ArmNN
 25.11
Loading...
Searching...
No Matches
ClGatherNdWorkload Class Reference

#include <ClGatherNdWorkload.hpp>

Inheritance diagram for ClGatherNdWorkload:
[legend]
Collaboration diagram for ClGatherNdWorkload:
[legend]

Public Member Functions

 ClGatherNdWorkload (const GatherNdQueueDescriptor &descriptor, const WorkloadInfo &info, const arm_compute::CLCompileContext &clCompileContext)
virtual void Execute () const override
Public Member Functions inherited from ClBaseWorkload< GatherNdQueueDescriptor >
 ClBaseWorkload (const GatherNdQueueDescriptor &descriptor, const WorkloadInfo &info)
void ReplaceInputTensorHandle (ITensorHandle *tensorHandle, unsigned int slot) override
void ReplaceOutputTensorHandle (ITensorHandle *tensorHandle, unsigned int slot) override
Public Member Functions inherited from BaseWorkload< GatherNdQueueDescriptor >
 BaseWorkload (const GatherNdQueueDescriptor &descriptor, const WorkloadInfo &info)
virtual const std::string & GetName () const override
void PostAllocationConfigure () override
const GatherNdQueueDescriptor & GetData () const
arm::pipe::ProfilingGuid GetGuid () const final
virtual bool SupportsTensorHandleReplacement () const override
Public Member Functions inherited from IWorkload
virtual ~IWorkload ()
virtual void RegisterDebugCallback (const DebugCallbackFunction &)
virtual armnn::Optional< armnn::MemoryRequirements > GetMemoryRequirements ()

Additional Inherited Members

Protected Member Functions inherited from ClBaseWorkload< GatherNdQueueDescriptor >
virtual void Reconfigure ()
Protected Attributes inherited from BaseWorkload< GatherNdQueueDescriptor >
GatherNdQueueDescriptor m_Data
const arm::pipe::ProfilingGuid m_Guid
const std::string m_Name

Detailed Description

Definition at line 22 of file ClGatherNdWorkload.hpp.

Constructor & Destructor Documentation

◆ ClGatherNdWorkload()

ClGatherNdWorkload ( const GatherNdQueueDescriptor & descriptor,
const WorkloadInfo & info,
const arm_compute::CLCompileContext & clCompileContext )

Calculate flattened indices: m_FlattenedIndices = indices * m_FlattenedCoeff. This could be done using MatMul instead of multiplication followed by reduce sum operation, but GeMM does not support s32 at the moment.

Call Gather with adequate shapes

Definition at line 109 of file ClGatherNdWorkload.cpp.

