CoreHelpers.h File Reference

Common helper classes and template function implementations. More...

#include <atomic>
#include <functional>
#include <string>
#include <typeinfo>
#include <type_traits>
#include <utility>
#include <nvneural/CoreTypes.h>
#include <nvneural/LayerTypes.h>
#include <nvneural/RefPtr.h>
#include <nvneural/RefObject.h>

Go to the source code of this file.

Classes
struct	std::hash< nvneural::NetworkBackendId >
	Specialization of std::hash for strongly typed network backends. More...

Enumerations
enum class	nvneural::TensorDimensionParseOrder : std::uint32_t { Tensor = 0 , Image = 1 }
	Order to read TensorDimension components in for TensorDimensionFromString. More...

Functions
TensorDimension	nvneural::align (const TensorDimension &dim, const TensorDimension &stepping) noexcept
	Rounds a TensorDimension object up to be a multiple of a provided set of strides. More...
constexpr std::size_t	nvneural::dataSize (TensorDataType dataType) noexcept
	Returns the size of an individual tensor element of a particular type, similar to sizeof. More...
ILogger *	nvneural::DefaultLogger ()
	Returns a pointer to the default logger for this module. More...
template<typename TDividend , typename TDivisor >
constexpr auto	nvneural::DivideRoundingUp (TDividend dividend, TDivisor divisor) noexcept
	Division operation, rounding up. More...
NeuralResult	nvneural::getInputLayerIndex (ILayer *pLayer, ILayer *pInputLayer, size_t *pIndex)
	Retrieves the layer index in the input list. More...
void	nvneural::SetDefaultLogger (ILogger *pDefaultLogger)
	Sets a default logger for the module.
std::vector< std::string >	nvneural::split (std::string strToSplit, char delimiter, bool skip_empty=false)
	Splits a string by a delimiter, returns a vector of the split strings. More...
std::vector< std::string >	nvneural::splitEscape (std::string strToSplit, char delimiter, bool skip_empty=false)
	Splits a string by a delimiter, returns a vector of the split strings. More...
NeuralResult	nvneural::TensorDimensionFromString (TensorDimension &outDim, const std::string &dimStr, TensorDimensionParseOrder parseOrder, size_t fillValue) noexcept
	Parses a TensorDimension from a UTF-8 string. More...
std::string	nvneural::TensorDimensionToString (const TensorDimension &dim) noexcept
	Converts a TensorDimension to a UTF-8 string. More...

Detailed Description

Common helper classes and template function implementations.

Enumeration Type Documentation

TensorDimensionParseOrder

enum nvneural::TensorDimensionParseOrder : std::uint32_t

strong

Order to read TensorDimension components in for TensorDimensionFromString.

Enumerator
Tensor	This order is used for parameters that typically describe tensor dimensions. Channels are parsed in this order: 1 value: C 2 values: H W 3 values: C H W 4 values: N C H W
Image	This order is used for parameters that typically describe image or kernel sizes. Channels are parsed in this order: 1 value: H + W (both components receive the same value) 2 values: W H 3 values: W H C 4 values: W H C N

Function Documentation

align()

TensorDimension nvneural::align ( const TensorDimension &  dim, const TensorDimension &  stepping  )

inlinenoexcept

Rounds a TensorDimension object up to be a multiple of a provided set of strides.

Dimensions are rounded individually by component.

Parameters

dim	Size to round up
stepping	Component multiples to round up to

dataSize()

constexpr std::size_t nvneural::dataSize ( TensorDataType  dataType )

inlineconstexprnoexcept

Returns the size of an individual tensor element of a particular type, similar to sizeof.

You should implement your own version of the function if working with custom types; unsupported types will return zero.

Parameters

dataType

Type to query

DefaultLogger()

ILogger * nvneural::DefaultLogger

(

)

Returns a pointer to the default logger for this module.

If no logger has been assigned via IPlugin::setDefaultLogger or SetDefaultLogger, the logger will silently ignore all messages sent to it.

DivideRoundingUp()

template<typename TDividend , typename TDivisor >

constexpr auto nvneural::DivideRoundingUp ( TDividend  dividend, TDivisor  divisor  )

inlineconstexprnoexcept

Division operation, rounding up.

The primary application of this helper is CUDA grid launches: assuming a grid of "dividend" threads to launch, and a block size of "divisor" threads per block, the return value of this function will be the appropriate number of blocks to launch.

Template Parameters

TDividend	Type of the dividend; should be integer-like
TDivisor	Type of the divisor; should be integer-like

Parameters

dividend	Value to divide
divisor	Value to divide by

getInputLayerIndex()

NeuralResult nvneural::getInputLayerIndex ( ILayer *  pLayer, ILayer *  pInputLayer, size_t *  pIndex  )

inline

Retrieves the layer index in the input list.

Parameters

pLayer	the layer to scan inputs from
pInputLayer	one of the input layer
pIndex	index output

Returns

Success if pInputLayer is part of the input list, else Failure (in that case *pIndex was untouched)

split()

std::vector<std::string> nvneural::split ( std::string  strToSplit, char  delimiter, bool  skip_empty = false  )

inline

Splits a string by a delimiter, returns a vector of the split strings.

This is generally used when splitting dimensions (IE 1x1 or 1x1,1x1). There is an optional flag that allows skipping of empty split strings.

Parameters

strToSplit	the string to split
delimiter	single character that denotes the delimiter
skip_empty	optional, skips returning empty split strings

Returns

a vector of the split strings

splitEscape()

std::vector<std::string> nvneural::splitEscape ( std::string  strToSplit, char  delimiter, bool  skip_empty = false  )

inline

Splits a string by a delimiter, returns a vector of the split strings.

This version considers escaped characters.

This is generally used when splitting text strings where escaped characters can be used in the text itself. There is an optional flag that allows skipping of empty split strings.

Parameters

strToSplit	the string to split
delimiter	single character that denotes the delimiter
skip_empty	optional, skips returning empty split strings

Returns

a vector of the split strings

TensorDimensionFromString()

NeuralResult nvneural::TensorDimensionFromString ( TensorDimension &  outDim, const std::string &  dimStr, TensorDimensionParseOrder  parseOrder, size_t  fillValue  )

noexcept

Parses a TensorDimension from a UTF-8 string.

The interpretations of strings like "2x3x4" are determined by the parse order. Alternately, dimensions may be specified explicitly ("2w4c"). Mixed use of implicit and explicit dimensions ("4n5x6") is not permitted within the same input string.

Parameters

outDim	TensorDimension variable receiving the parsed value.
dimStr	String to parse.
parseOrder	Order for parsing components if not explicit in the string.
fillValue	Value to fill unparsed components with.

TensorDimensionToString()

std::string nvneural::TensorDimensionToString ( const TensorDimension &  dim )

noexcept

Converts a TensorDimension to a UTF-8 string.

The string uses explicit component syntax for unambiguous round-trips through the TensorDimensionFromString function.

Parameters

dim	Dimension to convert