AthCUDA Namespace Reference

Namespaces
namespace	Details

Classes
class	Info
	Class providing information about the CUDA devices at runtime. More...
struct	DeviceInfo
class	IGPUSystemInfoSvc
class	IKernelRunnerSvc
	Interface for executing AthCUDA::IKernelTask tasks. More...
class	IKernelTask
	Abstract interface for setting up / running a CUDA task. More...
class	IStreamPoolSvc
	Interface for a CUDA stream pool service. More...
class	IStreamProvider
	Interface for components providing CUDA streams to (reentrant) algorithms. More...
class	IStreamSvc
	Interface for services providing CUDA streams to (reentrant) algorithms. More...
class	IStreamTool
	Interface for tools providing CUDA streams to (reentrant) algorithms. More...
class	StreamHolder
	Helper class for handling CUDA streams. More...
class	KernelStatus
	Helper object used for synchronising GPU kernel tasks. More...
class	AsyncCopyTool
	Tool providing an asynchronous vecmem::copy object for CUDA devices. More...
class	CopyTool
	Tool providing a synchronous vecmem::copy object for CUDA devices. More...
class	DeviceMemoryResourceTool
	Tool providing memory resource for device memory. More...
class	GPUSystemInfoSvc
class	HostMemoryResourceTool
	Tool providing memory resource for pinned host memory. More...
class	KernelRunnerSvc
	Service used for executing AthCUDA::IKernelTask tasks. More...
class	KernelRunnerSvcImplCallback
	Class used for calling AthCUDA::KernelRunnerSvc::setTaskFinished. More...
class	KernelRunnerSvcImpl
	CUDA implementation of AthCUDA::KernelRunnerSvc. More...
class	ManagedMemoryResourceTool
	Tool providing memory resource for managed host/device memory. More...
class	PerComponentStreamTool
	Tool providing a separate (single) CUDA stream to the component using it. More...
class	PerEventAndComponentStreamTool
	Tool providing a separate, per-event CUDA streams to the component using it. More...
class	PerEventStreamSvc
	Service providing CUDA streams to (reentrant) algorithms. More...
class	SingleStreamSvc
	Service providing CUDA streams to (reentrant) algorithms. More...
class	StreamPoolImpl
class	StreamPool
class	StreamPoolSvc
	Service managing a fixed number of CUDA streams for an application. More...
class	StreamPoolSvcImpl
	CUDA implementation of AthCUDA::StreamPoolSvc. More...
class	StreamSvcAdaptorTool
	Tool exposing a CUDA stream service as a CUDA stream tool. More...

Functions
std::ostream &	operator<< (std::ostream &out, const AthCUDA::Info &info)
	Print operator for AthCUDA::Info.
void	enqueueTask (std::function< void()> f)
	Enqueue a function to the TBB task arena used for CUDA calls.
tbb::task_arena &	taskArena ()

Function Documentation

enqueueTask()

void AthCUDA::enqueueTask

(

std::function< void()>

f

)

Enqueue a function to the TBB task arena used for CUDA calls.

Since many (most?) CUDA calls use global locks, in order to avoid waiting times in the threads, one needs to schedule all CUDA operations using a single task arena, which takes care of executing all operations one by one, using the TBB threads allocated for the rest of the job.

Definition at line 20 of file TaskArena.cxx.

                                            {
 
      taskArena().enqueue( f );
   }

operator<<()

std::ostream & AthCUDA::operator<<	(	std::ostream &	out,
		const AthCUDA::Info &	info )

Print operator for AthCUDA::Info.

taskArena()

tbb::task_arena & AthCUDA::taskArena

(

)

Definition at line 14 of file TaskArena.cxx.

                              {
 
      static tbb::task_arena arena ATLAS_THREAD_SAFE ( 1, 0 );
      return arena;
   }