rabit::engine Namespace Reference

core interface of the engine More...

Namespaces
namespace	mpi
	namespace that contains stubs to be compatible with MPI

Data Structures
class	AllreduceBase
	implementation of basic Allreduce engine More...
class	AllreduceMock
class	IEngine
	interface of core Allreduce engine More...
class	MPIEngine
	implementation of engine using MPI More...
struct	ThreadLocalEntry
	entry to to easily hold returning information More...

Typedefs
using	Manager = AllreduceBase
using	EngineThreadLocal = dmlc::ThreadLocalStore< ThreadLocalEntry >

Functions
bool	Init (int argc, char *argv[])
	initializes the engine module
bool	Finalize ()
	finalizes the engine module
IEngine *	GetEngine ()
	singleton method to get engine
void	Allgather (void *sendrecvbuf, size_t total_size, size_t slice_begin, size_t slice_end, size_t size_prev_slice)
	Allgather function, each node have a segment of data in the ring of sendrecvbuf, the data provided by current node k is [slice_begin, slice_end), the next node's segment must start with slice_end after the call of Allgather, sendrecvbuf_ contains all the contents including all segments use a ring based algorithm.
void	Allreduce_ (void *sendrecvbuf, size_t type_nbytes, size_t count, IEngine::ReduceFunction red, mpi::DataType dtype, mpi::OpType op, IEngine::PreprocFunction prepare_fun=nullptr, void *prepare_arg=nullptr)
	perform in-place Allreduce, on sendrecvbuf this is an internal function used by rabit to be able to compile with MPI do not use this function directly
size_t	ParseUnit (const char *name, const char *val)
MPI::Datatype	GetType (mpi::DataType dtype)
MPI::Op	GetOp (mpi::OpType otype)

Variables
MPIEngine	manager

Detailed Description

core interface of the engine

Function Documentation

Allgather()

void rabit::engine::Allgather	(	void *	sendrecvbuf,
		size_t	total_size,
		size_t	slice_begin,
		size_t	slice_end,
		size_t	size_prev_slice
	)

Allgather function, each node have a segment of data in the ring of sendrecvbuf, the data provided by current node k is [slice_begin, slice_end), the next node's segment must start with slice_end after the call of Allgather, sendrecvbuf_ contains all the contents including all segments use a ring based algorithm.

Parameters

sendrecvbuf	buffer for both sending and receiving data, it is a ring conceptually
total_size	total size of data to be gathered
slice_begin	beginning of the current slice
slice_end	end of the current slice
size_prev_slice	size of the previous slice i.e. slice of node (rank - 1) % world_size

Allreduce_()

void rabit::engine::Allreduce_	(	void *	sendrecvbuf,
		size_t	type_nbytes,
		size_t	count,
		IEngine::ReduceFunction	red,
		mpi::DataType	dtype,
		mpi::OpType	op,
		IEngine::PreprocFunction	prepare_fun = nullptr,
		void *	prepare_arg = nullptr
	)

perform in-place Allreduce, on sendrecvbuf this is an internal function used by rabit to be able to compile with MPI do not use this function directly

Parameters

sendrecvbuf	buffer for both sending and receiving data
type_nbytes	the number of bytes the type has
count	number of elements to be reduced
reducer	reduce function
dtype	the data type
op	the reduce operator type
prepare_func	Lazy preprocessing function, lazy prepare_fun(prepare_arg) will be called by the function before performing Allreduce, to initialize the data in sendrecvbuf_. If the result of Allreduce can be recovered directly, then prepare_func will NOT be called
prepare_arg	argument used to pass into the lazy preprocessing function.

Finalize()

bool rabit::engine::Finalize

(

void

)

finalizes the engine module

finalize syncrhonization module

Init()

bool rabit::engine::Init	(	int	argc,
		char *	argv[]
	)

initializes the engine module

initialize the synchronization module

intiialize the synchronization module