openswmm::twoD::GpuExecutionOwner Class Reference

#include <GpuExecutionOwner.hpp>

Collaboration diagram for openswmm::twoD::GpuExecutionOwner:

Public Member Functions
	GpuExecutionOwner ()
	Start a dedicated synchronous-dispatch worker.
	~GpuExecutionOwner ()
	Stop and join a nonresident execution owner after queued work drains.
	GpuExecutionOwner (const GpuExecutionOwner &)=delete
template<class F>
auto	invoke (F &&fn) -> std::invoke_result_t< F >
	Execute one callable on the controlling thread and wait for completion.

Static Public Member Functions
static std::shared_ptr< GpuExecutionOwner >	shared ()
	Acquire the process-resident controlling thread for matched GPU plugins.

Constructor & Destructor Documentation

GpuExecutionOwner() [1/2]

openswmm::twoD::GpuExecutionOwner::GpuExecutionOwner ( )

inline

Start a dedicated synchronous-dispatch worker.

Postcondition

Constructs an empty queue and a worker running work(). Thread creation may throw std::system_error; no partially live owner is returned.

Note

Normally use shared(). This constructor does not initialize Kokkos or load a plugin.

Here is the caller graph for this function:

~GpuExecutionOwner()

openswmm::twoD::GpuExecutionOwner::~GpuExecutionOwner ( )

inline

Stop and join a nonresident execution owner after queued work drains.

Precondition

No concurrent or later invoke() calls; destructor must run outside its worker thread. Never destroy under a DLL loader lock.

Postcondition

Sets stopping, wakes the worker, drains already queued tasks and joins it. Process-resident shared() instances intentionally avoid teardown-time destruction.

GpuExecutionOwner() [2/2]

openswmm::twoD::GpuExecutionOwner::GpuExecutionOwner ( const GpuExecutionOwner & )

delete

Here is the call graph for this function:

Member Function Documentation

invoke()

template<class F>

auto openswmm::twoD::GpuExecutionOwner::invoke ( F && fn ) -> std::invoke_result_t<F>

inline

Execute one callable on the controlling thread and wait for completion.

Template Parameters

F	Callable type; must be invocable with no arguments.

Parameters

[in]

fn

Forwarded callable, moved/copied into an owned queued task; captures may borrow buffers only until invoke returns.

Returns

Callable result, including void/reference results according to std::invoke_result_t<F>. Exceptions from the callable are rethrown through the future; allocation/queue failures may also propagate.

Note

Reentrant calls from the worker execute inline. Other callers enqueue under a mutex and block; queued calls are serialized. Owner must remain alive, and destruction must not race this call. Does not retain caller buffers after the synchronous result is delivered.

shared()

static std::shared_ptr< GpuExecutionOwner > openswmm::twoD::GpuExecutionOwner::shared ( )

inlinestatic

Acquire the process-resident controlling thread for matched GPU plugins.

Returns

Shared ownership handle to the one intentionally resident owner; a static allocation retains it through process teardown.

Note

First use allocates and starts a thread and may throw. Factory must pin the containing module before first use, avoiding thread join under the Windows loader lock. Later acquisitions are synchronized by C++ static initialization.

---

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

• /home/runner/work/openswmm.engine/openswmm.engine/src/engine/2d/solver/GpuExecutionOwner.hpp