Allocator using a named Umpire allocator over a CUDA device resource. More...

Inheritance diagram for gko::UmpireCudaAllocator:

Collaboration diagram for gko::UmpireCudaAllocator:

Public Member Functions
	UmpireCudaAllocator (std::string name)

void *	allocate (size_type num_bytes) override

void	deallocate (void *ptr) override

Public Member Functions inherited from gko::UmpireAllocatorBase
const std::string &	get_name () const
	Name of the underlying Umpire allocator.

size_type	get_current_size () const
	Bytes currently allocated from the Umpire allocator and not yet deallocated, by any of its users (not only through this object).

size_type	get_actual_size () const
	Bytes the Umpire allocator holds from its memory resource. More...

size_type	get_high_watermark () const
	Largest value get_current_size() has reached so far.

size_type	get_allocation_count () const
	Number of allocations from the Umpire allocator that are still live.

void	release ()
	Returns the pool's completely unused blocks to its memory resource.

Detailed Description

Allocator using a named Umpire allocator over a CUDA device resource.

The pool is not stream-ordered; see UmpireAllocatorBase.

Constructor & Destructor Documentation

gko::UmpireCudaAllocator::UmpireCudaAllocator ( std::string name )

explicit

Parameters

name	name of an allocator registered with Umpire.

Exceptions

Error	if `name` is not registered, or its resource is not device-accessible.
NotCompiled	if Ginkgo was built without Umpire.

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