Co-thread for parallel processing

Co-thread for parallel processing

[Randi Botse]

My program use co-threads, because creating and reinitialize thread is
expensive, it first create threads and reuse them. The main thread
will send each co-thread a event (through condition variable)
simultaneously to do task in parallel way. Each co-thread has variety
work-time, let say it can be (in miliseconds) 200ms, 130ms, 317ms,
90ms etc to get result to be consumed by main thread.

Here I did to wake-up all threads in the main thread:

.....
for (i = 0; i < nthread; i++)
    pthread_cond_broadcast(&cond);
wait_all_worker(); /* wait for all worker threads to finish */
get_all_worker_result(); /* get result from all worker thread */
...


And worker-thread routines:

...
for (;;) {
  pthread_mutex_lock();
  pthread_cond_wait();
  pthread_mutex_unlock();

  do_work();
  fill_result();
}
....

My dilema is:

1. What the proper way to the main thread to wait for all co-threads
until finished?
2. After calling pthread_cond_signal() or pthread_cond_broadcast()
followed by sleep()/nanosleep(),
    the condition variable looks like not signaled, (the co-thread
routine doesn't run). What make this?

Re: Co-thread for parallel processing

[Bogdan Cristea]

On Friday 02 December 2011 10:23:49 you wrote:
> My program use co-threads, because creating and reinitialize thread is
> expensive, it first create threads and reuse them. The main thread
> will send each co-thread a event (through condition variable)
> simultaneously to do task in parallel way. Each co-thread has variety
> work-time, let say it can be (in miliseconds) 200ms, 130ms, 317ms,
> 90ms etc to get result to be consumed by main thread.
> 
> Here I did to wake-up all threads in the main thread:
> 
> .....
> for (i = 0; i < nthread; i++)
>     pthread_cond_broadcast(&cond);
> wait_all_worker(); /* wait for all worker threads to finish */
> get_all_worker_result(); /* get result from all worker thread */
> ...
> 
> 
> And worker-thread routines:
> 
> ...
> for (;;) {
>   pthread_mutex_lock();
>   pthread_cond_wait();
>   pthread_mutex_unlock();
> 
>   do_work();
>   fill_result();
> }
> ....
> 
> My dilema is:
> 
> 1. What the proper way to the main thread to wait for all co-threads
> until finished?

I would use  a semaphore which should be locked by the treads once the 
condition variable is set en unlocked when the result is available for the 
main thread.

> 2. After calling pthread_cond_signal() or pthread_cond_broadcast()
> followed by sleep()/nanosleep(),
>     the condition variable looks like not signaled, (the co-thread
> routine doesn't run). What make this?

I don't understand, please provide a working example

-- 
Bogdan 

Re: Co-thread for parallel processing

[Randi Botse]

> I would use  a semaphore...
Ok, let see what I can do with that.

2. Just forget that, that might be caused by my frustation... ;)

> 2. After calling pthread_cond_signal() or pthread_cond_broadcast()
> followed by sleep()/nanosleep(),
>     the condition variable looks like not signaled, (the co-thread
> routine doesn't run). What make this?

I don't understand, please provide a working example

> Bogdan

Re: Co-thread for parallel processing

[Srinivasa T N]

> 1. What the proper way to the main thread to wait for all co-threads
> until finished?
> 2. After calling pthread_cond_signal() or pthread_cond_broadcast()
> followed by sleep()/nanosleep(),
>      the condition variable looks like not signaled, (the co-thread
> routine doesn't run). What make this?
I think you need a pthread_cond_broadcast () after do_work () in each 
thread.

Regards,
Seenu.

Re: Co-thread for parallel processing

[Hendrik Visage]

On Fri, Dec 2, 2011 at 11:23 AM, Randi Botse <nightdecoder@gmail.com> wrote:
> 1. What the proper way to the main thread to wait for all co-threads
> until finished?

I'll do a semop() with sem_op=1 in each thread's start, and a
sem_op=-1 when each thread's finished.
Then in the main thread I'll wait with a sem_op = 0.

> 2. After calling pthread_cond_signal() or pthread_cond_broadcast()
> followed by sleep()/nanosleep(),
>    the condition variable looks like not signaled, (the co-thread
> routine doesn't run). What make this?

Please show the code snippet in the co-thread that is not working as
well as the mainthread's method you try to wake up the co-threads
--
To unsubscribe from this list: send the line "unsubscribe linux-c-programming" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html