Yeah; the cid needs to be unique from the communicator passed to win_create 
because the win functions will call collectives on their communicator.  It 
would be unfortunate to have user collectives and window collectives colliding 
(unfortunate meaning violating the standard here).  It just makes a heap of 
things easier to get a unique CID per window, although it does limit both the 
total number of windows and the total number of communicators you can have.

Brian

On 8/31/15, 7:28 AM, "users on behalf of 
abhisek...@gmail.com<mailto:abhisek...@gmail.com>" 
<users-boun...@open-mpi.org<mailto:users-boun...@open-mpi.org> on behalf of 
abhisek...@gmail.com<mailto:abhisek...@gmail.com>> wrote:

RIght, my mistake, I missed the new cid generation in the comm duplicator 
calls. Thanks a lot!


--
Abhisek
Live Long and Prosper

On Mon, Aug 31, 2015 at 9:25 AM, Todd Kordenbrock 
<thkgc...@gmail.com<mailto:thkgc...@gmail.com>> wrote:
Hi Abhisek,

Both the portals4 and pt2pt OSC components call ompi_comm_dup() near the 
beginning of component_select(), so each window has it's own communicator.  The 
duplicated communicator has a unique contextid and therefore the window has 
unique match bits/hash key.

Thanks,
todd



On Mon, Aug 31, 2015 at 5:59 AM, Jeff Squyres (jsquyres) 
<jsquy...@cisco.com<mailto:jsquy...@cisco.com>> wrote:
On Aug 27, 2015, at 11:55 AM, abhisek...@gmail.com<mailto:abhisek...@gmail.com> 
wrote:
>
> I want to know if it is allowed in MPI one-sided communication to open 
> multiple windows simultaneously using  the same communicator.

Yes.

> The standard does not seem to forbid it as far as I can see, but when I look 
> at the open mpi  osc components (rdma and portals4) it looks like a window is 
> always identified by the context-id in the communicator (for example the 
> match bits in portals4 component use the communicator cid to distinguish 
> among windows, there is a hash-table in the rdma component that uses the 
> communicator cid as the key to store open modules). It seems to me that these 
> components may not be able to support multiple open windows opened with the 
> same communicator.

Nathan/Todd will have to answer that.  I'd be (very) surprised if the code did 
not support multiple windows that were generated from the same communicator.

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

_______________________________________________
users mailing list
us...@open-mpi.org<mailto:us...@open-mpi.org>
Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users
Link to this post: 
http://www.open-mpi.org/community/lists/users/2015/08/27509.php


_______________________________________________
users mailing list
us...@open-mpi.org<mailto:us...@open-mpi.org>
Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users
Link to this post: 
http://www.open-mpi.org/community/lists/users/2015/08/27510.php





Reply via email to