From 7bfd38a721360183f3392f9ab35db18a0dd7fef8 Mon Sep 17 00:00:00 2001 From: Nikhil Raj Date: Fri, 19 Aug 2022 15:23:36 +0100 Subject: Update Doxygen for 22.08 Release Signed-off-by: Nikhil Raj Change-Id: I4789fe868e0492839be1482e5cee3642ed90d756 --- .../_neon_convolution3d_workload_8cpp_source.xhtml | 152 +++++++++++++++++++++ 1 file changed, 152 insertions(+) create mode 100644 22.08/_neon_convolution3d_workload_8cpp_source.xhtml (limited to '22.08/_neon_convolution3d_workload_8cpp_source.xhtml') diff --git a/22.08/_neon_convolution3d_workload_8cpp_source.xhtml b/22.08/_neon_convolution3d_workload_8cpp_source.xhtml new file mode 100644 index 0000000000..d131780cbb --- /dev/null +++ b/22.08/_neon_convolution3d_workload_8cpp_source.xhtml @@ -0,0 +1,152 @@ + + + + + + + + + + + + + +ArmNN: src/backends/neon/workloads/NeonConvolution3dWorkload.cpp Source File + + + + + + + + + + + + + + + + +
+
+ + + + ArmNN + + + +
+
+  22.08 +
+
+
+ + + + + + + +
+
+ +
+
+
+ +
+ +
+
+ + +
+ +
+ +
+
+
NeonConvolution3dWorkload.cpp
+
+
+Go to the documentation of this file.
1 //
2 // Copyright © 2021 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 
6 #include "NeonConvolution3dWorkload.hpp"
7 
8 #include <aclCommon/ArmComputeTensorUtils.hpp>
9 #include <aclCommon/ArmComputeUtils.hpp>
10 #include <armnn/utility/PolymorphicDowncast.hpp>
11 #include <armnn/backends/TensorHandle.hpp>
12 #include <neon/workloads/NeonWorkloadUtils.hpp>
13 
14 #include <arm_compute/runtime/NEON/functions/NEConv3D.h>
15 
16 #include <armnn/Types.hpp>
17 #include <Half.hpp>
18 
19 namespace armnn
20 {
21 
22 using namespace armcomputetensorutils;
23 
24 arm_compute::Status NeonConvolution3dWorkloadValidate(const TensorInfo& input,
25  const TensorInfo& output,
26  const Convolution3dDescriptor& descriptor,
27  const TensorInfo& weights,
28  const Optional<TensorInfo>& biases,
29  bool isFastMathEnabled,
30  const ActivationDescriptor* activationDescriptor)
31 {
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;
36  if (descriptor.m_BiasEnabled)
37  {
38  ARMNN_ASSERT(biases.has_value());
39 
40  aclBiasesInfo = BuildArmComputeTensorInfo(biases.value(), descriptor.m_DataLayout);
41  optionalAclBiasesInfo = &aclBiasesInfo;
42  }
43  const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output, descriptor.m_DataLayout);
44 
45  const arm_compute::Conv3dInfo aclConv3DInfo = ComputeConv3DInfo(descriptor,
46  isFastMathEnabled,
47  activationDescriptor);
48 
49  return arm_compute::NEConv3D::validate(&aclInputInfo,
50  &aclWeightsInfo,
51  optionalAclBiasesInfo,
52  &aclOutputInfo,
53  aclConv3DInfo);
54 }
55 
56 NeonConvolution3dWorkload::NeonConvolution3dWorkload(const Convolution3dQueueDescriptor& descriptor,
57  const WorkloadInfo& info,
58  std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager,
59  const bool isFastMathEnabled)
60  : NeonBaseWorkload<Convolution3dQueueDescriptor>(descriptor, info)
61 {
62  IgnoreUnused(memoryManager);
63 
64  using arm_compute::NEConv3D;
65  uint32_t numInputs = m_Data.m_Parameters.m_BiasEnabled ? 3: 2;
66  m_Data.ValidateInputsOutputs("NeonConvolution3dWorkload", numInputs, 1);
67 
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;
71  if (m_Data.m_Parameters.m_BiasEnabled)
72  {
73  biasesPtr = &PolymorphicDowncast<IAclTensorHandle *>(m_Data.m_Inputs[2])->GetTensor();
74  }
75  arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
76 
77  arm_compute::DataLayout aclDataLayout = ConvertDataLayout(m_Data.m_Parameters.m_DataLayout);
78  input.info()->set_data_layout(aclDataLayout);
79  weights.info()->set_data_layout(aclDataLayout);
80  output.info()->set_data_layout(aclDataLayout);
81 
82  const arm_compute::Conv3dInfo aclConv3DInfo = ComputeConv3DInfo(descriptor, isFastMathEnabled);
83 
84  auto convolutionLayer = std::make_unique<arm_compute::NEConv3D>();
