[Mpi3-rma] Draft Plenary Presentation for MPI RMA
dinan at mcs.anl.gov
Fri Feb 4 12:22:45 CST 2011
These look good. Here's some feedback:
Slide 4: RAW (Read after Write), WAR, WAW are commonly used terms for
this, should we use those?
Slide 21: UPC allows one-sided allocation of distributed shared data
through the one-sided call "shared void *upc_global_alloc(size_t
nblocks, size_t nbytes)". I think Dan was pointing out that this is
impossible in MPI-2. (Still don't think we can support this in MPI-3.)
Slide 22: UPC doesn't allow sharing data on the stack, only heap data
can be shared. In general, this is kind of an iffy thing, might
encounter some resistance to suggest doing this. Linux is going to
initially map the stack to the zero page which is marked copy-on-write,
so beyond the parts you have used, the stack doesn't exist.
Slide 27: ARMCI has per-operation non-blocking calls. Blocking calls
and waits give you local completion and only fence gives you remote
Slide 30: Conflicting relaxed operations issued by a single source will
appear in program order when observed by that source. IIRC, a third
party need not observe this same order. I think that the result of
conflicting relaxed operations issued from different sources is undefined.
On 2/3/11 4:16 PM, Rajeev Thakur wrote:
> This looks very good, particularly the slides at the end that show how it addresses published criticisms, comparison to ARMCI, and discussion of UPC and Co-Array Fortran.
> Some comments:
> slide 3: "allocate memory on window" -> "allocate memory for window"
> slide 3: "attach memory..." -> dynamically attach/detach memory ...
> slide 3: "Collective Win_lock_all": Win_lock_all is specifically defined as non-collective; all refers to all targets from one origin
> slide 14 last line: tho -> to
> slide 29: last bullet refers to "proposal 1"
> On Feb 3, 2011, at 2:55 PM, William Gropp wrote:
>> I've posted a draft of the plenary presentation on the RMA Wiki. Comments welcome; if you want to edit it, send a message to this list first to grad the edit token. ( https://svn.mpi-forum.org/trac/mpi-forum-web/attachment/wiki/RmaWikiPage/rma-overview-2-11a.pptx )
>> William Gropp
>> Deputy Director for Research
>> Institute for Advanced Computing Applications and Technologies
>> Paul and Cynthia Saylor Professor of Computer Science
>> University of Illinois Urbana-Champaign
>> mpi3-rma mailing list
>> mpi3-rma at lists.mpi-forum.org
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
More information about the mpiwg-rma