112 : ClBaseWorkload<GatherNdQueueDescriptor>(descriptor, info)
113{
114 m_Data.ValidateInputsOutputs("ClGatherNdWorkload", 2, 1);
115
116 TensorInfo paramsInfo = info.m_InputTensorInfos[0];
117 TensorInfo indicesInfo = info.m_InputTensorInfos[1];
118 TensorInfo outputInfo = info.m_OutputTensorInfos[0];
119
120 arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
121 arm_compute::ICLTensor& indices = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
122 arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
123
124 // Calculate ND, K, W, C.
125 std::map<std::string, unsigned int> keyIndices = CalculateGatherNdKeyIndices(paramsInfo, indicesInfo);
126
127 /// Calculate flattened indices: m_FlattenedIndices = indices * m_FlattenedCoeff.
128 /// This could be done using MatMul instead of multiplication followed by reduce sum operation,
129 /// but GeMM does not support s32 at the moment.
130
131 // Prepare the tensor to store the output of the reduce_sum operation
132 armnn::TensorInfo flattenedIndices_Info = indicesInfo;
133 flattenedIndices_Info.SetShape({ keyIndices["W"] });
134 BuildArmComputeTensor(m_FlattenedIndices, flattenedIndices_Info);
135 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_FlattenedIndices);
136
137 // Reshape indices into { W, ND }
138 armnn::TensorInfo indicesInfoReshape = indicesInfo;
139 indicesInfoReshape.SetShape({ keyIndices["W"], keyIndices["ND"] });
140 BuildArmComputeTensor(m_IndicesReshaped, indicesInfoReshape);
141 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_IndicesReshaped);
142
143 // Calculate the m_FlattenedCoeff
144 TensorShape paramsShape = paramsInfo.GetShape();
145 std::vector<int32_t> flattenedCoeff(keyIndices["ND"], 1);
146 for (unsigned int i = 1; i < keyIndices["ND"]; ++i)
147 {
148 flattenedCoeff[i - 1] = static_cast<int32_t>(paramsShape[i]);
149 }
150 for (unsigned int i = keyIndices["ND"] - 1; i > 0; --i)
151 {
152 flattenedCoeff[i - 1] *= flattenedCoeff[i];
153 }
154 armnn::TensorInfo flattenedCoeff_Info = indicesInfo;
155 flattenedCoeff_Info.SetShape({ keyIndices["ND"] });
156 BuildArmComputeTensor(m_FlattenedCoeff, flattenedCoeff_Info);
157 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_FlattenedCoeff);
158 CopyArmComputeClTensorData<int32_t>(m_FlattenedCoeff, flattenedCoeff.data());
159
160 // Prepare the tensor to store the output of the multiplication
161 armnn::TensorInfo outputMul_Info = indicesInfo;
162 outputMul_Info.SetShape({ keyIndices["W"], keyIndices["ND"] });
163 BuildArmComputeTensor(m_OutputMul, outputMul_Info);
164 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_OutputMul);
165
166 // Reshape indices to the mul layer input shape
167 m_ReshapeIndicesLayer.configure(&indices, &m_IndicesReshaped);
168
169 // Multiply
170 m_MulLayer.configure(clCompileContext,
171 &m_IndicesReshaped,
172 &m_FlattenedCoeff,
173 &m_OutputMul,
174 1.0f,
175 arm_compute::ConvertPolicy::WRAP,
176 arm_compute::RoundingPolicy::TO_ZERO,
177 arm_compute::ActivationLayerInfo());
178
179 // Reduce Sum
180 const std::vector<unsigned int> armnnReduceAxes(1, 1);
181 arm_compute::Coordinates coords = BuildArmComputeReductionCoordinates(m_OutputMul.info()->num_dimensions(),
182 outputMul_Info.GetNumDimensions(),
183 armnnReduceAxes);
184 m_ReduceSumLayer.configure(clCompileContext,
185 &m_OutputMul,
186 &m_FlattenedIndices,
187 static_cast<unsigned int>(coords[0]),
188 arm_compute::ReductionOperation::SUM,
189 false);
190
191 /// Call Gather with adequate shapes
192 // Reshape params into { K, C }
193 armnn::TensorInfo paramsInfoReshape = paramsInfo;
194 paramsInfoReshape.SetShape({ keyIndices["K"], keyIndices["C"] });
195 BuildArmComputeTensor(m_InputGather, paramsInfoReshape);
196 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_InputGather);
197
198 // Reshape input to the gather params input shape
199 m_ReshapeInputLayer.configure(&input, &m_InputGather);
200
201 // Reshape output to have the shape given by gather { W, C }
202 // (the original outputInfo has the shape given by gatherNd)
203 armnn::TensorInfo outputGather_Info = outputInfo;
204 outputGather_Info.SetShape({ keyIndices["W"], keyIndices["C"] });
205 BuildArmComputeTensor(m_OutputGather, outputGather_Info);
206 armcomputetensorutils::InitialiseArmComputeTensorEmpty(m_OutputGather);
207 {
208 ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID("ClGatherNdWorkload_configure");
209 auto aclAxis = ComputeAclAxis(0, paramsInfoReshape);
210 m_GatherLayer.configure(clCompileContext, &m_InputGather, &m_FlattenedIndices, &m_OutputGather, aclAxis);
211 }
212
213 // Reshape output to the original output shape
214 m_ReshapeOutputLayer.configure(clCompileContext, &m_OutputGather, &output);
215};
ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID
#define ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID(label)
Creates a profiling event that uses GetGuid() and GetName() from the calling class.
Definition ClWorkloadUtils.hpp:36
armnn::TensorInfo::GetNumDimensions
unsigned int GetNumDimensions() const
Definition Tensor.hpp:197
armnn::TensorInfo::SetShape
void SetShape(const TensorShape &newShape)
Definition Tensor.hpp:195
armnn::ComputeAclAxis
int ComputeAclAxis(const int &armnnAxis, const armnn::TensorInfo &tensor)
Function to convert ArmNN axis (left to right) to ACL axis (right to left) ranging from [-rank,...
Definition ArmComputeUtils.hpp:246
armnn::CalculateGatherNdKeyIndices
std::map< std::string, unsigned int > CalculateGatherNdKeyIndices(TensorInfo inputInfo0, TensorInfo inputInfo1)
Calculates the key index values needed for GatherNd: N, ND, K, W, C (N is always 1)
Definition WorkloadUtils.cpp:313

References ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID, armnn::CalculateGatherNdKeyIndices(), ClBaseWorkload< GatherNdQueueDescriptor >::ClBaseWorkload(), armnn::ComputeAclAxis(), armnn::CopyArmComputeClTensorData(), TensorInfo::GetNumDimensions(), TensorInfo::GetShape(), IClTensorHandle::GetTensor(), armnn::info, BaseWorkload< GatherNdQueueDescriptor >::m_Data, and TensorInfo::SetShape().

Member Function Documentation

◆ Execute()

void Execute ( ) const
overridevirtual

Implements IWorkload.

Definition at line 217 of file ClGatherNdWorkload.cpp.

218{
219 ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID("ClGatherNdWorkload_Execute");
220 RunClFunction(m_ReshapeInputLayer, CHECK_LOCATION());
221 RunClFunction(m_ReshapeIndicesLayer, CHECK_LOCATION());
222 RunClFunction(m_MulLayer, CHECK_LOCATION());
223 RunClFunction(m_ReduceSumLayer, CHECK_LOCATION());
224 RunClFunction(m_GatherLayer, CHECK_LOCATION());
225 RunClFunction(m_ReshapeOutputLayer, CHECK_LOCATION());
226}
CHECK_LOCATION
#define CHECK_LOCATION()
Definition Exceptions.hpp:203
armnn::RunClFunction
void RunClFunction(arm_compute::IFunction &function, const CheckLocation &location)
Definition ClWorkloadUtils.hpp:167

References ARMNN_SCOPED_PROFILING_EVENT_CL_NAME_GUID, CHECK_LOCATION, and armnn::RunClFunction().

The documentation for this class was generated from the following files: