caffe2 Namespace Reference

Simple registry implementation in Caffe2 that uses static variables to register object creators during program initialization time. More...

Namespaces
	enforce_detail
	Rich logging messages.
	gpu_single_thread
	Code for special net type that uses one executor -thread per GPU.

Data Structures
class	ArgumentHelper
	A helper class to index into arguments. More...
class	AsyncDAGNet
class	AvgExportedStat
class	Blob
	Blob is a general container that hosts a typed pointer. More...
class	BlobDeserializerBase
	BlobDeserializerBase is an abstract class that deserializes a blob from a BlobProto or a TensorProto. More...
class	BlobSerializerBase
	BlobSerializerBase is an abstract class that serializes a blob to a string. More...
class	BlobsQueue
struct	BlobStatGetter
struct	BlobStatRegistry
class	Caffe2FlagParser
struct	CompiledExecutionStep
struct	CPUAllocator
class	CPUContext
	The CPU Context, representing the bare minimum of what a Context class in Caffe2 should implement. More...
class	CUDAContext
class	CudaRTCFunction
class	cudnnFilterDescWrapper
class	CuDNNHandles
	CuDNNHandles wraps around cudnnHandle_t so they can be properly destructed when threads exit. More...
class	CuDNNState
class	cudnnTensorDescWrapper
	cudnnTensorDescWrapper is the placeholder that wraps around a cudnnTensorDescriptor_t, allowing us to do descriptor change as-needed during runtime. More...
class	cudnnTypeWrapper
	cudnnTypeWrapper is a wrapper class that allows us to refer to the cudnn type in a template function. More...
class	cudnnTypeWrapper< double >
class	cudnnTypeWrapper< float >
class	cudnnTypeWrapper< float16 >
struct	CuDNNWorkspace
	CuDNNWorkspace is a wrapper around a raw cuda pointer that holds the cudnn scratch space. More...
class	CuDNNWrapper
	CuDNNWrapper is a class that wraps the cudnn handles and cudnn workspaces. More...
class	DAGNet
class	DAGNetBase
struct	DefaultCPUAllocator
class	DefaultEngine
class	DetailedExportedStat
class	DeviceGuard
struct	DeviceTypeRegisterer
struct	DispatchHelper
struct	DispatchHelper< FixedValues< FirstVal, Values... >, ExtraArgs... >
struct	DispatchHelper< FixedValues<>, ExtraArgs... >
class	ElementwiseRTCOp
	A GPU operator that can generate limited elementwise operations. More...
class	EnforceNotMet
class	ExportedStat
struct	ExportedStatValue
class	FixedDivisor
class	FixedDivisor< int32_t >
struct	FixedValues
struct	GenericTensorImplementation
class	GradientMakerBase
struct	GradientNotImplementedYet
	A helper class to indicate that the gradient mechanism is not ready. More...
struct	GradientOpsMeta
	A struct that holds the gradient operators and related gradient maps. More...
struct	GradientWrapper
class	InitRegisterer
class	LoggerVoidify
class	MaxPoolGradientRTCOp
class	MaxPoolRTCOp
class	MemoryAllocationReporter
class	MessageLogger
class	MKLContext
	The MKL Context, which is largely the same as the CPUContext. More...
class	MPICommonWorldWrapper
	A simple wrapper over an MPI common world. More...
class	MPIDataTypeWrapper
class	NetBase
class	NoGradient
	A helper class to indicate that the operator does not need gradient computation. More...
class	Operator
class	OperatorBase
class	OpSchema
	A class to record the schema of an op. More...
class	OpSchemaRegistry
	A registry to hold all the operator schemas. More...
struct	PinnedCPUAllocator
	An allocator that does the CPU memory allocation with pinned memory. More...
class	Predictor
class	QTensor
class	QTensorDeserializer
class	QTensorSerializer
class	Registerer
class	Registry
	A template class that allows one to register classes by keys. More...
class	ScopeGuardImpl
class	ScopeGuardImplBase
	ScopeGuard is a general implementation of the "Initialization is Resource Acquisition" idiom. More...
class	SignalHandler
class	SimpleNet
class	SimpleQueue
class	SkipIndices
class	SkipIndices<>
struct	Stat
class	StatRegistry
	Holds a map of atomic counters keyed by name. More...
class	StatValue
struct	StopOnSignal
class	StringDeserializer
	StringDeserializer is the deserializer for Strings. More...
class	StringSerializer
	StringSerializer is the serializer for String. More...
class	Tensor
	Tensor is the basic class in Caffe2 that stores a contiguous memory with its shape information. More...
class	TensorDeserializer
	TensorDeserializer is the deserializer for Tensors. More...
class	TensorPrinter
class	TensorSerializer
	TensorSerializer is the serializer for Tensors. More...
struct	TensorTypes
struct	TensorTypes2
class	ThreadLocalCUDAObjects
	A struct to host thread-local cuda objects. More...
struct	ThrowInTheTowelIfGradientIsCalled
	A helper class to indicate that the operator should have no gradient. More...
class	Timer
	A simple timer object for measuring time. More...
class	TypeMeta
	TypeMeta is a thin class that allows us to store the type of a container such as a blob, or the data type of a tensor, with a unique run-time id. More...
struct	TypeNameRegisterer
class	UnsupportedOperatorFeature
class	Workspace
	Workspace is a class that holds all the related objects created during runtime: (1) all blobs, and (2) all instantiated networks. More...
class	ZmqContext
class	ZmqMessage
class	ZmqSocket

