Mila::Dnn::Compute::DynamicMemoryResource Class Reference

A class that represents a dynamically-determined memory resource. More...

Inheritance diagram for Mila::Dnn::Compute::DynamicMemoryResource:

Collaboration diagram for Mila::Dnn::Compute::DynamicMemoryResource:

Public Member Functions
	DynamicMemoryResource (Compute::DeviceType device_type=Compute::DeviceType::Cuda)
	Constructs a DynamicMemoryResource based on the device type.
bool	is_device_accessible () const noexcept
	Checks if the memory resource is device-accessible.
bool	is_host_accessible () const noexcept
	Checks if the memory resource is host-accessible.

Protected Member Functions
void *	do_allocate (std::size_t size, std::size_t alignment) override
	Allocates memory of the specified size with the given alignment.
void	do_deallocate (void *ptr, std::size_t size, std::size_t alignment) override
	Deallocates previously allocated memory.
bool	do_is_equal (const std::pmr::memory_resource &other) const noexcept override
	Checks if this memory resource is equal to another memory resource.

Private Attributes
std::variant< Compute::CpuMemoryResource, Compute::CudaMemoryResource >	resource_variant_

Detailed Description

A class that represents a dynamically-determined memory resource.

This class serves as an adapter between the runtime selection of memory resources (via DeviceContext) and the compile-time requirements of the Tensor class, which requires a specific memory resource type rather than a variant.

Constructor & Destructor Documentation

DynamicMemoryResource()

Mila::Dnn::Compute::DynamicMemoryResource::DynamicMemoryResource ( Compute::DeviceType  device_type = Compute::DeviceType::Cuda )

inlineexplicit

Constructs a DynamicMemoryResource based on the device type.

Parameters

device_type

The type of device to create the memory resource for.

Member Function Documentation

do_allocate()

void * Mila::Dnn::Compute::DynamicMemoryResource::do_allocate ( std::size_t  size, std::size_t  alignment  )

inlineoverrideprotected

Allocates memory of the specified size with the given alignment.

This delegates to the appropriate memory resource based on the device type.

Parameters

size	The size in bytes to allocate.
alignment	The alignment requirement for the allocation.

Returns

void* Pointer to the allocated memory.

do_deallocate()

void Mila::Dnn::Compute::DynamicMemoryResource::do_deallocate ( void *  ptr, std::size_t  size, std::size_t  alignment  )

inlineoverrideprotected

Deallocates previously allocated memory.

This delegates to the appropriate memory resource based on the device type.

Parameters

ptr	Pointer to the memory to deallocate.
size	The size in bytes of the allocation.
alignment	The alignment of the allocation.

do_is_equal()

bool Mila::Dnn::Compute::DynamicMemoryResource::do_is_equal ( const std::pmr::memory_resource &  other ) const

inlineoverrideprotectednoexcept

Checks if this memory resource is equal to another memory resource.

Parameters

other

The other memory resource to compare with.

Returns

bool True if the memory resources are equal, false otherwise.

is_device_accessible()

bool Mila::Dnn::Compute::DynamicMemoryResource::is_device_accessible ( ) const

inlinenoexcept

Checks if the memory resource is device-accessible.

Returns

bool True if the memory is accessible from the device (GPU).

is_host_accessible()

bool Mila::Dnn::Compute::DynamicMemoryResource::is_host_accessible ( ) const

inlinenoexcept

Checks if the memory resource is host-accessible.

Returns

bool True if the memory is accessible from the host (CPU).

Member Data Documentation

resource_variant_

std::variant<Compute::CpuMemoryResource, Compute::CudaMemoryResource> Mila::Dnn::Compute::DynamicMemoryResource::resource_variant_

private

---

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

• /home/runner/work/Mila/Mila/Mila/Src/Dnn/Compute/DynamicMemoryResource.ixx