Hi all, So to fix the recently discovered problem in the separate model and win_allocate_shared, the formed ad-hoc working group proposes the following: Add the following sentence to https://svn.mpi-forum.org/trac/mpi-forum-web/raw-attachment/ticket/284/one-s... page 46 line 47 after ".. obey the following rules.": "For shared memory windows (cf. Section~\ref{sec:alloc_shared}), a store/load that accesses a memory region that is logically associated with a remote process is equivalent to a put/get in the following rules." And add the following advice to implementors on page 47 line 33 after "... made visible to the other).": "The semantic rules for shared memory windows in the MPI_RMA_SEPARATE model allow concurrent load/store accesses from two distinct processes to the same target window that is remote to both processes. An implementation cannot expose shared memory windows on hardware that does not support this access mode." I'm sure the wording can be tweaked (please propose changes here) but this is what we came up with. Please comment! We plan to discuss this later in the full forum because there are some trade-offs involved here. 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