Typedefs
typedef int64_t	TIndex
template<typename Key , typename Value >
using	CaffeMap = std::map< Key, Value >
typedef Tensor< CUDAContext >	TensorCUDA
typedef Registry< std::string, OperatorBase, const OperatorDef &, Workspace * > *(*	RegistryFunction) ()
typedef ScopeGuardImplBase &&	ScopeGuard
	This is largely unneeded if you just use auto for your guards.
using	ExportedStatList = std::vector< ExportedStatValue >
	Holds names and values of counters exported from a StatRegistry.
using	ExportedStatMap = std::unordered_map< std::string, int64_t >
typedef Tensor< CPUContext >	TensorCPU
typedef TypeMeta(*	TypeCall) (void *)
typedef vector< TIndex >(*	ShapeCall) (void *, bool &shares_data, size_t &capacity)
typedef intptr_t	CaffeTypeId
template<typename T >
using	EigenMatrixMap = Eigen::Map< Eigen::Matrix< T, Eigen::Dynamic, Eigen::Dynamic > >
template<typename T >
using	EigenArrayMap = Eigen::Map< Eigen::Array< T, Eigen::Dynamic, Eigen::Dynamic > >
template<typename T >
using	EigenVectorMap = Eigen::Map< Eigen::Matrix< T, Eigen::Dynamic, 1 > >
template<typename T >
using	EigenVectorArrayMap = Eigen::Map< Eigen::Array< T, Eigen::Dynamic, 1 > >
template<typename T >
using	ConstEigenMatrixMap = Eigen::Map< const Eigen::Matrix< T, Eigen::Dynamic, Eigen::Dynamic > >
template<typename T >
using	ConstEigenArrayMap = Eigen::Map< const Eigen::Array< T, Eigen::Dynamic, Eigen::Dynamic > >
template<typename T >
using	ConstEigenVectorMap = Eigen::Map< const Eigen::Matrix< T, Eigen::Dynamic, 1 > >
template<typename T >
using	ConstEigenVectorArrayMap = Eigen::Map< const Eigen::Array< T, Eigen::Dynamic, 1 > >

Enumerations
enum	CudaMemoryPoolType { NONE = 0, CNMEM = 1, CUB = 2 }
enum	StorageOrder { UNKNOWN = 0, NHWC = 1, NCHW = 2 }

