[Mpi3-hybridpm] [Mpi3-rma] MPI_Win_Allocate_shared proposal draft

Rajeev Thakur thakur at mcs.anl.gov
Mon Aug 29 13:22:46 CDT 2011


> 6:39 - I would like the sentence to clearly say that it is the user's responsibility to pass a communicator whose processes can share memory. How about changing it to 
> 
> "It is the user's responsibility to ensure that the communicator comm represents a group of processes that can create a shared memory segment that can be accessed by all processes in the group."
> 
> Then we should say what happens if the communicator contains processes that cannot share memory.

If the new function MPI_Comm_split_type passes, we need to remember to provide a reference here to it, because it is directly relevant for this use case.

Rajeev


On Aug 29, 2011, at 1:16 PM, Rajeev Thakur wrote:

> Looks good. Some comments:
> 
> 6:14 - baseptrs -> baseptr
> 
> 6:26 - DISP_UNIT -> SIZE
> 
> 6:30 - "among all processes" -> "among all processes in comm"
> 
> 6:31 - "returns" -> "and returns"
> 
> 6:32 - "be target" -> "be the target"
> 
> 6:33 - "accessed" -> "accesses"
> 
> 6:33 - change to "remote processes; the base pointers for other processes can be queried..."
> 
> 6:35 - Much of the sentence "The returned memory..." seems to be a repetition of what has been said before.
> 
> 6:39 - I would like the sentence to clearly say that it is the user's responsibility to pass a communicator whose processes can share memory. How about changing it to 
> 
> "It is the user's responsibility to ensure that the communicator comm represents a group of processes that can create a shared memory segment that can be accessed by all processes in the group."
> 
> Then we should say what happens if the communicator contains processes that cannot share memory.
> 
> 7:10 - "as info key" -> "as an info key"
> 
> 7:35 - MPI_WIN_FLAVOR_SHM is not defined in 11.2.6
> 
> 
> Since the default allocation is supposed to be contiguous, is the intended implementation method supposed to be something like the following:
>   * wait for everyone to call the function
>   * gather everyone's local size; add it up
>   * allocate single, large shared memory
>   * return local pointers everywhere
> 
> 
> On Aug 29, 2011, at 12:12 AM, Torsten Hoefler wrote:
> 
>> Hello all,
>> 
>> We just posted a new update to the proposal (thanks to Bill for the
>> comments!) to the ticket at
>> https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/ticket/284/
>> 
>> All the 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
> 
> 
> _______________________________________________
> 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