rma: optimized dynamic window for shm operations

[mpichbot]

Originally by monicasee on 2014-09-03 16:31:51 -0500

---

If user attaches buffers which are allocated by win_allocated_shared, such buffer should be internally detected as a SHM buffer in order to enable fast SHM RMA. However, current implementation does not do that. Also see ticket #2161.

However, we may have to change the method for distinguishing SHM targets in operation functions (check node_id). Because we don't know whether user will attach a private buffer for another local process.

[hzhou]

We currently have very little support for dynamic window optimizations. I am renaming this ticket since detecting or creating mechanism for optimized shm path for dynamic window is the the first step of any optimization