base_device::DeviceContext Class Reference

Singleton class to manage GPU device context and initialization. More...

#include <device.h>

Collaboration diagram for base_device::DeviceContext:

Public Member Functions
void	init ()
	Initialize GPU device binding.
int	get_device_id () const
	Get the bound GPU device ID.
int	get_device_count () const
	Get the total number of GPU devices on this node.
int	get_local_rank () const
	Get the local MPI rank within the node.
	DeviceContext (const DeviceContext &)=delete
DeviceContext &	operator= (const DeviceContext &)=delete

Static Public Member Functions
static DeviceContext &	instance ()
	Get the singleton instance of DeviceContext.

Private Member Functions
	DeviceContext ()=default
	~DeviceContext ()=default

Private Attributes
bool	initialized_ = false
bool	gpu_enabled_ = false
int	device_id_ = -1
int	device_count_ = 0
int	local_rank_ = 0
std::mutex	init_mutex_

Detailed Description

Singleton class to manage GPU device context and initialization.

This class provides a centralized way to:

1. Initialize GPU device binding (only once)
2. Query GPU device state (device_id, device_count, etc.)
3. Ensure thread-safe initialization

Usage: // Initialize (call once after MPI init and after determining device=gpu) DeviceContext::instance().init(MPI_COMM_WORLD);

// Query device info int dev_id = DeviceContext::instance().get_device_id();

Constructor & Destructor Documentation

DeviceContext() [1/2]

base_device::DeviceContext::DeviceContext ( const DeviceContext &  )

delete

DeviceContext() [2/2]

base_device::DeviceContext::DeviceContext ( )

privatedefault

~DeviceContext()

base_device::DeviceContext::~DeviceContext ( )

privatedefault

Member Function Documentation

get_device_count()

int base_device::DeviceContext::get_device_count ( ) const

inline

Get the total number of GPU devices on this node.

Returns

Number of GPU devices, or 0 if not initialized

get_device_id()

int base_device::DeviceContext::get_device_id ( ) const

inline

Get the bound GPU device ID.

Returns

Device ID (0-based), or -1 if not initialized

Here is the caller graph for this function:

get_local_rank()

int base_device::DeviceContext::get_local_rank ( ) const

inline

Get the local MPI rank within the node.

Returns

Local rank, or 0 if not initialized

init()

void base_device::DeviceContext::init

(

)

Initialize GPU device binding.

This function:

1. Gets the local rank within the node using MPI_COMM_TYPE_SHARED (MPI_COMM_WORLD)
2. Queries the number of available GPU devices
3. Binds the current process to a GPU device (local_rank % device_count)

Note

This function should only be called once. Subsequent calls are no-ops.

This function should only be called when device=gpu is confirmed.

In MPI builds, uses MPI_COMM_WORLD internally.

Here is the call graph for this function:

Here is the caller graph for this function:

instance()

DeviceContext & base_device::DeviceContext::instance ( )

static

Get the singleton instance of DeviceContext.

Returns

Reference to the singleton instance

Here is the call graph for this function:

Here is the caller graph for this function:

operator=()

DeviceContext & base_device::DeviceContext::operator= ( const DeviceContext &  )

delete

Member Data Documentation

device_count_

int base_device::DeviceContext::device_count_ = 0

private

device_id_

int base_device::DeviceContext::device_id_ = -1

private

gpu_enabled_

bool base_device::DeviceContext::gpu_enabled_ = false

private

init_mutex_

std::mutex base_device::DeviceContext::init_mutex_

private

initialized_

bool base_device::DeviceContext::initialized_ = false

private

local_rank_

int base_device::DeviceContext::local_rank_ = 0

private

---

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

• /home/runner/work/abacus-develop/abacus-develop/source/source_base/module_device/device.h
• /home/runner/work/abacus-develop/abacus-develop/source/source_base/module_device/device.cpp