TNL::Atomic< T, Devices::Cuda > Class Template Reference

Atomic wrapper for the CUDA/HIP device. More...

#include <TNL/Atomic.h>

Public Types
using	value_type = T
	Type of the atomic value.

Public Member Functions
__cuda_callable__	Atomic () noexcept=default
	Default constructor.
__cuda_callable__	Atomic (const Atomic &desired) noexcept
	Copy constructor.
__cuda_callable__ constexpr	Atomic (T desired) noexcept
	Constructor initializing the atomic value to desired.
__cuda_callable__ bool	compare_exchange_strong (T &expected, T desired) noexcept
	Strong compare-and-exchange operation.
__cuda_callable__ bool	compare_exchange_weak (T &expected, T desired) noexcept
	Weak compare-and-exchange operation.
__cuda_callable__ T	exchange (T desired) noexcept
	Atomically exchanges the stored value with desired.
__cuda_callable__ T	fetch_add (T arg)
	Atomically adds arg to the stored value and returns the old value.
__cuda_callable__ T	fetch_and (T arg)
	Atomically performs bitwise AND with arg and returns the old value.
__cuda_callable__ T	fetch_max (T arg) noexcept
	Atomically updates the stored value to the maximum of the current value and arg.
__cuda_callable__ T	fetch_min (T arg) noexcept
	Atomically updates the stored value to the minimum of the current value and arg.
__cuda_callable__ T	fetch_or (T arg)
	Atomically performs bitwise OR with arg and returns the old value.
__cuda_callable__ T	fetch_sub (T arg)
	Atomically subtracts arg from the stored value and returns the old value.
__cuda_callable__ T	fetch_xor (T arg)
	Atomically performs bitwise XOR with arg and returns the old value.
constexpr bool	is_always_lock_free () const noexcept
	Returns true if the atomic operations are always lock-free.
bool	is_lock_free () const noexcept
	Returns true if the atomic operations are lock-free.
__cuda_callable__ T	load () const noexcept
	Atomically loads the stored value.
__cuda_callable__	operator T () const noexcept
	Conversion operator returning the stored value.
__cuda_callable__ T	operator&= (T arg) noexcept
	Atomically performs bitwise AND with arg and returns the new value.
__cuda_callable__ T	operator++ () noexcept
	Pre-increment operator.
__cuda_callable__ T	operator++ (int) noexcept
	Post-increment operator.
__cuda_callable__ T	operator+= (T arg) noexcept
	Atomically adds arg to the stored value and returns the new value.
__cuda_callable__ T	operator-- () noexcept
	Pre-decrement operator.
__cuda_callable__ T	operator-- (int) noexcept
	Post-decrement operator.
__cuda_callable__ T	operator-= (T arg) noexcept
	Atomically subtracts arg from the stored value and returns the new value.
__cuda_callable__ Atomic &	operator= (const Atomic &desired) noexcept
	Copy assignment operator.
__cuda_callable__ T	operator= (T desired) noexcept
	Assigns desired to the atomic value.
__cuda_callable__ T	operator^= (T arg) noexcept
	Atomically performs bitwise XOR with arg and returns the new value.
__cuda_callable__ T	operator|= (T arg) noexcept
	Atomically performs bitwise OR with arg and returns the new value.
__cuda_callable__ void	store (T desired) noexcept
	Atomically replaces the stored value with desired.

Protected Attributes
T	value
	The stored atomic value.

Detailed Description

template<typename T>
class TNL::Atomic< T, Devices::Cuda >

Atomic wrapper for the CUDA/HIP device.

Implements the same interface as std::atomic<T> using CUDA/HIP built-in atomic operations. The copy operations are not atomic.

---

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

• src/TNL/Atomic.h