NeonBackendModelContext Class Reference

The NeonBackendModelContext is used to pass in Neon specific backend ModelOptions. More...

#include <NeonBackendModelContext.hpp>

Inheritance diagram for NeonBackendModelContext:

Public Member Functions
	NeonBackendModelContext (const ModelOptions &modelOptions)
bool	IsFastMathEnabled () const
Public Member Functions inherited from IBackendModelContext
virtual	~IBackendModelContext ()

Detailed Description

The NeonBackendModelContext is used to pass in Neon specific backend ModelOptions.

The supported backend ModelOptions are:

• "FastMathEnabled"
  Using the fast_math flag can lead to performance improvements in fp32 and fp16 layers but may result in
  results with reduced or different precision. The fast_math flag will not have any effect on int8 performance.

Definition at line 17 of file NeonBackendModelContext.hpp.

Constructor & Destructor Documentation

NeonBackendModelContext()

NeonBackendModelContext

(

const ModelOptions &

modelOptions

)

Definition at line 25 of file NeonBackendModelContext.cpp.

References armnn::ParseOptions().

    : m_IsFastMathEnabled(false)
{
   if (!modelOptions.empty())
   {
       ParseOptions(modelOptions, "CpuAcc", [&](std::string name, const BackendOptions::Var& value)
       {
           if (name == "FastMathEnabled")
           {
               m_IsFastMathEnabled |= ParseBool(value, false);
           }
       });
   }
}

Member Function Documentation

IsFastMathEnabled()

bool IsFastMathEnabled

(

)

const

Definition at line 40 of file NeonBackendModelContext.cpp.

Referenced by NeonWorkloadFactory::CreateConvolution2d(), and NeonLayerSupport::IsConvolution2dSupported().

{
    return m_IsFastMathEnabled;
}

---

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

• src/backends/neon/NeonBackendModelContext.hpp
• src/backends/neon/NeonBackendModelContext.cpp