One operation-tail completion shared by multiple CUDA buffer accesses. More...

#include <buffer.hpp>

Public Member Functions
	AccessCompletion (AccessCompletion &&) noexcept=default

	AccessCompletion (AccessCompletion const &)=delete

	AccessCompletion (Stream const &stream) noexcept
	Record the current stream tail, synchronizing only if publication fails.

AccessCompletion &	operator= (AccessCompletion &&) noexcept=default

AccessCompletion &	operator= (AccessCompletion const &)=delete

template<class Access >
void	release (Access &access) const noexcept
	Release one participating access through the shared completion.

Detailed Description

One operation-tail completion shared by multiple CUDA buffer accesses.

Record this after successfully submitting all work for one operation, then release every participating access through it. If event publication fails, construction synchronizes the stream and subsequent releases use the already-synchronized path.

Constructor & Destructor Documentation

◆ AccessCompletion() [1/3]

uni20::cuda::AccessCompletion::AccessCompletion ( Stream const & stream )

inlineexplicitnoexcept

Record the current stream tail, synchronizing only if publication fails.

◆ AccessCompletion() [2/3]

uni20::cuda::AccessCompletion::AccessCompletion ( AccessCompletion const & )

delete

◆ AccessCompletion() [3/3]

uni20::cuda::AccessCompletion::AccessCompletion ( AccessCompletion && )

defaultnoexcept

Member Function Documentation

◆ operator=() [1/2]

AccessCompletion & uni20::cuda::AccessCompletion::operator= ( AccessCompletion && )

defaultnoexcept

◆ operator=() [2/2]

AccessCompletion & uni20::cuda::AccessCompletion::operator= ( AccessCompletion const & )

delete

◆ release()

template<class Access >

void uni20::cuda::AccessCompletion::release ( Access & access ) const

inlinenoexcept

Release one participating access through the shared completion.

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

buffer.hpp

Public Member Functions