#include <RemoteDevice.h>

Inheritance diagram for executive::RemoteDevice:

Collaboration diagram for executive::RemoteDevice:

Classes
class	ConnectionManager
	A singleton to manage communication. More...
class	MemoryAllocation
	An interface to a managed memory allocation. More...
Public Types
typedef std::map< void , MemoryAllocation >	AllocationMap
	A map of memory allocations.
Public Member Functions
	RemoteDevice (unsigned int id, const PropertiesData &props, unsigned int flags=0)
	Sets the device properties, connects to an ocelot server.
	~RemoteDevice ()
	Clears all state.
Device::MemoryAllocation *	getMemoryAllocation (const void *address, AllocationType type) const
	Get the allocation containing a pointer or 0.
Device::MemoryAllocation *	getGlobalAllocation (const std::string &module, const std::string &name)
	Get the address of a global by stream.
Device::MemoryAllocation *	allocate (size_t size)
	Allocate some new dynamic memory on this device.
Device::MemoryAllocation *	allocateHost (size_t size, unsigned int flags)
	Make this a host memory allocation.
Device::MemoryAllocation *	registerHost (void *pointer, size_t size, unsigned int flags)
	Register a host memory allocation.
void	free (void *pointer)
	Free an existing non-global allocation.
MemoryAllocationVector	getNearbyAllocations (void *pointer) const
	Get nearby allocations to a pointer.
MemoryAllocationVector	getAllAllocations () const
	Get all allocations, host, global, and device.
void	clearMemory ()
	Wipe all memory allocations, but keep modules.
void *	glRegisterBuffer (unsigned int buffer, unsigned int flags)
	Registers an opengl buffer with a resource.
void *	glRegisterImage (unsigned int image, unsigned int target, unsigned int flags)
	Registers an opengl image with a resource.
void	unRegisterGraphicsResource (void *resource)
	Unregister a resource.
void	mapGraphicsResource (void **resource, int count, unsigned int stream)
	Map a graphics resource for use with this device.
void *	getPointerToMappedGraphicsResource (size_t &size, void *resource)
	Get a pointer to a mapped resource along with its size.
void	setGraphicsResourceFlags (void *resource, unsigned int flags)
	Change the flags of a mapped resource.
void	unmapGraphicsResource (void **resource, int count, unsigned int stream)
	Unmap a mapped resource.
void	load (const ir::Module *module)
	Load a module, must have a unique name.
void	unload (const std::string &name)
	Unload a module by name.
ExecutableKernel *	getKernel (const std::string &module, const std::string &kernel)
	Get a translated kernel from the device.
unsigned int	createEvent (int flags)
	Create a new event.
void	destroyEvent (unsigned int event)
	Destroy an existing event.
bool	queryEvent (unsigned int event)
	Query to see if an event has been recorded (yes/no)
void	recordEvent (unsigned int event, unsigned int stream)
	Record something happening on an event.
void	synchronizeEvent (unsigned int event)
	Synchronize on an event.
float	getEventTime (unsigned int start, unsigned int end)
	Get the elapsed time in ms between two recorded events.
unsigned int	createStream ()
	Create a new stream.
void	destroyStream (unsigned int stream)
	Destroy an existing stream.
bool	queryStream (unsigned int stream)
	Query the status of an existing stream (ready/not)
void	synchronizeStream (unsigned int stream)
	Synchronize a particular stream.
void	setStream (unsigned int stream)
	Sets the current stream.
void	select ()
	Select this device as the current device. Only one device is allowed to be selected at any time.
bool	selected () const
	is this device selected?
void	unselect ()
	Deselect this device.
void	bindTexture (void *pointer, const std::string &moduleName, const std::string &textureName, const textureReference &ref, const cudaChannelFormatDesc &desc, const ir::Dim3 &size)
	Binds a texture to a memory allocation at a pointer.
void	unbindTexture (const std::string &moduleName, const std::string &textureName)
	unbinds anything bound to a particular texture
void *	getTextureReference (const std::string &moduleName, const std::string &textureName)
	Get's a reference to an internal texture.
void	launch (const std::string &module, const std::string &kernel, const ir::Dim3 &grid, const ir::Dim3 &block, size_t sharedMemory, const void argumentBlock, size_t argumentBlockSize, const trace::TraceGeneratorVector &traceGenerators=trace::TraceGeneratorVector(), const ir::ExternalFunctionSet externals=0)
	helper function for launching a kernel
cudaFuncAttributes	getAttributes (const std::string &module, const std::string &kernel)
	Get the function attributes of a specific kernel.
unsigned int	getLastError ()
	Get the last error from this device.
void	synchronize ()
	Wait until all asynchronous operations have completed.
void	limitWorkerThreads (unsigned int threads)
	Limit the worker threads used by this device.
void	setOptimizationLevel (translator::Translator::OptimizationLevel level)
	Set the optimization level for kernels in this device.
Static Public Member Functions
static DeviceVector	createDevices (unsigned int flags, int computeCapability)
	Allocate a new device for each remote device.
static unsigned int	deviceCount (int computeCapability)
	Determine the number of CUDA GPUs in the system.
Static Public Attributes
static ConnectionManager	connectionManager
Friends
class	RemoteDevice::MemoryAllocation

Detailed Description

class defining a remote Ocelot device

Member Typedef Documentation

typedef std::map<void*, MemoryAllocation*> executive::RemoteDevice::AllocationMap

A map of memory allocations.

