Implementation class for RmmResourceAdaptor. More...

#include <rmm_resource_adaptor_impl.hpp>

Public Member Functions
	RmmResourceAdaptorImpl (cuda::mr::any_resource< cuda::mr::device_accessible > primary_mr, std::optional< cuda::mr::any_resource< cuda::mr::device_accessible >> fallback_mr)
	Construct with primary and optional fallback memory resource. More...

	RmmResourceAdaptorImpl (RmmResourceAdaptorImpl const &)=delete

	RmmResourceAdaptorImpl (RmmResourceAdaptorImpl &&)=delete

RmmResourceAdaptorImpl &	operator= (RmmResourceAdaptorImpl const &)=delete

RmmResourceAdaptorImpl &	operator= (RmmResourceAdaptorImpl &&)=delete

bool	operator== (RmmResourceAdaptorImpl const &other) const noexcept
	Equality comparison (identity-based). More...

rmm::device_async_resource_ref	get_upstream_resource () const noexcept
	Get a reference to the primary upstream resource. More...

std::optional< rmm::device_async_resource_ref >	get_fallback_resource () const noexcept
	Get a reference to the fallback upstream resource. More...

ScopedMemoryRecord	get_main_record () const
	Returns a copy of the main memory record. More...

std::int64_t	current_allocated () const noexcept
	Get the total current allocated memory from both primary and fallback. More...

void	begin_scoped_memory_record ()
	Begin recording a new scoped memory usage record for the current thread. More...

ScopedMemoryRecord	end_scoped_memory_record ()
	End the current scoped memory record and return it. More...

void *	allocate (cuda::stream_ref stream, std::size_t bytes, std::size_t alignment=rmm::CUDA_ALLOCATION_ALIGNMENT)
	Allocate memory asynchronously on the given stream. More...

void	deallocate (cuda::stream_ref stream, void *ptr, std::size_t bytes, std::size_t alignment=rmm::CUDA_ALLOCATION_ALIGNMENT) noexcept
	Deallocate memory asynchronously on the given stream. More...

void *	allocate_sync (std::size_t bytes, std::size_t alignment=rmm::CUDA_ALLOCATION_ALIGNMENT)
	Allocate memory synchronously. More...

void	deallocate_sync (void *ptr, std::size_t bytes, std::size_t alignment=rmm::CUDA_ALLOCATION_ALIGNMENT) noexcept
	Deallocate memory synchronously. More...

Friends
void	get_property (RmmResourceAdaptorImpl const &, cuda::mr::device_accessible) noexcept
	Tag this resource as device-accessible for the CCCL concept.

Detailed Description

Implementation class for RmmResourceAdaptor.

Holds all mutable state for memory tracking and fallback allocation. This class satisfies the CCCL cuda::mr::resource concept and is held by RmmResourceAdaptor via cuda::mr::shared_resource for reference-counted ownership.

Definition at line 35 of file rmm_resource_adaptor_impl.hpp.

Constructor & Destructor Documentation

◆ RmmResourceAdaptorImpl()

rapidsmpf::detail::RmmResourceAdaptorImpl::RmmResourceAdaptorImpl	(	cuda::mr::any_resource< cuda::mr::device_accessible >	primary_mr,
		std::optional< cuda::mr::any_resource< cuda::mr::device_accessible >>	fallback_mr
	)

Construct with primary and optional fallback memory resource.

Parameters

primary_mr	The primary memory resource.
fallback_mr	Optional fallback memory resource.

Member Function Documentation

◆ allocate()

void* rapidsmpf::detail::RmmResourceAdaptorImpl::allocate	(	cuda::stream_ref	stream,
		std::size_t	bytes,
		std::size_t	alignment = `rmm::CUDA_ALLOCATION_ALIGNMENT`
	)

Allocate memory asynchronously on the given stream.

Parameters

stream	The CUDA stream for the allocation.
bytes	Number of bytes to allocate.
alignment	Alignment requirement.

Returns: Pointer to the allocated memory.

◆ allocate_sync()

void* rapidsmpf::detail::RmmResourceAdaptorImpl::allocate_sync	(	std::size_t	bytes,
		std::size_t	alignment = `rmm::CUDA_ALLOCATION_ALIGNMENT`
	)

Allocate memory synchronously.

Parameters

bytes	Number of bytes to allocate.
alignment	Alignment requirement.

Returns: Pointer to the allocated memory.