85  convolutionLayer->configure(&input,
86  &weights,
87  biasesPtr,
88  &output,
89  aclConv3DInfo);
90 
91  // Add details for profiling output
92  WorkloadInfo detailsInfo;
93 
94  detailsInfo.m_InputTensorInfos = info.m_InputTensorInfos;
95  detailsInfo.m_OutputTensorInfos = info.m_OutputTensorInfos;
96 
97  // Report Profiling Details
98  ARMNN_REPORT_PROFILING_WORKLOAD_DESC("NeonConvolution3dWorkload_Construct",
99  descriptor.m_Parameters,
100  detailsInfo,
101  this->GetGuid());
102 
103  m_ConvolutionLayer.reset(convolutionLayer.release());
104 
105  ARMNN_ASSERT(m_ConvolutionLayer);
106 
107  m_ConvolutionLayer->prepare();
108 }
109 
110 void NeonConvolution3dWorkload::Execute() const
111 {
112  ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID("NeonConvolution3dWorkload_Execute", this->GetGuid());
113  m_ConvolutionLayer->run();
114 }
115 
116 } //namespace armnn
armnn::NeonConvolution3dWorkload::NeonConvolution3dWorkload
NeonConvolution3dWorkload(const Convolution3dQueueDescriptor &descriptor, const WorkloadInfo &info, std::shared_ptr< arm_compute::MemoryManagerOnDemand > &memoryManager, const bool isFastMathENabled=false)
Definition: NeonConvolution3dWorkload.cpp:56
+
armnn::ComputeConv3DInfo
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...
Definition: ArmComputeUtils.hpp:293
+
armnn::DataLayout
DataLayout
Definition: Types.hpp:62
+ +
ArmComputeUtils.hpp
+ + + +
armnn::BaseWorkload< Convolution3dQueueDescriptor >::GetGuid
arm::pipe::ProfilingGuid GetGuid() const final
Definition: Workload.hpp:61
+
ArmComputeTensorUtils.hpp
+
armnn::Convolution3dDescriptor::m_BiasEnabled
bool m_BiasEnabled
Enable/disable bias.
Definition: Descriptors.hpp:614
+
armnn::QueueDescriptor::ValidateInputsOutputs
void ValidateInputsOutputs(const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const
Definition: WorkloadData.cpp:474
+
armnn
Copyright (c) 2021 ARM Limited and Contributors.
Definition: 01_00_quick_start.dox:6
+
armnn::IgnoreUnused
void IgnoreUnused(Ts &&...)
Definition: IgnoreUnused.hpp:14
+ + +
PolymorphicDowncast.hpp
+ +
armnn::WorkloadInfo::m_InputTensorInfos
std::vector< TensorInfo > m_InputTensorInfos
Definition: WorkloadInfo.hpp:18
+ +
armnn::OptionalBase::has_value
bool has_value() const noexcept
Definition: Optional.hpp:53
+
Types.hpp
+
armnn::Status
Status
enumeration
Definition: Types.hpp:42
+
NeonWorkloadUtils.hpp
+
ARMNN_ASSERT
#define ARMNN_ASSERT(COND)
Definition: Assert.hpp:14
+ +
armnn::WorkloadInfo::m_OutputTensorInfos
std::vector< TensorInfo > m_OutputTensorInfos
Definition: WorkloadInfo.hpp:19
+
armnn::ActivationDescriptor
An ActivationDescriptor for the ActivationLayer.
Definition: Descriptors.hpp:36
+
armnn::Convolution3dDescriptor
A Convolution3dDescriptor for the Convolution3dLayer.
Definition: Descriptors.hpp:549
+
armnn::NeonConvolution3dWorkloadValidate
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)
Definition: NeonConvolution3dWorkload.cpp:24
+
armnn::Convolution3dDescriptor::m_DataLayout
DataLayout m_DataLayout
The data layout to be used (NDHWC, NCDHW).
Definition: Descriptors.hpp:616
+
Half.hpp
+
armnn::QueueDescriptor::m_Outputs
std::vector< ITensorHandle * > m_Outputs
Definition: WorkloadData.hpp:27
+ +
ARMNN_REPORT_PROFILING_WORKLOAD_DESC
#define ARMNN_REPORT_PROFILING_WORKLOAD_DESC(name, desc, infos, guid)
Definition: Profiling.hpp:227
+
armnn::WorkloadInfo
Contains information about TensorInfos of a layer.
Definition: WorkloadInfo.hpp:16
+
armnn::QueueDescriptor::m_Inputs
std::vector< ITensorHandle * > m_Inputs
Definition: WorkloadData.hpp:26
+
ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID
#define ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID(name, guid)
Definition: NeonWorkloadUtils.hpp:24
+
NeonConvolution3dWorkload.hpp
+
+
+ + + + -- cgit v1.2.1