Constructor & Destructor Documentation

executive::RemoteDevice::RemoteDevice	(	unsigned int	id,
		const PropertiesData &	props,
		unsigned int	flags = `0`
	)

Sets the device properties, connects to an ocelot server.

Sets the device properties, bind this to the cuda id.

executive::RemoteDevice::~RemoteDevice ( )

Clears all state.

Member Function Documentation

executive::Device::MemoryAllocation * executive::RemoteDevice::allocate ( size_t size ) [virtual]

Allocate some new dynamic memory on this device.

Implements executive::Device.

executive::Device::MemoryAllocation * executive::RemoteDevice::allocateHost	(	size_t	size,
		unsigned int	flags
	)		`[virtual]`

Make this a host memory allocation.

Implements executive::Device.

void executive::RemoteDevice::bindTexture	(	void *	pointer,
		const std::string &	moduleName,
		const std::string &	textureName,
		const textureReference &	ref,
		const cudaChannelFormatDesc &	desc,
		const ir::Dim3 &	size
	)		`[virtual]`

Binds a texture to a memory allocation at a pointer.

Implements executive::Device.

void executive::RemoteDevice::clearMemory ( ) [virtual]

Wipe all memory allocations, but keep modules.

Implements executive::Device.

executive::Device::DeviceVector executive::RemoteDevice::createDevices	(	unsigned int	flags,
		int	computeCapability
	)		`[static]`

Allocate a new device for each remote device.

unsigned int executive::RemoteDevice::createEvent ( int flags ) [virtual]

Create a new event.

Implements executive::Device.

unsigned int executive::RemoteDevice::createStream ( ) [virtual]

Create a new stream.

Implements executive::Device.

void executive::RemoteDevice::destroyEvent ( unsigned int event ) [virtual]

Destroy an existing event.

Implements executive::Device.

void executive::RemoteDevice::destroyStream ( unsigned int stream ) [virtual]

Destroy an existing stream.

Implements executive::Device.

unsigned int executive::RemoteDevice::deviceCount ( int computeCapability ) [static]

Determine the number of CUDA GPUs in the system.

Determine the number of remote GPUs in the system.

void executive::RemoteDevice::free ( void * pointer ) [virtual]

Free an existing non-global allocation.

Implements executive::Device.

executive::Device::MemoryAllocationVector executive::RemoteDevice::getAllAllocations ( ) const [virtual]

Get all allocations, host, global, and device.

Implements executive::Device.

cudaFuncAttributes executive::RemoteDevice::getAttributes	(	const std::string &	module,
		const std::string &	kernel
	)		`[virtual]`

Get the function attributes of a specific kernel.

Implements executive::Device.

float executive::RemoteDevice::getEventTime	(	unsigned int	start,
		unsigned int	end
	)		`[virtual]`

Get the elapsed time in ms between two recorded events.

Implements executive::Device.

executive::Device::MemoryAllocation * executive::RemoteDevice::getGlobalAllocation	(	const std::string &	module,
		const std::string &	name
	)		`[virtual]`

Get the address of a global by stream.

Implements executive::Device.

executive::ExecutableKernel * executive::RemoteDevice::getKernel	(	const std::string &	module,
		const std::string &	kernel
	)		`[virtual]`

Get a translated kernel from the device.

Implements executive::Device.

unsigned int executive::RemoteDevice::getLastError ( ) [virtual]

Get the last error from this device.

Implements executive::Device.

executive::Device::MemoryAllocation * executive::RemoteDevice::getMemoryAllocation	(	const void *	address,
		AllocationType	type
	)		const `[virtual]`

Get the allocation containing a pointer or 0.

Implements executive::Device.

executive::Device::MemoryAllocationVector executive::RemoteDevice::getNearbyAllocations ( void * pointer ) const [virtual]

Get nearby allocations to a pointer.

Implements executive::Device.

void * executive::RemoteDevice::getPointerToMappedGraphicsResource	(	size_t &	size,
		void *	resource
	)		`[virtual]`

Get a pointer to a mapped resource along with its size.

Implements executive::Device.

void * executive::RemoteDevice::getTextureReference	(	const std::string &	moduleName,
		const std::string &	textureName
	)		`[virtual]`

Get's a reference to an internal texture.

Implements executive::Device.

void * executive::RemoteDevice::glRegisterBuffer	(	unsigned int	buffer,
		unsigned int	flags
	)		`[virtual]`

Registers an opengl buffer with a resource.

Implements executive::Device.

void * executive::RemoteDevice::glRegisterImage	(	unsigned int	image,
		unsigned int	target,
		unsigned int	flags
	)		`[virtual]`

Registers an opengl image with a resource.

Implements executive::Device.

void executive::RemoteDevice::launch	(	const std::string &	module,
		const std::string &	kernel,
		const ir::Dim3 &	grid,
		const ir::Dim3 &	block,
		size_t	sharedMemory,
		const void *	argumentBlock,
		size_t	argumentBlockSize,
		const trace::TraceGeneratorVector &	traceGenerators = `trace::TraceGeneratorVector()`,
		const ir::ExternalFunctionSet *	externals = `0`
	)		`[virtual]`

helper function for launching a kernel

Parameters:

module	module name
kernel	kernel name
grid	grid dimensions
block	block dimensions
sharedMemory	shared memory size
argumentBlock	array of bytes for parameter memory
argumentBlockSize	number of bytes in parameter memory
traceGenerators	vector of trace generators to add and remove from kernel