OptimizerOptions Struct Reference

ArmNN performs an optimization on each model/network before it gets loaded for execution. More...

#include <INetwork.hpp>

Public Member Functions
	OptimizerOptions ()
	OptimizerOptions (bool reduceFp32ToFp16, bool debug, bool reduceFp32ToBf16, bool importEnabled, ModelOptions modelOptions={}, bool exportEnabled=false, bool debugToFile=false)
	OptimizerOptions (bool reduceFp32ToFp16, bool debug, bool reduceFp32ToBf16=false, ShapeInferenceMethod shapeInferenceMethod=armnn::ShapeInferenceMethod::ValidateOnly, bool importEnabled=false, ModelOptions modelOptions={}, bool exportEnabled=false, bool debugToFile=false, bool allowExpandedDims=false)
const std::string	ToString () const

Public Attributes
bool	m_ReduceFp32ToFp16
	Reduces all Fp32 operators in the model to Fp16 for faster processing. More...
bool	m_Debug
	Add debug data for easier troubleshooting. More...
bool	m_DebugToFile
	Pass debug data to separate output files for easier troubleshooting. More...
bool	m_ReduceFp32ToBf16
	@Note This feature has been replaced by enabling Fast Math in compute library backend options. More...
ShapeInferenceMethod	m_shapeInferenceMethod
	Infer output size when not available. More...
bool	m_ImportEnabled
	Enable Import. More...
ModelOptions	m_ModelOptions
	Enable Model Options. More...
bool	m_ProfilingEnabled
	Enable profiling dump of the optimizer phase. More...
bool	m_ExportEnabled
	Enable Export. More...
bool	m_AllowExpandedDims
	When calculating tensor sizes, dimensions of size == 1 will be ignored. More...

Detailed Description

ArmNN performs an optimization on each model/network before it gets loaded for execution.

OptimizerOptions provides a set of features that allows the user to customize this optimization on a per model basis.

Examples

CustomMemoryAllocatorSample.cpp.

Definition at line 137 of file INetwork.hpp.

Constructor & Destructor Documentation

OptimizerOptions() [1/3]

OptimizerOptions ( )

inline

Definition at line 139 of file INetwork.hpp.

        : m_ReduceFp32ToFp16(false)
        , m_Debug(false)
        , m_DebugToFile(false)
        , m_ReduceFp32ToBf16(false)
        , m_shapeInferenceMethod(armnn::ShapeInferenceMethod::ValidateOnly)
        , m_ImportEnabled(false)
        , m_ModelOptions()
        , m_ProfilingEnabled(false)
        , m_ExportEnabled(false)
        , m_AllowExpandedDims(false)
    {}

References armnn::ValidateOnly.

OptimizerOptions() [2/3]

OptimizerOptions ( bool  reduceFp32ToFp16, bool  debug, bool  reduceFp32ToBf16, bool  importEnabled, ModelOptions  modelOptions = {}, bool  exportEnabled = false, bool  debugToFile = false  )

inline

Definition at line 152 of file INetwork.hpp.

                                                 {}, bool exportEnabled = false, bool debugToFile = false)
        : m_ReduceFp32ToFp16(reduceFp32ToFp16)
        , m_Debug(debug)
        , m_DebugToFile(debugToFile)
        , m_ReduceFp32ToBf16(reduceFp32ToBf16)
        , m_shapeInferenceMethod(armnn::ShapeInferenceMethod::ValidateOnly)
        , m_ImportEnabled(importEnabled)
        , m_ModelOptions(modelOptions)
        , m_ProfilingEnabled(false)
        , m_ExportEnabled(exportEnabled)
        , m_AllowExpandedDims(false)
    {
    }

OptimizerOptions() [3/3]

OptimizerOptions ( bool  reduceFp32ToFp16, bool  debug, bool  reduceFp32ToBf16 = false, ShapeInferenceMethod  shapeInferenceMethod = armnn::ShapeInferenceMethod::ValidateOnly, bool  importEnabled = false, ModelOptions  modelOptions = {}, bool  exportEnabled = false, bool  debugToFile = false, bool  allowExpandedDims = false  )

inline

Definition at line 167 of file INetwork.hpp.

                                                                             {}, bool exportEnabled = false,
                     bool debugToFile = false, bool allowExpandedDims = false)
        : m_ReduceFp32ToFp16(reduceFp32ToFp16)
        , m_Debug(debug)
        , m_DebugToFile(debugToFile)
        , m_ReduceFp32ToBf16(reduceFp32ToBf16)
        , m_shapeInferenceMethod(shapeInferenceMethod)
        , m_ImportEnabled(importEnabled)
        , m_ModelOptions(modelOptions)
        , m_ProfilingEnabled(false)
        , m_ExportEnabled(exportEnabled)
        , m_AllowExpandedDims(allowExpandedDims)
    {
    }

Member Function Documentation

ToString()

const std::string ToString ( ) const

inline

Definition at line 184 of file INetwork.hpp.