◆ begin_scoped_memory_record()

void rapidsmpf::detail::RmmResourceAdaptorImpl::begin_scoped_memory_record ( )

Begin recording a new scoped memory usage record for the current thread.

This method pushes a new empty ScopedMemoryRecord onto the thread-local record stack, allowing for nested memory tracking scopes.

Must be paired with a matching call to end_scoped_memory_record().

See also: end_scoped_memory_record()

◆ current_allocated()

std::int64_t rapidsmpf::detail::RmmResourceAdaptorImpl::current_allocated ( ) const

noexcept

Get the total current allocated memory from both primary and fallback.

Returns: Total number of currently allocated bytes.

◆ deallocate()

void rapidsmpf::detail::RmmResourceAdaptorImpl::deallocate	(	cuda::stream_ref	stream,
		void *	ptr,
		std::size_t	bytes,
		std::size_t	alignment = `rmm::CUDA_ALLOCATION_ALIGNMENT`
	)

noexcept

Deallocate memory asynchronously on the given stream.

Parameters

stream	The CUDA stream for the deallocation.
ptr	Pointer to the memory to deallocate.
bytes	Number of bytes to deallocate.
alignment	Alignment of the original allocation.

◆ deallocate_sync()

void rapidsmpf::detail::RmmResourceAdaptorImpl::deallocate_sync	(	void *	ptr,
		std::size_t	bytes,
		std::size_t	alignment = `rmm::CUDA_ALLOCATION_ALIGNMENT`
	)

noexcept

Deallocate memory synchronously.

Parameters

ptr	Pointer to the memory to deallocate.
bytes	Number of bytes to deallocate.
alignment	Alignment of the original allocation.

◆ end_scoped_memory_record()

ScopedMemoryRecord rapidsmpf::detail::RmmResourceAdaptorImpl::end_scoped_memory_record ( )

End the current scoped memory record and return it.

Pops the top ScopedMemoryRecord from the thread-local stack and returns it. If this scope was nested within another (i.e. if begin_scoped_memory_record() was called multiple times in a row), the returned scope is automatically added as a subscope to the next scope remaining on the stack.

This allows nesting of scoped memory tracking, where each scope can contain one or more subscopes. When analyzing or reporting memory statistics, the memory usage of each scope can be calculated inclusive of its subscopes. This behavior mimics standard hierarchical memory profilers, where the total memory attributed to a scope includes all allocations made within it, plus those made in its nested regions.

Returns: The scope that was just ended.

Exceptions

std::out_of_range if called without a matching begin_scoped_memory_record().

See also: begin_scoped_memory_record()

◆ get_fallback_resource()

std::optional<rmm::device_async_resource_ref> rapidsmpf::detail::RmmResourceAdaptorImpl::get_fallback_resource ( ) const

noexcept

Get a reference to the fallback upstream resource.

This resource is used if the primary resource throws rmm::out_of_memory.

Returns: Optional reference to the fallback RMM memory resource.

◆ get_main_record()

ScopedMemoryRecord rapidsmpf::detail::RmmResourceAdaptorImpl::get_main_record ( ) const

Returns a copy of the main memory record.

The main record tracks memory statistics for the lifetime of the resource.

Returns: A copy of the current main memory record.

◆ get_upstream_resource()

rmm::device_async_resource_ref rapidsmpf::detail::RmmResourceAdaptorImpl::get_upstream_resource ( ) const

noexcept

Get a reference to the primary upstream resource.

Returns: Reference to the RMM memory resource.

◆ operator==()

bool rapidsmpf::detail::RmmResourceAdaptorImpl::operator== ( RmmResourceAdaptorImpl const & other ) const

noexcept

Equality comparison (identity-based).

Parameters

other The other impl to compare.

Returns: True if this and other are the same object.

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

/__w/rapidsmpf/rapidsmpf/cpp/include/rapidsmpf/detail/rmm_resource_adaptor_impl.hpp

Public Member Functions

Friends

Detailed Description

Constructor & Destructor Documentation

◆ RmmResourceAdaptorImpl()

Member Function Documentation

◆ allocate()

◆ allocate_sync()

◆ begin_scoped_memory_record()

◆ current_allocated()

◆ deallocate()

◆ deallocate_sync()

◆ end_scoped_memory_record()

◆ get_fallback_resource()

◆ get_main_record()

◆ get_upstream_resource()

◆ operator==()