guid)