Functions
void	ConvertToRawDataset (const string &input_db_name, const string &output_db_name)
void	ReadImage (std::ifstream *file, int *label, char *buffer)
void	WriteToDB (const string &filename, const int num_items, const int &offset, db::DB *db)
void	ConvertCIFAR ()
void	ConvertImageDataset (const string &input_folder, const string &list_filename, const string &output_db_name, const bool shuffle)
uint32_t	swap_endian (uint32_t val)
void	convert_dataset (const char *image_filename, const char *label_filename, const char *db_path, const int data_limit)
void	run ()
void	swap (Blob &lhs, Blob &rhs)
	CAFFE_DEFINE_TYPED_REGISTRY (BlobSerializerRegistry, CaffeTypeId, BlobSerializerBase)
	CAFFE_DEFINE_REGISTRY (BlobDeserializerRegistry, BlobDeserializerBase)
	CAFFE_DECLARE_TYPED_REGISTRY (BlobSerializerRegistry, CaffeTypeId, BlobSerializerBase)
unique_ptr< BlobSerializerBase >	CreateSerializer (CaffeTypeId id)
	CAFFE_DECLARE_REGISTRY (BlobDeserializerRegistry, BlobDeserializerBase)
unique_ptr< BlobDeserializerBase >	CreateDeserializer (const string &type)
template<typename T , typename... Args>
std::enable_if<!std::is_array< T >::value, std::unique_ptr< T > >::type	make_unique (Args &&... args)
template<typename T >
std::enable_if< std::is_array< T >::value, std::unique_ptr< T > >::type	make_unique (const size_t n)
template<typename T , typename... Args>
std::enable_if< std::extent< T >::value !=0, std::unique_ptr< T > >::type	make_unique (Args &&...)=delete
template<typename Dst , typename Src >
Dst	dynamic_cast_if_rtti (Src ptr)
size_t	cudnnCompiledVersion ()
size_t	cudnnRuntimeVersion ()
void	CheckCuDNNVersions ()
cudnnTensorFormat_t	GetCudnnTensorFormat (const StorageOrder &order)
	A wrapper function to convert the Caffe storage order to cudnn storage order enum values.
int	NumCudaDevices ()
	Returns the number of devices.
void	SetDefaultGPUID (const int deviceid)
	Sets the default GPU id for Caffe2. More...
int	GetDefaultGPUID ()
	Gets the default GPU id for Caffe2.
int	GetCurrentGPUID ()
	Gets the current GPU id. More...
int	GetGPUIDForPointer (const void *ptr)
	Gets the GPU id that the current pointer is located at.
const cudaDeviceProp &	GetDeviceProperty (const int device)
	Gets the device property for the given device.
void	DeviceQuery (const int deviceid)
	Runs a device query function and prints out the results to LOG(INFO).
bool	GetCudaPeerAccessPattern (vector< vector< bool > > *pattern)
	Return a peer access pattern by returning a matrix (in the format of a nested vector) of boolean values specifying whether peer access is possible. More...
const char *	cublasGetErrorString (cublasStatus_t error)
	Return a human readable cublas error string.
const char *	curandGetErrorString (curandStatus_t error)
	Return a human readable curand error string.
int	CudaVersion ()
	A runtime function to report the cuda version that Caffe2 is built with.
bool	HasCudaGPU ()
	Check if the current running session has a cuda gpu present. More...
int	CAFFE_GET_BLOCKS (const int N)
	Compute the number of blocks needed to run N threads.
CPUAllocator *	GetCPUAllocator ()
void	SetCPUAllocator (CPUAllocator *alloc)
CudaMemoryPoolType	GetCudaMemoryPoolType ()
	Gets the current memory pool type used by Caffe2. More...
	CAFFE_KNOWN_TYPE (db::DBReader)
	CAFFE_KNOWN_TYPE (db::Cursor)
	CAFFE_DEFINE_REGISTRY (Caffe2FlagsRegistry, Caffe2FlagParser, const string &)
void	SetUsageMessage (const string &str)
	Sets the usage message when a commandline tool is called with "--help".
const char *	UsageMessage ()
	Returns the usage message for the commandline tool set by SetUsageMessage.
bool	ParseCaffeCommandLineFlags (int *pargc, char ***pargv)
	Parses the commandline flags. More...
bool	CommandLineFlagsHasBeenParsed ()
	Checks if the commandline flags has already been passed.
	CAFFE_DECLARE_REGISTRY (Caffe2FlagsRegistry, Caffe2FlagParser, const string &)
bool	GlobalInit (int *pargc, char ***argv)
	Initialize the global environment of caffe2. More...
std::string	StripBasename (const std::string &full_path)
size_t	ReplaceAll (string &s, const char *from, const char *to)
void	SetStackTraceFetcher (std::function< string(void)> fetcher)
void	SetOperatorLogger (std::function< void(const OperatorDef &)> tracer)
std::function< void(const OperatorDef &)>	GetOperatorLogger ()
bool	InitCaffeLogging (int *argc, char **argv)
constexpr bool	IsUsingGoogleLogging ()
void	MakeStringInternal (std::stringstream &ss)
