The MemoryGrp abstract class provides a way of accessing distributed memory in a parallel machine. More...

#include <util/group/memory.h>

Inheritance diagram for sc::MemoryGrp:

Public Member Functions
	MemoryGrp (const Ref< KeyVal > &)

int	me () const
	Returns who I am.

int	n () const
	Returns how many nodes there are.

virtual void	set_localsize (size_t)=0
	Set the size of locally held memory. More...

size_t	localsize ()
	Returns the amount of memory residing locally on me().

virtual void *	localdata ()=0
	Returns a pointer to the local data.

distsize_t	localoffset ()
	Returns the global offset to this node's memory.

int	size (int node)
	Returns the amount of memory residing on node.

distsize_t	offset (int node)
	Returns the global offset to node's memory.

distsize_t	totalsize ()
	Returns the sum of all memory allocated on all nodes.

virtual void	activate ()
	Activate is called before the memory is to be used.

virtual void	deactivate ()
	Deactivate is called after the memory has been used.

virtual void *	obtain_writeonly (distsize_t offset, size_t size)=0
	This gives write access to the memory location. No locking is done.

virtual void *	obtain_readwrite (distsize_t offset, size_t size)=0
	Only one thread can have an unreleased obtain_readwrite at a time. More...

virtual void *	obtain_readonly (distsize_t offset, size_t size)=0
	This gives read access to the memory location. No locking is done.

virtual void	release_readonly (void *data, distsize_t offset, size_t size)=0
	This is called when read access is no longer needed.

virtual void	release_writeonly (void *data, distsize_t offset, size_t size)=0
	This is called when write access is no longer needed.

virtual void	release_readwrite (void *data, distsize_t offset, size_t size)=0
	This is called when read/write access is no longer needed. More...

virtual void	write (const void *data, distsize_t offset, size_t size)
	This is used to write data directly. More...

virtual void	sum_reduction (double *data, distsize_t doffset, size_t dsize)
	Perform a sum reduction on double data. More...

virtual void	sum_reduction_on_node (double *data, size_t doffset, size_t dsize, int node=-1)
	Perform a sum reduction on double data localized to a single node. More...

virtual void	sync ()=0
	Synchronizes all the nodes. More...

virtual void *	malloc_local (size_t nbyte)
	Allocate data that will be accessed locally only. More...

double *	malloc_local_double (size_t ndouble)
	Allocate double data that will be accessed locally only. More...

virtual void	free_local (void *&data)
	Free data that was allocated with malloc_local, and set the pointer to 0.

void	free_local_double (double *&data)
	Free data that was allocated with malloc_local_double, and set the pointer to 0

virtual void	catchup ()
	Processes outstanding requests. More...

virtual Ref< MemoryGrp >	clone (void)
	Returns a copy of this MemoryGrp specialization that provides an independent communication context. More...

virtual void	print (std::ostream &o=ExEnv::out0()) const
	Prints out information about the object.

Public Member Functions inherited from sc::DescribedClass
	DescribedClass (const DescribedClass &)

DescribedClass &	operator= (const DescribedClass &)

ClassDesc *	class_desc () const MPQC__NOEXCEPT
	This returns the unique pointer to the ClassDesc corresponding to the given type_info object. More...

const char *	class_name () const
	Return the name of the object's exact type.

int	class_version () const
	Return the version of the class.

Ref< DescribedClass >	ref ()
	Return this object wrapped up in a Ref smart pointer. More...

Public Member Functions inherited from sc::RefCount
size_t	identifier () const
	Return the unique identifier for this object that can be compared for different objects of different types. More...

int	lock_ptr () const
	Lock this object.

int	unlock_ptr () const
	Unlock this object.

void	use_locks (bool inVal)
	start and stop using locks on this object

refcount_t	nreference () const
	Return the reference count.

refcount_t	reference ()
	Increment the reference count and return the new count.

refcount_t	dereference ()
	Decrement the reference count and return the new count.

int	managed () const

void	unmanage ()
	Turn off the reference counting mechanism for this object. More...

Static Public Member Functions
static MemoryGrp *	initial_memorygrp (int &argc, char **argv)
	Create a memory group. More...

static MemoryGrp *	initial_memorygrp ()
	Create a memory group. More...

static void	set_default_memorygrp (const Ref< MemoryGrp > &)
	The default memory group contains the primary memory group to be used by an application.

static MemoryGrp *	get_default_memorygrp ()
	Returns the default memory group. More...

Protected Member Functions
void	obtain_local_lock (size_t start, size_t fence)

void	release_local_lock (size_t start, size_t fence)

Protected Member Functions inherited from sc::RefCount
	RefCount (const RefCount &)

RefCount &	operator= (const RefCount &)

Protected Attributes
int	me_

int	n_

distsize_t *	offsets_

int	debug_

Detailed Description

The MemoryGrp abstract class provides a way of accessing distributed memory in a parallel machine.

Several specializations are available. For one processor, ProcMemoryGrp provides a simple stub implementation. Parallel specializations include ShmMemoryGrp, MTMPIMemoryGrp, and ARMCIMemoryGrp. The particular specializations that work depend highly on the target hardware and software environment.

Member Function Documentation

◆ catchup()

virtual void sc::MemoryGrp::catchup ( )

virtual

Processes outstanding requests.

Some memory group implementations don't have access to real shared memory or even active messages. Instead, requests are processed whenever certain memory group routines are called. This can cause large latencies and buffer overflows. If this is a problem, then the catchup member can be called to process all outstanding requests.

Reimplemented in sc::MemoryGrpRegion.

