Re: [OMPI users] multi-threaded programming

[Jeff Squyres]

On Mar 8, 2011, at 12:34 PM, Eugene Loh wrote:

> Let's say you have multi-threaded MPI processes, you request 
> MPI_THREAD_MULTIPLE and get MPI_THREAD_MULTIPLE, and you use the self,sm,tcp 
> BTLs (which have some degree of threading support).  Is it okay to have an 
> [MPI_Isend|MPI_Irecv] on one thread be completed by an MPI_Wait on another 
> thread?  I'm assuming some sort of synchronization and memory barrier/flush 
> in between to protect against funny race conditions.
> 
> If it makes things any easier on you, we can do this multiple-choice style:
> 
> 1)  Forbidden by the MPI standard.
> 2)  Not forbidden by the MPI standard, but will not work with OMPI (not even 
> with the BTLs that claim to be multi-threaded).
> 3)  Works well with OMPI (provided you use a BTL that's multi-threaded).

I believe the current answer is #2, but it would be great if the answer could 
change to be a variant of #3:

3) Works well with OMPI (provided you use a BTL that's safe to use with 
MPI_THREAD_MULTIPLE)

(i.e., the BTL doesn't have to be multi-threaded, itself)

-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/

Re: [OMPI users] multi-threaded programming

[Eugene Loh]

Durga Choudhury wrote:

  A follow-up question (and pardon if this sounds stupid) is this:

If I want to make my process multithreaded, BUT only one thread has
anything to do with MPI (for example, using OpenMP inside MPI), then
the results will be correct EVEN IF #1 or #2 of Eugene holds true. Is
this correct?
  


I believe this is thoroughly covered by the standard (though I suppose
the same could have been said about my question).

In any case, for your situation, initialize MPI with
MPI_Init_thread().  Ask for thread level MPI_THREAD_FUNNELED and check
that that level is provided.  That should cover your case.  See the man
page for MPI_Init_thread().  My question should not have anything to do
with your case.

  On Tue, Mar 8, 2011 at 12:34 PM, Eugene Loh  wrote:
  
  
Let's say you have multi-threaded MPI processes, you request
MPI_THREAD_MULTIPLE and get MPI_THREAD_MULTIPLE, and you use the self,sm,tcp
BTLs (which have some degree of threading support).  Is it okay to have an
[MPI_Isend|MPI_Irecv] on one thread be completed by an MPI_Wait on another
thread?  I'm assuming some sort of synchronization and memory barrier/flush
in between to protect against funny race conditions.

If it makes things any easier on you, we can do this multiple-choice style:

1)  Forbidden by the MPI standard.
2)  Not forbidden by the MPI standard, but will not work with OMPI (not even
with the BTLs that claim to be multi-threaded).
3)  Works well with OMPI (provided you use a BTL that's multi-threaded).

It's looking like #2 to me, but I'm not sure.
___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users


  
  
___
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
  

Re: [OMPI users] multi-threaded programming

[Durga Choudhury]

A follow-up question (and pardon if this sounds stupid) is this:

If I want to make my process multithreaded, BUT only one thread has
anything to do with MPI (for example, using OpenMP inside MPI), then
the results will be correct EVEN IF #1 or #2 of Eugene holds true. Is
this correct?

Thanks
Durga

On Tue, Mar 8, 2011 at 12:34 PM, Eugene Loh  wrote:
> Let's say you have multi-threaded MPI processes, you request
> MPI_THREAD_MULTIPLE and get MPI_THREAD_MULTIPLE, and you use the self,sm,tcp
> BTLs (which have some degree of threading support).  Is it okay to have an
> [MPI_Isend|MPI_Irecv] on one thread be completed by an MPI_Wait on another
> thread?  I'm assuming some sort of synchronization and memory barrier/flush
> in between to protect against funny race conditions.
>
> If it makes things any easier on you, we can do this multiple-choice style:
>
> 1)  Forbidden by the MPI standard.
> 2)  Not forbidden by the MPI standard, but will not work with OMPI (not even
> with the BTLs that claim to be multi-threaded).
> 3)  Works well with OMPI (provided you use a BTL that's multi-threaded).
>
> It's looking like #2 to me, but I'm not sure.
> ___
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>

[OMPI users] multi-threaded programming

[Eugene Loh]

Let's say you have multi-threaded MPI processes, you request 
MPI_THREAD_MULTIPLE and get MPI_THREAD_MULTIPLE, and you use the 
self,sm,tcp BTLs (which have some degree of threading support).  Is it 
okay to have an [MPI_Isend|MPI_Irecv] on one thread be completed by an 
MPI_Wait on another thread?  I'm assuming some sort of synchronization 
and memory barrier/flush in between to protect against funny race 
conditions.


If it makes things any easier on you, we can do this multiple-choice style:

1)  Forbidden by the MPI standard.
2)  Not forbidden by the MPI standard, but will not work with OMPI (not 
even with the BTLs that claim to be multi-threaded).

3)  Works well with OMPI (provided you use a BTL that's multi-threaded).

It's looking like #2 to me, but I'm not sure.