Realm::Cuda::GPUreduceXferDes Class Reference

#include <cuda_internal.h>

Inheritance diagram for Realm::Cuda::GPUreduceXferDes:

Collaboration diagram for Realm::Cuda::GPUreduceXferDes:

Public Member Functions
	GPUreduceXferDes (uintptr_t _dma_op, Channel *_channel, NodeID _launch_node, XferDesID _guid, const std::vector< XferDesPortInfo > &inputs_info, const std::vector< XferDesPortInfo > &outputs_info, int _priority, XferDesRedopInfo _redop_info)
long	get_requests (Request **requests, long nr)
bool	progress_xd (GPUreduceChannel *channel, TimeLimit work_until)
bool	fast_reduction_kernel_mode (GPUreduceChannel *channel, const size_t max_bytes, XferPort *in_port, XferPort *out_port, const size_t in_span_start, const size_t out_span_start)
void	setup_redop_kernel (GPUreduceChannel *channel, void *params, const size_t in_span_start, const size_t out_span_start, const size_t in_elem_size, const size_t out_elem_size, const size_t elems, const bool has_transpose)
void	record_redop_advanced_kernel (GPU *gpu)
KernelVariantDesc	describe_kernel_variant (GPU *cpu, bool is_advanced)
bool	resolve_kernel_slot (GPU *gpu, void *host_proxy, CUfunction &kernel_out, CUfunction GPU::GPUReductionOpEntry::*cache_field)
Public Member Functions inherited from Realm::XferDes
	REALM_ALIGNED_TYPE_CONST (AlignedStorage, UnalignedStorage, 16)
	REALM_PMTA_DEFN (XferDes, IntrusivePriorityListLink< XferDes >, xd_link)
	REALM_PMTA_DEFN (XferDes, int, priority)
	XferDes (uintptr_t _dma_op, Channel *_channel, NodeID _launch_node, XferDesID _guid, const std::vector< XferDesPortInfo > &inputs_info, const std::vector< XferDesPortInfo > &outputs_info, int _priority, const void *_fill_data, size_t fill_size)
void	add_reference (void)
void	remove_reference (void)
void	add_update_pre_bytes_total_received (void)
virtual Event	request_metadata ()
virtual void	notify_request_read_done (Request *req)
virtual void	notify_request_write_done (Request *req)
virtual void	flush ()
long	default_get_requests (Request **requests, long nr, unsigned flags=0)
void	default_notify_request_read_done (Request *req)
void	default_notify_request_write_done (Request *req)
virtual void	update_bytes_read (int port_idx, size_t offset, size_t size)
virtual void	update_bytes_write (int port_idx, size_t offset, size_t size)
void	update_pre_bytes_write (int port_idx, size_t offset, size_t size)
void	update_pre_bytes_total (int port_idx, size_t pre_bytes_total)
void	update_next_bytes_read (int port_idx, size_t offset, size_t size)
void	begin_completion ()
void	mark_completed ()
unsigned	current_progress (void)
bool	check_for_progress (unsigned last_counter)
void	update_progress (void)
virtual bool	request_available ()
virtual Request *	dequeue_request ()
virtual void	enqueue_request (Request *req)
size_t	update_control_info (ReadSequenceCache *rseqcache)
size_t	get_addresses (size_t min_xfer_size, ReadSequenceCache *rseqcache)
size_t	get_addresses (size_t min_xfer_size, ReadSequenceCache *rseqcache, const InstanceLayoutPieceBase *&in_nonaffine, const InstanceLayoutPieceBase *&out_nonaffine)
bool	record_address_consumption (size_t total_read_bytes, size_t total_write_bytes)
void	replicate_fill_data (size_t new_size)

Protected Attributes
XferDesRedopInfo	redop_info
const ReductionOpUntyped *	redop
CUfunction	kernel
CUfunction	kernel_advanced
CUfunction	kernel_transpose
const void *	kernel_host_proxy
const void *	kernel_host_proxy_advanced
const void *	kernel_host_proxy_transpose
GPUStream *	stream
std::vector< GPU * >	src_gpus
std::vector< bool >	src_is_ipc
Protected Attributes inherited from Realm::XferDes
Mutex	available_req_mutex
std::queue< Request * >	available_reqs

