Abstract interface for setting up / running a CUDA task. More...

#include <IKernelTask.h>

Collaboration diagram for AthCUDA::IKernelTask:

Public Types
enum	KernelExecMode { Synchronous = 0, Asynchronous = 1 }
	Possible modes in which the kernel was executed. More...

Public Member Functions
virtual	~IKernelTask ()=default
	Virtual destructor, to make vtable happy. More...

virtual int	execute (StreamHolder &stream)=0
	Execute the kernel using a specific stream. More...

virtual int	finished (int code, KernelExecMode mode)=0
	Function called when an asynchronous execution finishes. More...

Detailed Description

Abstract interface for setting up / running a CUDA task.

The expectation is that the implementation of such a task would either schedule itself to CUDA for execution, or if no GPU is available (at the time), would execute its calculation on the CPU instead.

Author: Attila Krasznahorkay Attil.nosp@m.a.Kr.nosp@m.aszna.nosp@m.hork.nosp@m.ay@ce.nosp@m.rn.c.nosp@m.h

Definition at line 22 of file IKernelTask.h.

Member Enumeration Documentation

◆ KernelExecMode

enum AthCUDA::IKernelTask::KernelExecMode

Possible modes in which the kernel was executed.

Enumerator
Synchronous	The kernel was executed synchronously on the CPU.
Asynchronous	The kernel was executed asynchronously on a CUDA device.

Definition at line 29 of file IKernelTask.h.

                           {
          Synchronous  = 0,
          Asynchronous = 1
       };

Constructor & Destructor Documentation

◆ ~IKernelTask()

virtual AthCUDA::IKernelTask::~IKernelTask ( )

virtualdefault

Virtual destructor, to make vtable happy.

Member Function Documentation

◆ execute()

virtual int AthCUDA::IKernelTask::execute ( StreamHolder & stream )

pure virtual

Execute the kernel using a specific stream.

If the stream is "invalid", the task needs to be executed in the current thread on the CPU instead.

Parameters

stream The stream to use in the task execution

Returns: 0 if the call succeeded, something else otherwise

◆ finished()

virtual int AthCUDA::IKernelTask::finished	(	int	code,
		KernelExecMode	mode
	)

pure virtual

Function called when an asynchronous execution finishes.

This function is set up to be called automatically when the task's execution is offloaded to a GPU, and that execution has finished. It can be used to notify some listener about the finish of the task.