uni20::cuda::Runtime Class Reference

Scoped owner of the process-wide CUDA resource installation. More...

#include <runtime.hpp>

Public Member Functions
std::optional< int >	default_device () const noexcept
	Return the configured default CUDA device ordinal, if any.
DeviceResources &	default_device_resources () const
	Return the canonical resources for the configured default device.
DeviceResources &	device_resources (int device) const
	Return the canonical resources for an enrolled device.
bool	has_device (int device) const noexcept
	Return whether this runtime enrolled a CUDA device ordinal.
Runtime &	operator= (Runtime &&)=delete
Runtime &	operator= (Runtime const &)=delete
	Runtime (Runtime &&)=delete
	Runtime (Runtime const &)=delete
	~Runtime () noexcept

Friends
Runtime	initialize (RuntimeConfig)
	Install and return a scoped process-wide CUDA runtime owner.

Detailed Description

Scoped owner of the process-wide CUDA resource installation.

Construction installs globally discoverable per-device resources. Destruction first uninstalls lookup, then drains and destroys those resources. The owner must outlive every CUDA Tensor, buffer, task, stream, and provider-resource lease.

Constructor & Destructor Documentation

Runtime() [1/2]

uni20::cuda::Runtime::Runtime ( Runtime const &  )

delete

Runtime() [2/2]

uni20::cuda::Runtime::Runtime ( Runtime &&  )

delete

~Runtime()

uni20::cuda::Runtime::~Runtime ( )

noexcept

Member Function Documentation

default_device()

std::optional< int > uni20::cuda::Runtime::default_device ( ) const

inlinenoexcept

Return the configured default CUDA device ordinal, if any.

default_device_resources()

DeviceResources & uni20::cuda::Runtime::default_device_resources

(

)

const

Return the canonical resources for the configured default device.

device_resources()

DeviceResources & uni20::cuda::Runtime::device_resources

(

int

device

)

const

Return the canonical resources for an enrolled device.

has_device()

bool uni20::cuda::Runtime::has_device ( int  device ) const

noexcept

Return whether this runtime enrolled a CUDA device ordinal.

operator=() [1/2]

Runtime & uni20::cuda::Runtime::operator= ( Runtime &&  )

delete

operator=() [2/2]

Runtime & uni20::cuda::Runtime::operator= ( Runtime const &  )

delete

Friends And Related Symbol Documentation

initialize

Runtime initialize ( RuntimeConfig  )

friend

Install and return a scoped process-wide CUDA runtime owner.

Only one installation may be active. Retain the returned object at the application startup boundary for as long as CUDA objects exist.

---

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

• runtime.hpp
• runtime.cpp