C++ Runahead scheduling does not synchronize correctly at barriers

[wlruys]

Kernels are continuing past the barrier execution.
Easily reproducible with Jimmy's boltzmann app (or the timestepping synthetic graph).

[wlruys]

Event synchronize is called correctly before the barrier is marked as complete.

parla-experimental/src/c/backend/include/runtime.hpp

Lines 443 to 449 in 8d2326b

	void synchronize_events() {
	size_t num_events = this->events.size_unsafe();
	for (size_t i = 0; i < num_events; i++) {
	uintptr_t event_ptr = this->events.at_unsafe(i);
	event_synchronize(event_ptr);
	}
	}

Need to check if event synchronize synchronizes on cross-stream events? It should.

[wlruys]

Confirmed that Python runahead works correctly (and gets Jimmy's code to scale better). This is likely a wrong stream or event reference being passed?