This document contains a detailed description of which variables are accessed by the OpenMP threads in NEST-2.6.0 simulation.
Scheduler::update()
The main simulation runs through the function nest::Scheduler::update()
. The simulation progresses in discrete time windows. A certain number of timesteps sum up to the minimum synaptic delay, min_delay_
, also called a timeslice. During this interval of length min_delay_
, individual threads can work separately. Thus all spike events are delivered (by each thread) at the beginning of each timeslice, and spike events are gathered (from all threads) at the end of each timeslice.
Roughly, the call hierarchy inside Scheduler::update()
looks like the following:
- for each thread in parallel do until simulation is over
- Only at the beginning of a timeslice, call
Scheduler::deliver_events_()
- for each node (neuron/device) assigned to this thread do
- call
Node::update()
- call
- Synchronize threads
- Using a single thread, do the following
- After each
min_delay_
interval, callScheduler::gather_events_()
Scheduler::advance_time_()
- After each
- Only at the beginning of a timeslice, call
Scheduler::update()
Variables used by Each OpenMP thread t
accessess the following variables/functions:
Scheduler::nodes_vec_[t]
Node::update(clock_, from_step_, to_step_)
Important Variables for GPU Computation
Scheduler::net_
, the network objectScheduler::nodes_vec_
, vector containing neurons/devices assigned to each thread- Number of OpenMP threads