template<typename T >
void	MakeStringInternal (std::stringstream &ss, const T &t)
template<typename T , typename... Args>
void	MakeStringInternal (std::stringstream &ss, const T &t, const Args &... args)
template<typename... Args>
string	MakeString (const Args &... args)
template<>
string	MakeString (const string &str)
string	MakeString (const char *c_str)
template<class Container >
string	Join (const string &delimiter, const Container &v)
template<class T >
void	LogMessageFatal (const char *file, int line, const T &message)
template<typename T >
T &	CheckNotNullCommon (const char *file, int line, const char *names, T &t)
template<typename T >
T *	CheckNotNull (const char *file, int line, const char *names, T *t)
template<typename T >
T &	CheckNotNull (const char *file, int line, const char *names, T &t)
template<class First , class Second >
std::ostream &	operator<< (std::ostream &out, const std::pair< First, Second > &p)
template<class Iter >
void	PrintSequence (std::ostream &ss, Iter begin, Iter end)
	CAFFE_DEFINE_REGISTRY (NetRegistry, NetBase, const NetDef &, Workspace *)
unique_ptr< NetBase >	CreateNet (const NetDef &net_def, Workspace *ws)
	Creates a network, accessing / creating blobs in the given workspace. More...
	CAFFE_DECLARE_REGISTRY (NetRegistry, NetBase, const NetDef &, Workspace *)
	REGISTER_NET (async_dag, AsyncDAGNet)
unique_ptr< OperatorBase >	CreateOperator (const OperatorDef &operator_def, Workspace *ws)
std::map< int32_t, OperatorRegistry * > *	gDeviceTypeRegistry ()
	CAFFE_DEFINE_REGISTRY (CPUOperatorRegistry, OperatorBase, const OperatorDef &, Workspace *)
	CAFFE_REGISTER_DEVICE_TYPE (DeviceType::CPU, CPUOperatorRegistry)
	CAFFE_DEFINE_REGISTRY (CUDAOperatorRegistry, OperatorBase, const OperatorDef &, Workspace *)
	CAFFE_REGISTER_DEVICE_TYPE (DeviceType::CUDA, CUDAOperatorRegistry)
	CAFFE_DEFINE_REGISTRY (GradientRegistry, GradientMakerBase, const OperatorDef &, const vector< GradientWrapper > &)
GradientOpsMeta	GetGradientForOp (const OperatorDef &def, const vector< GradientWrapper > &g_output)
	Gets the GradientOpsMeta for the given operator def.
TensorShapes	InferBlobShapesAndTypesFromWorkspace (Workspace *ws, const vector< std::unique_ptr< NetDef >> &nets)
TensorShapes	InferBlobShapesAndTypesFromMap (const CaffeMap< std::string, std::vector< TIndex >> &blob_dimensions, const vector< std::unique_ptr< NetDef >> &nets)
	CAFFE2_DEFINE_TENSOR_TYPES_DISPATCHER (TensorTypes, DoRunWithType, DoRunWithOtherType) CAFFE2_DEFINE_TENSOR_TYPES_DISPATCHER(TensorTypes2
	CAFFE_DECLARE_REGISTRY (CPUOperatorRegistry, OperatorBase, const OperatorDef &, Workspace *)
	CAFFE_DECLARE_REGISTRY (CUDAOperatorRegistry, OperatorBase, const OperatorDef &, Workspace *)
	CAFFE_DECLARE_REGISTRY (GradientRegistry, GradientMakerBase, const OperatorDef &, const vector< GradientWrapper > &)
std::ostream &	operator<< (std::ostream &out, const OpSchema &schema)
TensorShape	CreateTensorShape (vector< int > dims, ::caffe2::TensorProto_DataType dt)
vector< TIndex >	GetDimsVector (const TensorShape &shape)
	CAFFE_KNOWN_TYPE (QTensor< CPUContext >)
template<typename FunctionType >
ScopeGuardImpl< typename std::decay< FunctionType >::type >	MakeGuard (FunctionType &&fn) noexcept(std::is_nothrow_constructible< typename std::decay< FunctionType >::type, FunctionType >::value)
ExportedStatMap	toMap (const ExportedStatList &stats)
	CAFFE_KNOWN_TYPE (Tensor< CPUContext >)
TypeCall	GetTypeCallFunction (CaffeTypeId id)
void	RegisterTypeCallFunction (CaffeTypeId id, TypeCall c)
ShapeCall	GetShapeCallFunction (CaffeTypeId id)
void	RegisterShapeCallFunction (CaffeTypeId id, ShapeCall c)
vector< TIndex >	ToVectorTIndex (const std::vector< int > &src)
	A utility function to convert vector<int> to vector<TIndex>.
TIndex	size_from_dim_ (int k, vector< TIndex > dims)
	Return product of all dimensions starting from K.
TIndex	size_to_dim_ (int k, vector< TIndex > dims)
int	canonical_axis_index_ (int axis_index, int ndims)
template<class Context >
TypeMeta	GetTensorType (void *c)
template<class Context >
vector< TIndex >	GetTensorShape (void *c, bool &shares_data, size_t &capacity)
std::map< CaffeTypeId, string > &	gTypeNames ()
std::set< string > &	gRegisteredTypeNames ()
string	Demangle (const char *name)
string	GetExceptionString (const std::exception &e)
	CAFFE_KNOWN_TYPE (float)
	CAFFE_KNOWN_TYPE (int)
	CAFFE_KNOWN_TYPE (std::string)
	CAFFE_KNOWN_TYPE (bool)
	CAFFE_KNOWN_TYPE (uint8_t)
