cuda::event Namespace Reference

CUDA timing functionality, via events and their related code (not including the event wrapper type event_t itself) More...

Namespaces
	ipc
	Definitions and functionality related to CUDA events (not including the event wrapper type event_t itself)

Typedefs
using	duration_t = ::std::chrono::duration< float, ::std::milli >
	The type used by the CUDA Runtime API to represent the time difference between pairs of events.
using	handle_t = CUevent
	The CUDA driver's raw handle for events.

Enumerations
enum	: bool {   sync_by_busy_waiting = false,   sync_by_blocking = true }
	Synchronization option for cuda::event_t 's. More...
enum	: bool {   dont_record_timings = false,   do_record_timings = true }
	Should the CUDA Runtime API record timing information for events as it schedules them?
enum	: bool {   not_interprocess = false,   interprocess = true,   single_process = not_interprocess }
	IPC usability option for {cuda::event_t}'s. More...

Functions
event_t	wrap (device::id_t device_id, context::handle_t context_handle, handle_t event_handle, bool take_ownership=false, bool hold_pc_refcount_unit=false) noexcept
	Wrap an existing CUDA event in a event_t instance. More...
::std::string	identify (const event_t &event)
duration_t	time_elapsed_between (const event_t &start, const event_t &end)
	Determine (inaccurately) the elapsed time between two events. More...
duration_t	time_elapsed_between (const ::std::pair< const event_t &, const event_t &> &event_pair)
event_t	create (const device_t &device, bool uses_blocking_sync=sync_by_busy_waiting, bool records_timing=do_record_timings, bool interprocess=not_interprocess)
	creates a new event on (the primary execution context of) a device. More...
event_t	create (const context_t &context, bool uses_blocking_sync=sync_by_busy_waiting, bool records_timing=do_record_timings, bool interprocess=not_interprocess)
	creates a new event. More...

Detailed Description

CUDA timing functionality, via events and their related code (not including the event wrapper type event_t itself)

Enumeration Type Documentation

anonymous enum

anonymous enum : bool

Synchronization option for cuda::event_t 's.

Enumerator
sync_by_busy_waiting	The thread calling event_.synchronize() will enter a busy-wait loop; this (might) minimize delay between kernel execution conclusion and control returning to the thread, but is very wasteful of CPU time.
sync_by_blocking	The thread calling event_.synchronize() will block - yield control of the CPU and will only become ready for execution after the kernel has completed its execution - at which point it would have to wait its turn among other threads. This does not waste CPU computing time, but results in a longer delay.

anonymous enum

anonymous enum : bool

IPC usability option for {cuda::event_t}'s.

Enumerator
not_interprocess	Can only be used by the process which created it.
interprocess	Can be shared between processes. Must not be able to record timings.

Function Documentation

create() [1/2]

event_t cuda::event::create ( const device_t &  device, bool  uses_blocking_sync = sync_by_busy_waiting, bool  records_timing = do_record_timings, bool  interprocess = not_interprocess  )

inline

creates a new event on (the primary execution context of) a device.

Parameters

device	The device on which to create the new stream
uses_blocking_sync	When synchronizing on this new event, shall a thread busy-wait for it, or block?
records_timing	Can this event be used to record time values (e.g. duration between events)?
interprocess	Can multiple processes work with the constructed event?

Returns

The constructed event proxy

Note

The created event will keep the device's primary context active while it exists.

create() [2/2]

event_t cuda::event::create ( const context_t &  context, bool  uses_blocking_sync = sync_by_busy_waiting, bool  records_timing = do_record_timings, bool  interprocess = not_interprocess  )

inline

creates a new event.

Parameters

context	The CUDA execution context in which to create the event
uses_blocking_sync	When synchronizing on this new event, shall a thread busy-wait for it, or block?
records_timing	Can this event be used to record time values (e.g. duration between events)?
interprocess	Can multiple processes work with the constructed event?

Returns

The constructed event proxy

Note

Even if the context happens to be primary, the created event will not keep this context alive.

time_elapsed_between()

duration_t cuda::event::time_elapsed_between ( const event_t &  start, const event_t &  end  )

inline

Determine (inaccurately) the elapsed time between two events.

Note

Q: Why the weird output type? A: This is what the CUDA Runtime API itself returns

Parameters

start	first timepoint event
end	second, later, timepoint event

Returns

the difference in the (inaccurately) measured time, in msec

wrap()

event_t cuda::event::wrap ( device::id_t  device_id, context::handle_t  context_handle, handle_t  event_handle, bool  take_ownership = false, bool  hold_pc_refcount_unit = false  )

inlinenoexcept

Wrap an existing CUDA event in a event_t instance.

Note

This is a named constructor idiom, existing of direct access to the ctor of the same signature, to emphasize that a new event is not created.

Parameters

device_id	Index of the device to which the event relates
context_handle	Handle of the context in which this event was created
event_handle	handle of the pre-existing event
take_ownership	When set to false, the CUDA event will not be destroyed along with proxy; use this setting when temporarily working with a stream existing irrespective of the current context and outlasting it. When set to true, the proxy class will act as it does usually, destroying the event when being destructed itself.
hold_pc_refcount_unit	when the event's context is a device's primary context, this controls whether that context must be kept active while the event continues to exist.

Returns

an event wrapper associated with the specified event