In openMP how do I ensure threads are synchronized before continuing?

1
votes

I am using a #pragma omp barrier to ensure that all my parallel threads meet up at the same point before continuing (no fancy conditionally branching code, just straight loop), but I am surmising that the barrier pragma does not actually guarantee synchronicity, just completion as these are the results I am getting:

0: func() size: 64 Time: 0.000414 Start: 1522116688.801262 End: 1522116688.801676
1: func() size: 64 Time: 0.000828 Start: 1522116688.801263 End: 1522116688.802091

thread 0 is starting about a microsecond faster than thread 1, giving it the somewhat unrealistic completion time of 0.414 msec, incidentally in a single core/thread run the run time averages around 0.800 msec. (please forgive me if my units are off, it is late).

My Question is: Is there a way to ensure in openMP that threads are all started at the same time? Or would I have to bring in another library like pthread in order to have this functionality?

1
copenmpbenchmarkingintel
Can you post your loop? As far as I know there are not parallel libraries that allow you to specify threads start at the same time. – user9478968
What is the actual problem you are trying to solve? Please post a minimal reproducible example to illustrate your problem. You absolutely cannot ensure that multiple threads start at the same time with anything other than a hard realtime system. – Zulan
How would you want to guarantee that? What if you have more threads than cores? What if other processes are running on the same system (there are always some, such as background o.s. processes)? – Daniel Langr
You might get closer to simultaneous resumption of all threads if you have them busy-waiting (aka spinning) on a shared-memory atomic flag. No OS involvement in waking them up, just a store from one core which all the other cores will see at nearly the same time if they're all sitting in a pause / cmp byte [flag], 0 / je .retry busy-wait loop. (Write that loop in C with atomics and _mm_pause() if you want). – Peter Cordes

1 Answers

1
votes

The barrier statement in OpenMP, as in other languages, ensures no thread progresses until all threads reach the barrier.

It does not specify the order in which threads begin to execute again. As far as I know manually scheduling threads is not possible in OpenMP or Pthread libraries(see comment below).