	CAFFE_KNOWN_TYPE (int8_t)
	CAFFE_KNOWN_TYPE (uint16_t)
	CAFFE_KNOWN_TYPE (int16_t)
	CAFFE_KNOWN_TYPE (int64_t)
	CAFFE_KNOWN_TYPE (float16)
	CAFFE_KNOWN_TYPE (double)
	CAFFE_KNOWN_TYPE (char)
	CAFFE_KNOWN_TYPE (std::unique_ptr< std::mutex >)
	CAFFE_KNOWN_TYPE (std::unique_ptr< std::atomic< bool >>)
	CAFFE_KNOWN_TYPE (std::vector< int64_t >)
	CAFFE_KNOWN_TYPE (std::vector< unsigned long >)
TensorProto::DataType	TypeMetaToDataType (const TypeMeta &meta)
const TypeMeta &	DataTypeToTypeMeta (const TensorProto::DataType &dt)
StorageOrder	StringToStorageOrder (const string &str)
constexpr char	NameScopeSeparator ()
struct	CAFFE2_ALIGNED (2) __f16
string	GetUniqueName ()
template<typename T_IN , typename T_OUT , class Context >
bool	TransformOnGPU (Tensor< Context > &X, Tensor< Context > *Y, T_OUT mean, T_OUT std, Context *context)
	CAFFE_KNOWN_TYPE (MPICommonWorldWrapper)
std::mutex &	MPIMutex ()
MPI_Comm	GlobalMPIComm ()
	Gets the global MPI communicator used by Caffe2. More...
void	SetGlobalMPIComm (MPI_Comm new_comm)
	Sets the global MPI communicator. More...
int	MPICommSize (MPI_Comm comm)
	A helper function to return the size of the given communicator.
int	MPICommRank (MPI_Comm comm)
	A helper function to return the rank of the given communicator.
void	MPISetupPeers (const int replicas, const string &role, const string &job_path)
	A function used to perform peer setup so one does not need to use mpirun / mpiexec to run the binary. More...
void	CheckInitializedMPI ()
	CAFFE_DECLARE_REGISTRY (MKLOperatorRegistry, OperatorBase, const OperatorDef &, Workspace *)
void	MurmurHash3_x86_32 (const void *key, int len, uint32_t seed, void *out)
void	MurmurHash3_x86_128 (const void *key, const int len, uint32_t seed, void *out)
void	MurmurHash3_x64_128 (const void *key, const int len, const uint32_t seed, void *out)
std::string	DeviceTypeName (const int32_t &d)
bool	ReadStringFromFile (const char *filename, string *str)
bool	WriteStringToFile (const string &str, const char *filename)
bool	ReadProtoFromTextFile (const char *filename, Message *proto)
void	WriteProtoToTextFile (const Message &proto, const char *filename)
bool	ReadProtoFromBinaryFile (const char *filename, MessageLite *proto)
void	WriteProtoToBinaryFile (const MessageLite &proto, const char *filename)
template<>
Argument	MakeArgument (const string &name, const MessageLite &value)
const Argument &	GetArgument (const OperatorDef &def, const string &name)
bool	GetFlagArgument (const OperatorDef &def, const string &name, bool def_value)
Argument *	GetMutableArgument (const string &name, const bool create_if_missing, OperatorDef *def)
bool	ReadProtoFromBinaryFile (const string filename, MessageLite *proto)
void	WriteProtoToBinaryFile (const MessageLite &proto, const string &filename)
string	ProtoDebugString (const Message &proto)
bool	ReadProtoFromTextFile (const string filename, Message *proto)
void	WriteProtoToTextFile (const Message &proto, const string &filename)
bool	ReadProtoFromFile (const char *filename, Message *proto)
bool	ReadProtoFromFile (const string &filename, Message *proto)
template<class IterableInputs , class IterableOutputs , class IterableArgs >
OperatorDef	CreateOperatorDef (const string &type, const string &name, const IterableInputs &inputs, const IterableOutputs &outputs, const IterableArgs &args, const DeviceOption &device_option, const string &engine)
template<class IterableInputs , class IterableOutputs , class IterableArgs >
OperatorDef	CreateOperatorDef (const string &type, const string &name, const IterableInputs &inputs, const IterableOutputs &outputs, const IterableArgs &args)
template<class IterableInputs , class IterableOutputs >
OperatorDef	CreateOperatorDef (const string &type, const string &name, const IterableInputs &inputs, const IterableOutputs &outputs)
bool	HasArgument (const OperatorDef &def, const string &name)
template<typename T >
Argument	MakeArgument (const string &name, const T &value)
template<typename T >
void	AddArgument (const string &name, const T &value, OperatorDef *def)
std::vector< std::string >	split (char separator, const std::string &string)
size_t	editDistance (const std::string &s1, const std::string &s2, size_t max_distance)
int32_t	editDistanceHelper (const char *s1, size_t s1_len, const char *s2, size_t s2_len, std::vector< size_t > &current, std::vector< size_t > &previous, std::vector< size_t > &previous1, size_t max_distance)

