[Mpi3-hybridpm] [Mpi3-rma] Alternative Proposal for Shared Memory Support
Rajeev Thakur
thakur at mcs.anl.gov
Tue Mar 15 21:22:49 CDT 2011
All can be used, but none are needed, right?
Currently, in the unified model you can do:
lock
put(A)
unlock
barrier barrier
x = A
With shared memory, you could do
A=10
barrier barrier
x = A
Right?
If the above is true, it might be a bit confusing to users that in the RMA chapter you use put, get, and synchronization functions for some things and can directly read/write memory for some other things.
Rajeev
On Mar 15, 2011, at 9:17 PM, Torsten Hoefler wrote:
> On Tue, Mar 15, 2011 at 09:03:40PM -0500, Rajeev Thakur wrote:
>> Since processes sharing memory can/will access it directly using
>> memory references (not put/get), and this proposal is for the unified
>> memory model, we need to say whether any of the synchronization
>> functions in the RMA chapter are needed at all for such usage.
> Well, put/get to local windows are also allowed, so this would not be an
> exception. It's even nicely defined because the chapter defines the
> behavior for overlapping windows, and a shared window could be treated
> semantically like multiple overlapping windows. I am not sure if all are
> needed but I would say that all can be used.
>
> Yes, load/store accesses would probably be the most common use (plus
> maybe the new (portable) atomics!).
>
> Best,
> Torsten
>
> --
> bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ -----
> Torsten Hoefler | Performance Modeling and Simulation Lead
> Blue Waters Directorate | University of Illinois (UIUC)
> 1205 W Clark Street | Urbana, IL, 61801
> NCSA Building | +01 (217) 244-7736
> _______________________________________________
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
More information about the mpiwg-hybridpm
mailing list