Merge branch 'master' into feature/sig-add-set

content/backmatter.tex

@@ -664,6 +664,10 @@ \section{Version 1.6}
   update a remote flag without associated data transfer of a put-with-signal operation.
 \ChangelogRef{subsec:shmem_signal_add, subsec:shmem_signal_set}%
 %
+\item Clarified that \OPR{Fence} operations only guarantee ordering for
+    operations that are performed on the same context.
+\ChangelogRef{subsec:shmem_fence}%
+%
 \item Added a team-based pointer query routine:
   \FUNC{shmem\_team\_ptr}.
 \ChangelogRef{subsec:shmem_team_ptr}%
@@ -678,10 +682,19 @@ \section{Version 1.6}
     operations for team-based reductions.
 \ChangelogRef{teamreducetypes}%
 %
+\item Added fine grained completion routine: \FUNC{shmem\_pe\_quiet}.
+\ChangelogRef{subsec:shmem_pe_quiet}%
+%
 \item Split the listings for the \FUNC{shmem\_\{malloc, free, realloc, align\}}
   functions from a single entry in \openshmem[1.5] into separate entries.
 \ChangelogRef{subsec:shmem_malloc, subsec:shmem_free, subsec:shmem_realloc,
   subsec:shmem_align}%
+\item Corrected the level argument's recommended value in API notes for
+    \FUNC{shmem\_pcontrol} to indicate that the value should be greater than
+    2 to enable profiling with profile library defined effects and
+    additional arguments.
+\ChangelogRef{subsec:shmem_pcontrol}
+%
 \end{itemize}
 
 \section{Version 1.5}

content/shmem_fence.tex

@@ -24,7 +24,7 @@
     \ac{PE} on the given context prior
     to the call to \FUNC{shmem\_fence} are guaranteed to be delivered before any
     subsequent operations on symmetric data
-    objects to the same \ac{PE}. \FUNC{shmem\_fence} guarantees order of delivery,
+    objects to the same \ac{PE} on the same context. \FUNC{shmem\_fence} guarantees order of delivery,
     not completion. It does not guarantee order of delivery of nonblocking
     \GET{} or values fetched by nonblocking \ac{AMO} routines.
     If \VAR{ctx} has the value \CONST{SHMEM\_CTX\_INVALID}, no operation is

content/shmem_pcontrol.tex

@@ -37,7 +37,7 @@
   \item \texttt{level == 1} Profiling is enabled at the default level of detail.
   \item \texttt{level == 2} Profiling is enabled and profile buffers are 
   flushed if available.
-  \item \texttt{level >= 2} Profiling is enabled with profile library defined 
+  \item \texttt{level > 2} Profiling is enabled with profile library defined 
   effects and additional arguments.
   \end{itemize}
 

content/shmem_pe_quiet.tex

@@ -0,0 +1,47 @@
+\apisummary{
+    Waits for completion of all outstanding memory store, blocking 
+    \PUT{}, \ac{AMO}, and \emph{put-with-signal}, as well as 
+    nonblocking \PUT{}, \emph{put-with-signal}, and \GET{} routines 
+    to symmetric data objects issued by the calling \ac{PE} at the target
+    \acp{PE}.
+}
+
+\begin{apidefinition}
+
+\begin{Csynopsis}
+void @\FuncDecl{shmem\_pe\_quiet}@(const int *target_pes, size_t npes);
+void @\FuncDecl{shmem\_ctx\_pe\_quiet}@(shmem_ctx_t ctx, const int *target_pes, size_t npes);
+\end{Csynopsis}
+
+\begin{apiarguments}
+    \apiargument{IN}{ctx}{A context handle specifying the context on which to perform the operation.
+        When this argument is not provided, the operation is performed on
+        the default context.}
+	\apiargument{IN}{target\_pes}{Address of target \ac{PE} array where the
+    operations need to be completed}
+	\apiargument{IN}{npes}{The number of \acp{PE} in the target \ac{PE} array}
+
+\end{apiarguments}
+
+\apidescription{
+    The \FUNC{shmem\_pe\_quiet} ensures completion of memory store, blocking 
+    \PUT{}, \ac{AMO}, and \emph{put-with-signal}, as well as  nonblocking
+    \PUT{}, \emph{put-with-signal}, and \GET{} routines on the symmetric data
+    objects issued by the calling \ac{PE} to the target \acp{PE} and on the
+    given context. If \VAR{npes} is set to 0, the \VAR{target\_pes} is ignored
+    and the routine returns immediately.
+
+	The completion and visibility semantics of these operations are the same as the
+	\FUNC{shmem\_quiet} routine. However, it applies only to the target
+    \acp{PE}, i.e., the operations to the target \acp{PE} are guaranteed to be
+    complete and visible to all \acp{PE} when \FUNC{shmem\_pe\_quiet} returns.
+}
+\apireturnvalues{
+    None.
+}
+
+\apinotes{
+    On certain platforms, when \FUNC{shmem\_pe\_quiet} is invoked on a set of
+    \acp{PE}, the performance might be equivalent to \FUNC{shmem\_quiet}.
+}
+\end{apidefinition}

content/shmem_reductions.tex

@@ -283,9 +283,11 @@ \subsubsubsection{PROD}
     provides one element for each reduction.  The results of the reductions are placed in the
     \dest{} array on all \acp{PE} participating in the reduction.
 
+    The same \source{} and \dest{} arrays must be passed by all PEs that
+    participate in the collective.
     The \source{} and \dest{} arguments must either be the same symmetric
     address, or two different symmetric addresses corresponding to buffers that
-    do not overlap in memory. That is, they must be completely overlapping or
+    do not overlap in memory. That is, they must be completely overlapping (sometimes referred to as an ``in place'' reduction) or
     completely disjoint.
 
     Team-based reduction routines operate over all \acp{PE} in the provided team argument. All

main_spec.tex

@@ -493,6 +493,9 @@ \subsubsection{\textbf{SHMEM\_FENCE}}\label{subsec:shmem_fence}
 \subsubsection{\textbf{SHMEM\_QUIET}}\label{subsec:shmem_quiet}
 \input{content/shmem_quiet.tex}
 
+\subsubsection{\textbf{SHMEM\_PE\_QUIET}}\label{subsec:shmem_pe_quiet}
+\input{content/shmem_pe_quiet.tex}
+
 \subsubsection{Synchronization and Communication Ordering in OpenSHMEM}
 \input{content/synchronization_model.tex}