[Mpi3-rma] RMA telecon, 1/24, noon Central
balaji at mcs.anl.gov
Sat Jan 22 21:04:22 CST 2011
On 01/22/2011 08:19 PM, James Dinan wrote:
> Is the justification for the window argument that a system could use
> both RMA methods and thus have some windows that are MPI_RMA_UNIFIED and
> some that are MPI_RMA_SEPARATE (maybe host memory and accelerator memory
> or something like that)? If this is the intent, have we also added a
> mechanism (e.g. info argument) to request an RMA mode for a window?
Yes, different windows can have different memory models (for example
MPI_COMM_SELF might have MPI_RMA_UNIFIED even for non-cache-coherent
I'm not sure I understand the need for an "info" argument. An "info" is
not used for requesting something from MPI -- it's just to tell MPI what
the application will be doing. E.g., you can tell MPI that you don't
want to use some features allowing it to optimize other features.
I'm assuming that you didn't really mean "info", and meant just a
separate parameter that asks for one type of memory model or other. In
this case, I'm not sure MPI can give you one or the other on demand.
Depending on the hardware platform, you get one or the other. You can
query for which one you are getting, which the MPI_RMA_QUERY function is
meant to provide.
More information about the mpiwg-rma