SampleMLP Class Reference

The SampleMLP class implements the MNIST API sample. More...

Collaboration diagram for SampleMLP:

Public Member Functions
	SampleMLP (const SampleMLPParams &params)
bool	build ()
	Function builds the network engine. More...
bool	infer ()
	Runs the TensorRT inference engine for this sample. More...
bool	teardown ()
	Cleans up any state created in the sample class. More...

Private Types
template<typename T >
using	SampleUniquePtr = std::unique_ptr< T, samplesCommon::InferDeleter >

Private Member Functions
bool	constructNetwork (SampleUniquePtr< nvinfer1::IBuilder > &builder, SampleUniquePtr< nvinfer1::INetworkDefinition > &network, SampleUniquePtr< nvinfer1::IBuilderConfig > &config)
	Uses the API to create the MLP Network. More...
bool	processInput (const samplesCommon::BufferManager &buffers)
	Reads the input and stores the result in a managed buffer. More...
bool	verifyOutput (const samplesCommon::BufferManager &buffers)
	Classifies digits and verify result. More...
std::map< std::string, std::pair< nvinfer1::Dims, nvinfer1::Weights > >	loadWeights (const std::string &file)
	Loads weights from weights file. More...
nvinfer1::Dims	loadShape (std::ifstream &input)
	Loads shape from weights file. More...
void	transposeWeights (nvinfer1::Weights &wts, int hiddenSize)
	Transpose weights. More...
nvinfer1::ILayer *	addMLPLayer (nvinfer1::INetworkDefinition *network, nvinfer1::ITensor &inputTensor, int32_t hiddenSize, nvinfer1::Weights wts, nvinfer1::Weights bias, nvinfer1::ActivationType actType, int idx)
	Add an MLP layer. More...

Private Attributes
SampleMLPParams	mParams
	The parameters for the sample. More...
int	mNumber {0}
	The number to classify. More...
std::map< std::string, std::pair< nvinfer1::Dims, nvinfer1::Weights > >	mWeightMap
	The weight name to weight value map. More...
std::shared_ptr< nvinfer1::ICudaEngine >	mEngine
	The TensorRT engine used to run the network. More...
std::vector< SampleUniquePtr< nvinfer1::IHostMemory > >	weightsMemory
	Host weights memory holder. More...

Detailed Description

The SampleMLP class implements the MNIST API sample.

It creates the network for MNIST classification using the API

Member Typedef Documentation

SampleUniquePtr

template<typename T >

using SampleMLP::SampleUniquePtr = std::unique_ptr<T, samplesCommon::InferDeleter>

private

Constructor & Destructor Documentation

SampleMLP()

SampleMLP::SampleMLP ( const SampleMLPParams &  params )

inline

Member Function Documentation

build()

bool SampleMLP::build

(

)

Function builds the network engine.

Creates the network, configures the builder and creates the network engine.

This function creates the MLP network by using the API to create a model and builds the engine that will be used to run MNIST (mEngine)

Returns

Returns true if the engine was created successfully and false otherwise

Here is the call graph for this function:

infer()

bool SampleMLP::infer

(

)

Runs the TensorRT inference engine for this sample.

This function is the main execution function of the sample. It allocates the buffer, sets inputs and executes the engine.

Here is the call graph for this function:

teardown()

bool SampleMLP::teardown

(

)

Cleans up any state created in the sample class.

constructNetwork()

bool SampleMLP::constructNetwork ( SampleUniquePtr< nvinfer1::IBuilder > &  builder, SampleUniquePtr< nvinfer1::INetworkDefinition > &  network, SampleUniquePtr< nvinfer1::IBuilderConfig > &  config  )

private

Uses the API to create the MLP Network.

Parameters

network	Pointer to the network that will be populated with the MLP network
builder	Pointer to the engine builder

Here is the call graph for this function:

Here is the caller graph for this function:

processInput()

bool SampleMLP::processInput ( const samplesCommon::BufferManager &  buffers )

private

Reads the input and stores the result in a managed buffer.

Here is the call graph for this function:

Here is the caller graph for this function:

verifyOutput()

bool SampleMLP::verifyOutput ( const samplesCommon::BufferManager &  buffers )

private

Classifies digits and verify result.

Returns

whether the classification output matches expectations

Here is the call graph for this function:

Here is the caller graph for this function:

loadWeights()

std::map< std::string, std::pair< nvinfer1::Dims, nvinfer1::Weights > > SampleMLP::loadWeights ( const std::string &  file )

private

Loads weights from weights file.

Our weight files are in a very simple space delimited format. type is the integer value of the DataType enum in NvInfer.h. <number of buffers> for each buffer: [name] [type] [size] <data x size in hex>

Here is the call graph for this function:

Here is the caller graph for this function:

loadShape()

nvinfer1::Dims SampleMLP::loadShape ( std::ifstream &  input )

private

Loads shape from weights file.

Here is the caller graph for this function:

transposeWeights()

void SampleMLP::transposeWeights ( nvinfer1::Weights &  wts, int  hiddenSize  )

private

Transpose weights.

Here is the caller graph for this function:

addMLPLayer()

nvinfer1::ILayer * SampleMLP::addMLPLayer ( nvinfer1::INetworkDefinition *  network, nvinfer1::ITensor &  inputTensor, int32_t  hiddenSize, nvinfer1::Weights  wts, nvinfer1::Weights  bias, nvinfer1::ActivationType  actType, int  idx  )

private

Add an MLP layer.

The addMLPLayer function is a simple helper function that creates the combination required for an MLP layer. By replacing the implementation of this sequence with various implementations, then then it can be shown how TensorRT optimizations those layer sequences.

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

mParams

SampleMLPParams SampleMLP::mParams

private

The parameters for the sample.

mNumber

int SampleMLP::mNumber {0}

private

The number to classify.

mWeightMap

std::map<std::string, std::pair<nvinfer1::Dims, nvinfer1::Weights> > SampleMLP::mWeightMap

private

The weight name to weight value map.

mEngine

std::shared_ptr<nvinfer1::ICudaEngine> SampleMLP::mEngine

private

The TensorRT engine used to run the network.

weightsMemory

std::vector<SampleUniquePtr<nvinfer1::IHostMemory> > SampleMLP::weightsMemory

private

Host weights memory holder.

---

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

• sampleMLP.cpp