ArmNN
 24.02
NeonBatchMatMulWorkload Class Reference

#include <NeonBatchMatMulWorkload.hpp>

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

Public Member Functions

 NeonBatchMatMulWorkload (const BatchMatMulQueueDescriptor &descriptor, const WorkloadInfo &info, const bool isFastMathEnabled)
 
virtual void Execute () const override
 
- Public Member Functions inherited from NeonBaseWorkload< BatchMatMulQueueDescriptor >
 NeonBaseWorkload (const BatchMatMulQueueDescriptor &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< BatchMatMulQueueDescriptor >
 BaseWorkload (const BatchMatMulQueueDescriptor &descriptor, const WorkloadInfo &info)
 
virtual const std::string & GetName () const override
 
void ExecuteAsync (ExecutionData &executionData) override
 
void PostAllocationConfigure () override
 
const BatchMatMulQueueDescriptorGetData () const
 
arm::pipe::ProfilingGuid GetGuid () const final
 
virtual bool SupportsTensorHandleReplacement () const override
 
- Public Member Functions inherited from IWorkload
virtual ~IWorkload ()
 
virtual arm::pipe::ProfilingGuid GetGuid () const =0
 
virtual bool SupportsTensorHandleReplacement () const =0
 
virtual const std::string & GetName () const =0
 
virtual void RegisterDebugCallback (const DebugCallbackFunction &)
 
virtual armnn::Optional< armnn::MemoryRequirementsGetMemoryRequirements ()
 

Additional Inherited Members

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

Detailed Description

Definition at line 22 of file NeonBatchMatMulWorkload.hpp.

Constructor & Destructor Documentation

◆ NeonBatchMatMulWorkload()

NeonBatchMatMulWorkload ( const BatchMatMulQueueDescriptor descriptor,
const WorkloadInfo info,
const bool  isFastMathEnabled 
)

Definition at line 57 of file NeonBatchMatMulWorkload.cpp.

60  : NeonBaseWorkload<BatchMatMulQueueDescriptor>(descriptor, info)
61 {
62  if (descriptor.m_Parameters.m_AdjointX || descriptor.m_Parameters.m_AdjointY )
63  {
64  throw Exception("Support for adjoint not implemented.");
65  }
66  if (descriptor.m_Parameters.m_DataLayoutX != armnn::DataLayout::NCHW
67  || descriptor.m_Parameters.m_DataLayoutY != armnn::DataLayout::NCHW )
68  {
69  throw Exception("Only supported the MatMul in the last 2 dimensions");
70  }
71 
72  m_Data.ValidateInputsOutputs("NeonBatchMatMulWorkload", 2, 1);
73 
74  arm_compute::ITensor& inputX = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
75  arm_compute::ITensor& inputY = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
76  arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
77 
78  // GeMM dispatches kernel handles dynamic inputs differently to static so this flag needs to be set
79  inputX.info()->set_are_values_constant(false);
80  inputY.info()->set_are_values_constant(false);
81 
82  const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
83 
84  arm_compute::MatMulInfo matMulInfo;
85  matMulInfo.adj_lhs(descriptor.m_Parameters.m_TransposeX);
86  matMulInfo.adj_rhs(descriptor.m_Parameters.m_TransposeY);
87 
88  arm_compute::CpuMatMulSettings settings;
89  settings.fast_math(isFastMathEnabled);
90 
91  m_MatMulLayer.configure(&inputX, &inputY, &output, matMulInfo, settings, activationInfo);
92 
93  // Report Profiling Details
94  WorkloadInfo detailsInfo;
95  detailsInfo.m_InputTensorInfos = info.m_InputTensorInfos;
96  detailsInfo.m_OutputTensorInfos = info.m_OutputTensorInfos;
97  ARMNN_REPORT_PROFILING_WORKLOAD_DESC("NeonBatchMatMulWorkload_Construct",
98  descriptor.m_Parameters,
99  detailsInfo,
100  GetGuid());
101 }

References ARMNN_REPORT_PROFILING_WORKLOAD_DESC, armnn::ConvertAdditionalInfoToAclActivationLayerInfo(), BaseWorkload< BatchMatMulQueueDescriptor >::GetGuid(), armnn::info, BatchMatMulDescriptor::m_AdjointX, BatchMatMulDescriptor::m_AdjointY, BaseWorkload< BatchMatMulQueueDescriptor >::m_Data, BatchMatMulDescriptor::m_DataLayoutX, BatchMatMulDescriptor::m_DataLayoutY, QueueDescriptor::m_Inputs, WorkloadInfo::m_InputTensorInfos, QueueDescriptor::m_Outputs, WorkloadInfo::m_OutputTensorInfos, QueueDescriptorWithParameters< LayerDescriptor >::m_Parameters, BatchMatMulDescriptor::m_TransposeX, BatchMatMulDescriptor::m_TransposeY, armnn::NCHW, and QueueDescriptor::ValidateInputsOutputs().

Member Function Documentation

◆ Execute()

void Execute ( ) const
overridevirtual

Implements IWorkload.

Definition at line 103 of file NeonBatchMatMulWorkload.cpp.

104 {
105  ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID("NeonBatchMatMulWorkload_Execute");
106  m_MatMulLayer.run();
107 }

References ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID.


The documentation for this class was generated from the following files:
armnn::QueueDescriptor::ValidateInputsOutputs
void ValidateInputsOutputs(const std::string &descName, unsigned int numExpectedIn, unsigned int numExpectedOut) const
Definition: WorkloadData.cpp:446
armnn::ConvertAdditionalInfoToAclActivationLayerInfo
arm_compute::ActivationLayerInfo ConvertAdditionalInfoToAclActivationLayerInfo(const QueueDescriptor &queueDescriptor)
Definition: ArmComputeUtils.hpp:105
armnn::BoostLogSeverityMapping::info
@ info
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::BaseWorkload< BatchMatMulQueueDescriptor >::GetGuid
arm::pipe::ProfilingGuid GetGuid() const final
Definition: Workload.hpp:67
armnn::BaseWorkload< BatchMatMulQueueDescriptor >::m_Data
BatchMatMulQueueDescriptor m_Data
Definition: Workload.hpp:89
ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID
#define ARMNN_SCOPED_PROFILING_EVENT_NEON_NAME_GUID(label)
Creates a profiling event that uses GetGuid() and GetName() from the calling class.
Definition: NeonWorkloadUtils.hpp:32
armnn::QueueDescriptor::m_Inputs
std::vector< ITensorHandle * > m_Inputs
Definition: WorkloadData.hpp:26
armnn::DataLayout::NCHW
@ NCHW