Additional Inherited Members
Public Types inherited from Realm::XferDes
enum	{ XFERDES_NO_GUID = 0 }
typedef IntrusivePriorityList< XferDes, int, REALM_PMTA_USE(XferDes, xd_link), REALM_PMTA_USE(XferDes, priority), DummyLock >	XferDesList
typedef SequenceCache<&XferDes::update_bytes_read >	ReadSequenceCache
typedef SequenceCache<&XferDes::update_bytes_write >	WriteSequenceCache
Public Attributes inherited from Realm::XferDes
uintptr_t	dma_op
XferDesQueue *	xferDes_queue
NodeID	launch_node
atomic< bool >	iteration_completed
atomic< int64_t >	bytes_write_pending
atomic< bool >	transfer_completed
uint64_t	current_in_port_mask
uint64_t	current_out_port_mask
uint64_t	current_in_port_remain
uint64_t	current_out_port_remain
std::vector< XferPort >	input_ports
std::vector< XferPort >	output_ports
ControlPortState	input_control
ControlPortState	output_control
uint64_t	max_req_size
int	priority
XferDesID	guid
XferDesKind	kind
Channel *	channel
void *	fill_data
size_t	fill_size
size_t	orig_fill_size
AlignedStorage	inline_fill_storage
Mutex	xd_lock
Mutex	update_read_lock
Mutex	update_write_lock
atomic< unsigned >	progress_counter
atomic< unsigned >	reference_count
unsigned	nb_update_pre_bytes_total_calls_expected
atomic< unsigned >	nb_update_pre_bytes_total_calls_received
IntrusivePriorityListLink< XferDes >	xd_link
DeferredXDEnqueue	deferred_enqueue
Static Public Attributes inherited from Realm::XferDes
static const size_t	ALIGNED_FILL_STORAGE_SIZE = 32
Protected Member Functions inherited from Realm::XferDes
virtual	~XferDes ()

Constructor & Destructor Documentation

GPUreduceXferDes()

Realm::Cuda::GPUreduceXferDes::GPUreduceXferDes	(	uintptr_t	_dma_op,
		Channel *	_channel,
		NodeID	_launch_node,
		XferDesID	_guid,
		const std::vector< XferDesPortInfo > &	inputs_info,
		const std::vector< XferDesPortInfo > &	outputs_info,
		int	_priority,
		XferDesRedopInfo	_redop_info
	)

Member Function Documentation

describe_kernel_variant()

KernelVariantDesc Realm::Cuda::GPUreduceXferDes::describe_kernel_variant	(	GPU *	cpu,
		bool	is_advanced
	)

fast_reduction_kernel_mode()

bool Realm::Cuda::GPUreduceXferDes::fast_reduction_kernel_mode	(	GPUreduceChannel *	channel,
		const size_t	max_bytes,
		XferPort *	in_port,
		XferPort *	out_port,
		const size_t	in_span_start,
		const size_t	out_span_start
	)

get_requests()

long Realm::Cuda::GPUreduceXferDes::get_requests ( Request **  requests, long  nr  )

virtual

Implements Realm::XferDes.

progress_xd()

bool Realm::Cuda::GPUreduceXferDes::progress_xd	(	GPUreduceChannel *	channel,
		TimeLimit	work_until
	)

record_redop_advanced_kernel()

void Realm::Cuda::GPUreduceXferDes::record_redop_advanced_kernel

(

GPU *

gpu

)

resolve_kernel_slot()

bool Realm::Cuda::GPUreduceXferDes::resolve_kernel_slot	(	GPU *	gpu,
		void *	host_proxy,
		CUfunction &	kernel_out,
		CUfunction GPU::GPUReductionOpEntry::*	cache_field
	)

setup_redop_kernel()

void Realm::Cuda::GPUreduceXferDes::setup_redop_kernel	(	GPUreduceChannel *	channel,
		void *	params,
		const size_t	in_span_start,
		const size_t	out_span_start,
		const size_t	in_elem_size,
		const size_t	out_elem_size,
		const size_t	elems,
		const bool	has_transpose
	)

Member Data Documentation

kernel

CUfunction Realm::Cuda::GPUreduceXferDes::kernel

protected

kernel_advanced

CUfunction Realm::Cuda::GPUreduceXferDes::kernel_advanced

protected

kernel_host_proxy

const void* Realm::Cuda::GPUreduceXferDes::kernel_host_proxy

protected

kernel_host_proxy_advanced

const void* Realm::Cuda::GPUreduceXferDes::kernel_host_proxy_advanced

protected

kernel_host_proxy_transpose

const void* Realm::Cuda::GPUreduceXferDes::kernel_host_proxy_transpose

protected

kernel_transpose

CUfunction Realm::Cuda::GPUreduceXferDes::kernel_transpose

protected

redop

const ReductionOpUntyped* Realm::Cuda::GPUreduceXferDes::redop

protected

redop_info

XferDesRedopInfo Realm::Cuda::GPUreduceXferDes::redop_info

protected

src_gpus

std::vector<GPU *> Realm::Cuda::GPUreduceXferDes::src_gpus

protected

src_is_ipc

std::vector<bool> Realm::Cuda::GPUreduceXferDes::src_is_ipc

protected

stream

GPUStream* Realm::Cuda::GPUreduceXferDes::stream

protected

---

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

• /home/runner/work/realm/realm/realm-src/src/realm/cuda/cuda_internal.h