To: J3                                                     J3/15-198
From: Nick Maclaren
Subject: Making event ordering consistency explicit
Date: 2015 August 05
Reference: 15-183

Discussion

The problem arises from 7.2p2 "The effect of each change is as if the
atomic subroutine ATOMIC_ADD were executed with a variable that stores
the event count as its ATOM argument."

This reads as if it derives the semantics of EVENT POST and EVENT WAIT
from the semantics of atomic variables, but it does not, because they
need more consistency than is delivered by atomic variables.  The
following wording resolves the ambiguity.


Edits to WG5/N2056:

[17:4+] Add a new paragraph after 7.1p1 (Events/Introduction):

    "Segments ordered by EVENT POST and EVENT WAIT are included in the
    partial order specified in subclause 2.3.5 of ISO/IEC 1539-1:2010."

[17:26+] Add a new Note at the end of 7.2 EVENT_TYPE:

    "NOTE:
    The updates of atomic variables are coherent but not necessarily
    consistent, so a processor might have to use extra synchronization
    to obtain the consistency required for the segments ordered by
    EVENT POST and EVENT WAIT statements."