[Mpi3-ft] FTWG Con Call 2013-08-20

Aurélien Bouteiller bouteill at icl.utk.edu
Tue Aug 20 13:47:10 CDT 2013


All,


Here is (I think) a correct code using Get_failed (combined Ack-GetAck fn). 


General idea: the atomic shared variable counts number of threads doing ANY_SRC. As long as somebody else wants to do ANY_SRC, a thread is not allowed to do GET_FAILED. Once that becomes possible, it sets the counter to negative to prevent anybody else from entering an ANYSRC critical section while updating the shared grp (a condition is used to resolve that situation later on). 

Please comment. 


pthread_main() {
  threads_in_ANYSRC=0;

if(0 > atomic_inc(threads_in_ANYSRC)) {
  cond_wait(grp_update) // wait for some other get_failed to proceed
  if(!happy(grp)) { 
    atomic_dec(threads_in_ANYSRC)
    goto cleanup
  }
}
rc = MPI_RECV(ANY)
if(rc == MPI_ERR_PROC_FAILED) { 
  while(1 != atomic_cas(threads_in_ANYSRC, 1, num_threads-2)); // if 0, take the token and disable get_failed for others (threads_inANY<0)
  grp = MPI_COMM_GET_FAILED(comm);
  if(!happy(grp)) {
    atomic_inc(threads_in_ANYSRC, num_threads+1) // quit ANYSRC critical section and do something else
    cond_signal(grp_update)
    goto cleanup
  }
  else {
    atomic_inc(threads_in_ANYSRC, num_threads+2) // stay in ANYSRC critical section
    cond_signal(grp_update)
  }
}
rc = MPI_RECV(ANY)
…
atomic_dec(threads_in_ANYSRC)





Le 19 août 2013 à 16:00, Wesley Bland <wbland at mcs.anl.gov> a écrit :

> Dear WG members,
> 
> This is a reminder that according to our planning, we are having our regular phone meeting tomorrow at 3pm EDT.
> 
> NOTE THE NEW CALL-IN NUMBER. This a permanent change from the old number.
> 
> Date: August 20,
> Time: 3pm EDT/New York
> Dial-in information: 712-432-0360
> Code: 623998#
> 
> Agenda:
> * Slides to present at the Madrid meeting (slides attached to this email)
> * Plan for Madrid meeting
> 
> Next Meetings:
> * September 3, 2013
> * September 17, 2013
> * October 1, 2013
> * October 15, 2013
> 
> <2013-09 MPI Forum ULFM.pptx>
> _______________________________________________
> mpi3-ft mailing list
> mpi3-ft at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft

--
* Dr. Aurélien Bouteiller
* Researcher at Innovative Computing Laboratory
* University of Tennessee
* 1122 Volunteer Boulevard, suite 309b
* Knoxville, TN 37996
* 865 974 9375











More information about the mpiwg-ft mailing list