Clean up Initialization and Finalization for DrvX

[ywwu928]

Description

Important -- Read Before Creating a Pull Request

PR description

1. The power on and off part is not of interest in simulation (according to Mattan), so use natural pointers instead of GlobalPtr for those accesses to speed up. This will be emulating that there are special hardware such that when a pod has no more task, it will go to sleep directly. When a task is put onto the pod, there will be hardware mechanism to wake it up. This is a reasonable assumption confirmed with Mattan.

2. Replace waitAll() at the end of a program with endExecution(). This is an empty function in DrvX which can get rid of a lot of unwanted overhead. I tried the same thing in PREP but it won't work.

3. Termination is now determined by the following condition:

   • all CPs have finished distributing work
   • no more remaining tasks on all cores
     In order to achieve this, there is one variable pod_remaining_tasks_ per pod to keep track. The reason it is per pod is because work stealing can only steal from the same pod. If there are no more remaining tasks on a pod, the cores should go to sleep and not try to steal task which creates read overhead. Incrementing the remaining tasks is moved to execute_on call and will increment the task of the worker core instead of the task creation core. decrementing the remaining task is moved to operator()() of task where the task is actually executed.

Checklist

• The additions follow the code standards in the developer guide.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.