GPU timing functions

[jrcccc]

Just a suggestion on how to implement GPU kernel timing functions using Events:

cuda Event_t startTime, stopTime;
float timeInMs;
cudaEventCreate(&startTime);
cudaEventCreate(&stopTime);

And before / after every kernel launch:

cudaEventRecord(startTime);
launch_kernel<<<blablabla>>>(...);
cudaEventRecord(stopTime);


cudaEventSynchronize(stopTime);

cudaEventElapsedTime(&timeInMs, startTime, stopTime);

And of course equivalent hip functions.