◆ clone()

virtual Ref<MemoryGrp> sc::MemoryGrp::clone ( void )

virtual

Returns a copy of this MemoryGrp specialization that provides an independent communication context.

This will throw for specializations that do not reimplement it.

Reimplemented in sc::MTMPIMemoryGrp, and sc::MemoryGrpRegion.

◆ get_default_memorygrp()

static MemoryGrp* sc::MemoryGrp::get_default_memorygrp ( )

static

Returns the default memory group.

If the default memory group has not yet been set, then one is created. The particular specialization used is determined by configuration options and which specializations are being used for MessageGrp and ThreadGrp.

◆ initial_memorygrp() [1/2]

static MemoryGrp* sc::MemoryGrp::initial_memorygrp ( )

static

Create a memory group.

See also: initial_memorygrp(int&,char**)

◆ initial_memorygrp() [2/2]

static MemoryGrp* sc::MemoryGrp::initial_memorygrp	(	int &	argc,
		char **	argv
	)

static

Create a memory group.

This routine looks for a -memorygrp argument, and then the environmental variable MEMORYGRP to decide which specialization of MemoryGrp would be appropriate. The argument to -memorygrp or the value of the environmental variable should be either string for a ParsedKeyVal constructor or a classname. The default ThreadGrp and MessageGrp objects should be initialized before this is called.

◆ malloc_local()

virtual void* sc::MemoryGrp::malloc_local ( size_t nbyte )

virtual

Allocate data that will be accessed locally only.

Using this for data that will be used for global operations can improve efficiency. Data allocated in this way must be freed with free_local_double.

Reimplemented in sc::MemoryGrpRegion, and sc::ARMCIMemoryGrp.

◆ malloc_local_double()

double* sc::MemoryGrp::malloc_local_double ( size_t ndouble )

Allocate double data that will be accessed locally only.

See also: malloc_local

◆ obtain_readwrite()

virtual void* sc::MemoryGrp::obtain_readwrite	(	distsize_t	offset,
		size_t	size
	)

pure virtual

Only one thread can have an unreleased obtain_readwrite at a time.

The actual memory region locked can be larger than that requested. If the memory region is already locked this will block. For this reason, data should be held as read/write for as short a time as possible.

Implemented in sc::ActiveMsgMemoryGrp, sc::ShmMemoryGrp, sc::RDMAMemoryGrp, sc::MemoryGrpRegion, and sc::ProcMemoryGrp.

◆ release_readwrite()

virtual void sc::MemoryGrp::release_readwrite	(	void *	data,
		distsize_t	offset,
		size_t	size
	)

pure virtual

This is called when read/write access is no longer needed.

The memory will be unlocked.

Implemented in sc::ActiveMsgMemoryGrp, sc::ShmMemoryGrp, sc::RDMAMemoryGrp, sc::MemoryGrpRegion, and sc::ProcMemoryGrp.

◆ set_localsize()

virtual void sc::MemoryGrp::set_localsize ( size_t )

pure virtual

Set the size of locally held memory.

When memory is accessed using a global offset counting starts at node 0 and proceeds up to node n() - 1.

Implemented in sc::ActiveMsgMemoryGrp, sc::ShmMemoryGrp, sc::MemoryGrpRegion, sc::ARMCIMemoryGrp, sc::ProcMemoryGrp, sc::RDMAMemoryGrp, and sc::MsgMemoryGrp.

◆ sum_reduction()

virtual void sc::MemoryGrp::sum_reduction	(	double *	data,
		distsize_t	doffset,
		size_t	dsize
	)

virtual

Perform a sum reduction on double data.

Parameters

data	the contribution to sum into the global array.
doffset	the global offset in terms of doubles.
dsize	the size in terms of doubles.

Reimplemented in sc::ActiveMsgMemoryGrp, and sc::RDMAMemoryGrp.

◆ sum_reduction_on_node()

virtual void sc::MemoryGrp::sum_reduction_on_node	(	double *	data,
		size_t	doffset,
		size_t	dsize,
		int	node = `-1`
	)

virtual

Perform a sum reduction on double data localized to a single node.

Parameters

data	the contribution to sum into the global array.
doffset	the local offset on the node in terms of doubles.
dsize	the size in terms of doubles.
node	the node holding the data. The default is the local node.

Reimplemented in sc::ActiveMsgMemoryGrp, sc::RDMAMemoryGrp, and sc::MemoryGrpRegion.

◆ sync()

virtual void sc::MemoryGrp::sync ( )

pure virtual

Synchronizes all the nodes.

This is useful after remote memory writes to be certain that all of the writes have completed and the data can be accessed locally, for example.

Implemented in sc::MTMPIMemoryGrp, sc::MemoryGrpRegion, sc::ARMCIMemoryGrp, sc::ProcMemoryGrp, and sc::MsgMemoryGrp.

◆ write()

virtual void sc::MemoryGrp::write	(	const void *	data,
		distsize_t	offset,
		size_t	size
	)

virtual

This is used to write data directly.

The default implementation uses obtain_writeonly(). More efficient implementations will avoid copies

Reimplemented in sc::ActiveMsgMemoryGrp.

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

src/lib/util/group/memory.h

Public Member Functions

Static Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Member Function Documentation

◆ catchup()

◆ clone()

◆ get_default_memorygrp()

◆ initial_memorygrp() [1/2]

◆ initial_memorygrp() [2/2]

◆ malloc_local()

◆ malloc_local_double()

◆ obtain_readwrite()

◆ release_readwrite()

◆ set_localsize()

◆ sum_reduction()

◆ sum_reduction_on_node()

◆ sync()

◆ write()