    {
        std::stringstream stream;
        stream << "OptimizerOptions: \n";
        stream << "\tReduceFp32ToFp16: " << m_ReduceFp32ToFp16 << "\n";
        stream << "\tReduceFp32ToBf16: " << m_ReduceFp32ToBf16 << "\n";
        stream << "\tDebug: " << m_Debug << "\n";
        stream << "\tDebug to file: " << m_DebugToFile << "\n";
        stream << "\tShapeInferenceMethod: " <<
        (m_shapeInferenceMethod == ShapeInferenceMethod::ValidateOnly ? "ValidateOnly" : "InferAndValidate") << "\n";
        stream << "\tImportEnabled: " << m_ImportEnabled << "\n";
        stream << "\tExportEnabled: " << m_ExportEnabled << "\n";
        stream << "\tProfilingEnabled: " << m_ProfilingEnabled << "\n";
        stream << "\tAllowExpandedDims: " << m_AllowExpandedDims << "\n";
 
        stream << "\tModelOptions: \n";
        for (auto optionsGroup : m_ModelOptions)
        {
            for (size_t i=0; i < optionsGroup.GetOptionCount(); i++)
            {
                const armnn::BackendOptions::BackendOption option = optionsGroup.GetOption(i);
                stream << "\t\tBackend: "  << optionsGroup.GetBackendId() << "\n"
                       << "\t\t\tOption: " << option.GetName() << "\n"
                       << "\t\t\tValue: "  << std::string(option.GetValue().ToString()) << "\n";
            }
        }
 
        return stream.str();
    }

References BackendOptions::BackendOption::GetName(), BackendOptions::BackendOption::GetValue(), OptimizerOptions::m_AllowExpandedDims, OptimizerOptions::m_Debug, OptimizerOptions::m_DebugToFile, OptimizerOptions::m_ExportEnabled, OptimizerOptions::m_ImportEnabled, OptimizerOptions::m_ModelOptions, OptimizerOptions::m_ProfilingEnabled, OptimizerOptions::m_ReduceFp32ToBf16, OptimizerOptions::m_ReduceFp32ToFp16, OptimizerOptions::m_shapeInferenceMethod, BackendOptions::Var::ToString(), and armnn::ValidateOnly.

Referenced by armnn::Optimize().

Member Data Documentation

m_AllowExpandedDims

bool m_AllowExpandedDims

When calculating tensor sizes, dimensions of size == 1 will be ignored.

Definition at line 247 of file INetwork.hpp.

Referenced by OptimizerOptions::ToString().

m_Debug

bool m_Debug

Add debug data for easier troubleshooting.

Definition at line 222 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

m_DebugToFile

bool m_DebugToFile

Pass debug data to separate output files for easier troubleshooting.

Definition at line 225 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

m_ExportEnabled

bool m_ExportEnabled

Enable Export.

Definition at line 244 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

m_ImportEnabled

bool m_ImportEnabled

Enable Import.

Examples

CustomMemoryAllocatorSample.cpp.

Definition at line 235 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

m_ModelOptions

ModelOptions m_ModelOptions

Enable Model Options.

Definition at line 238 of file INetwork.hpp.

Referenced by armnn::Optimize(), ArmnnDriverImpl::PrepareArmnnModel(), ArmnnDriverImpl::PrepareArmnnModelFromCache(), and OptimizerOptions::ToString().

m_ProfilingEnabled

bool m_ProfilingEnabled

Enable profiling dump of the optimizer phase.

Definition at line 241 of file INetwork.hpp.

Referenced by armnn::Optimize(), ArmnnDriverImpl::PrepareArmnnModel(), ArmnnDriverImpl::PrepareArmnnModelFromCache(), and OptimizerOptions::ToString().

m_ReduceFp32ToBf16

bool m_ReduceFp32ToBf16

@Note This feature has been replaced by enabling Fast Math in compute library backend options.

This is currently a placeholder option

Definition at line 229 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

m_ReduceFp32ToFp16

bool m_ReduceFp32ToFp16

Reduces all Fp32 operators in the model to Fp16 for faster processing.

@Note This feature works best if all operators of the model are in Fp32. ArmNN will add conversion layers between layers that weren't in Fp32 in the first place or if the operator is not supported in Fp16. The overhead of these conversions can lead to a slower overall performance if too many conversions are required.

Definition at line 219 of file INetwork.hpp.

Referenced by armnn::Optimize(), ArmnnDriverImpl::PrepareArmnnModel(), ArmnnDriverImpl::PrepareArmnnModelFromCache(), and OptimizerOptions::ToString().

m_shapeInferenceMethod

ShapeInferenceMethod m_shapeInferenceMethod

Infer output size when not available.

Definition at line 232 of file INetwork.hpp.

Referenced by armnn::Optimize(), and OptimizerOptions::ToString().

---

The documentation for this struct was generated from the following file:

• include/armnn/INetwork.hpp