LayerTypes.h File Reference

Interface types needed by layer objects. More...

#include <nvneural/CoreTypes.h>

Go to the source code of this file.

Classes
class	nvneural::IBypassLayer
	IBypassLayer is an optional interface that marks a layer as one that can potentially forward its input and be skipped. More...
class	nvneural::IConcatenationLayer
	IConcatenationLayer is an optional interface implemented by layers that marks them as concatenation layers. More...
class	nvneural::IConcatenationLayer2
	IConcatenationLayer2 is an optional interface implemented by layers that marks them as concatenation layers. More...
class	nvneural::IInnerActivationLayer
	IInnerActivationLayer is an optional interface implemented by layers that fuse pairs of operations. More...
class	nvneural::IInputLayer
	IInputLayer is an optional interface implemented by layers that marks them as input layers. More...
class	nvneural::ILayer
	ILayer is the base class for neural network layers. More...
class	nvneural::IRuntimeOptionsLayer
	IRuntimeOptionsLayer is an optional interface that marks a layer as one that uses runtime options. More...
class	nvneural::IStatusMessageLayer
	IStatusMessageLayer is an optional interface intended for ILayer to provide status messages. More...

Namespaces
	nvneural::runtimeoptionsattributes
	Enumeration describing the runtime options modifiers, see runtimeoptionstype.
	nvneural::runtimeoptionstype
	Enumeration describing the runtime options given to the host, see IRuntimeOptionsLayer.

Enumerations
enum class	nvneural::WeightsQuery : std::uint32_t { Required = 1 , Aligned = 2 }
	WeightsQuery describes the different types of queries for weights data. More...

Variables
const char *const	nvneural::runtimeoptionstype::booltf = "boolTF"
	Requires nothing, sent "true" or "false".
const char *const	nvneural::runtimeoptionsattributes::list = "list"
	Defines a list of string, ex. "list":["option_1", "option_2"].
const char *const	nvneural::runtimeoptionsattributes::max = "max"
	Defines a max value, ex. "max":"1" or "max":"1.0".
const char *const	nvneural::runtimeoptionsattributes::min = "min"
	Defines a min value, ex. "min":"0" or "min":"0.5".
const char *const	nvneural::runtimeoptionstype::numfloat = "numFloat"
	Requires nothing, sent float.
const char *const	nvneural::runtimeoptionstype::numint = "numInt"
	Requires nothing, sent int.
const char *const	nvneural::runtimeoptionsattributes::step = "step"
	Defines a step value, ex. "step":"1" or "step":"0.5".
const char *const	nvneural::runtimeoptionstype::stringlist = "stringList"
	Requires runtimeoptionsattributes::list, sent ordinal.
const char *const	nvneural::runtimeoptionstype::textedit = "textEdit"
	Requires nothing, sent string.
const char *const	nvneural::runtimeoptionstype::textlabel = "textLabel"
	Requires nothing, sent nothing, send string.
const char *const	nvneural::runtimeoptionstype::trigger = "trigger"
	Requires nothing, sent empty string.

Detailed Description

Interface types needed by layer objects.

Enumeration Type Documentation

WeightsQuery

enum nvneural::WeightsQuery : std::uint32_t

strong

WeightsQuery describes the different types of queries for weights data.

ILayer::weightsDimensions implementations passed an unknown weights query should return a zero-valued TensorDimension.

Enumerator
Required	Represents the "official" dimensions expected from compatible weights. External weights data not matching this dimension may be rejected by the weights loader. Weights loaders such as FilesystemWeightsLoader may use this size to synthesize random weights for profiling or validation purposes.
Aligned	Represents the "actual" dimensions expected from compatible weights. During upload, network backends will automatically pad weights up to this size, allowing the kernels used by the layer implementations to assume aligned storage when performing inference.