Variables
DoRunWithOtherType2 typedef Registry< std::string, OperatorBase, const OperatorDef &, Workspace * >	OperatorRegistry
const int	kCIFARSize = 32
const int	kCIFARImageNBytes = kCIFARSize * kCIFARSize * 3
const int	kCIFAR10BatchSize = 10000
const int	kCIFAR10TestDataSize = 10000
const int	kCIFAR10TrainBatches = 5
const int	kCIFAR100TrainDataSize = 50000
const int	kCIFAR100TestDataSize = 10000
constexpr auto	kTensorBlobType = "Tensor"
constexpr auto	kChunkIdSeparator = "#%"
constexpr int	kDefaultChunkSize = -1
constexpr int	kNoChunking = 0
constexpr int	CAFFE_CUDA_NUM_THREADS = 512
constexpr int	CAFFE_MAXIMUM_NUM_BLOCKS = 4096
constexpr size_t	gCaffe2Alignment = 32
std::function< void(const OperatorDef &)>	GetOperatorLogger ()
	DoRunWithType2
constexpr int	kCannotComputeNumOutputs = -1
constexpr auto	kQTensorBlobQType = "QTensor"
constexpr int	k_limit_default_ = 1000
	float16

Detailed Description

Simple registry implementation in Caffe2 that uses static variables to register object creators during program initialization time.

Copyright (c) 2016, NVIDIA CORPORATION, All rights reserved.

Copyright 2016 Facebook.

Author

Tudor Bosman (tudor.[email protected][email protected].[email protected].com)

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Function Documentation

CreateNet()

unique_ptr< NetBase > caffe2::CreateNet	(	const NetDef &	net_def,
		Workspace *	ws
	)

Creates a network, accessing / creating blobs in the given workspace.

Note that this is different from Workspace::CreateNet. The latter adds the created net object to the workspace's net map, while this function returns a standalone net object.

Definition at line 66 of file net.cc.

GetCudaMemoryPoolType()

CudaMemoryPoolType caffe2::GetCudaMemoryPoolType

(

)

Gets the current memory pool type used by Caffe2.

The memory pool is set up during caffe2's global initialization time.

GetCudaPeerAccessPattern()

bool caffe2::GetCudaPeerAccessPattern

(

vector< vector< bool > > *

pattern

)

Return a peer access pattern by returning a matrix (in the format of a nested vector) of boolean values specifying whether peer access is possible.

This function returns false if anything wrong happens during the query of the GPU access pattern.

Definition at line 159 of file common_gpu.cc.

GetCurrentGPUID()

int caffe2::GetCurrentGPUID

(

)

Gets the current GPU id.

This is a simple wrapper around cudaGetDevice().

Definition at line 93 of file common_gpu.cc.

GlobalInit()

bool caffe2::GlobalInit	(	int *	pargc,
		char ***	argv
	)

Initialize the global environment of caffe2.

Caffe2 uses a registration pattern for initialization functions. Custom initialization functions should take the signature bool (func)(int, char***) where the pointers to argc and argv are passed in. Caffe2 then runs the initialization in three phases: (1) Functions registered with REGISTER_CAFFE2_EARLY_INIT_FUNCTION. Note that since it is possible the logger is not initialized yet, any logging in such early init functions may not be printed correctly. (2) Parses Caffe-specific commandline flags, and initializes caffe logging. (3) Functions registered with REGISTER_CAFFE2_INIT_FUNCTION. If there is something wrong at each stage, the function returns false. If the global initialization has already been run, the function returns false as well.

Definition at line 15 of file init.cc.

GlobalMPIComm()

MPI_Comm caffe2::GlobalMPIComm

(

)

Gets the global MPI communicator used by Caffe2.

In default, this is MPI_COMM_WORLD unless you call SetGlobalMPIComm().

Definition at line 20 of file mpi_common.cc.

HasCudaGPU()

bool caffe2::HasCudaGPU ( )

inline

Check if the current running session has a cuda gpu present.

Note that this is different from having caffe2 built with cuda. Building Caffe2 with cuda only guarantees that this function exists. If there are no cuda gpus present in the machine, or there are hardware configuration problems like an insufficient driver, this function will still return false, meaning that there is no usable GPU present.

Definition at line 64 of file common_gpu.h.

MPISetupPeers()

void caffe2::MPISetupPeers	(	const int	replicas,
		const string &	role,
		const string &	job_path
	)

A function used to perform peer setup so one does not need to use mpirun / mpiexec to run the binary.

Note that if you use mpirun or mpiexec to set up the common world, do not use this function - MPI_Init would have already set that up.

This also assumes that you have a common path (like NFS) that multiple instances can read from.

Inputs: replicas (int): the number of replicas that mpi will run with. role (string): the role of this process, "server" or "client". job_path (string): a file name that the server will write its port into and the clients will read the server's port from.

Definition at line 94 of file mpi_common.cc.

ParseCaffeCommandLineFlags()

bool caffe2::ParseCaffeCommandLineFlags	(	int *	pargc,
		char ***	pargv
	)

Parses the commandline flags.

This command parses all the commandline arguments passed in via pargc and argv. Once it is finished, partc and argv will contain the remaining commandline args that caffe2 does not deal with. Note that following convention, argv[0] contains the binary name and is not parsed.

Definition at line 55 of file flags.cc.

SetDefaultGPUID()

void caffe2::SetDefaultGPUID

(

const int

deviceid

)

Sets the default GPU id for Caffe2.

If an operator is set to run on Cuda GPU but no gpu id is given, we will use the default gpu id to run the operator. Before this function is explicitly called, GPU 0 will be the default GPU id.

Definition at line 80 of file common_gpu.cc.

SetGlobalMPIComm()

void caffe2::SetGlobalMPIComm

(

MPI_Comm

new_comm

)

Sets the global MPI communicator.

Caffe2 takes over the ownership of the passed in communicator.

Definition at line 24 of file mpi_common.cc.