Introduce support for local complete

content/shmem_local_complete.tex

@@ -0,0 +1,52 @@
+\apisummary{
+    Waits for data to be copied out of the \VAR{source} array on all 
+    outstanding non-blocking \OPR{Put} and non-blocking \OPR{put-with-signal} 
+    issued by a \ac{PE}.
+}
+
+\begin{apidefinition}
+
+\begin{Csynopsis}
+void @\FuncDecl{shmem\_local\_complete}@(void);
+void @\FuncDecl{shmem\_ctx\_local\_complete}@(shmem_ctx_t ctx);
+\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.}
+\end{apiarguments}
+
+\apidescription{
+    The \FUNC{shmem\_local\_complete} routine ensures local completion of all 
+    non-blocking \OPR{Put} and non-blocking \OPR{put-with-signal} operations 
+    issued by a \ac{PE}. Local completion guarantees the reusability of the
+    \VAR{source} buffers associated with a \ac{PE} issuing the operation. 
+    Local completion does not guarantee any ordering and/or delivery of 
+    completion with any visibility guarantees on all \acp{PE}. 
+    Return from \FUNC{shmem\_local\_complete} just guarantees that the data has
+    been copied out of the \VAR{source} array on all previously posted
+    non-blocking \OPR{Put} and non-blocking \OPR{put-with-signal} operations
+    in the local \ac{PE}. Memory ordering routines as supported in
+    Section~\ref{subsec:memory_order} is still required to provide
+    mechanims to ensure ordering and/or delivery of completions on the
+    non-blocking \OPR{Put} and non-blocking \OPR{put-with-signal} operations.
+}
+
+
+\apireturnvalues{
+    None.
+}
+
+\begin{apiexamples}
+
+\apicexample
+    {The following example uses \FUNC{shmem\_quiet} in a \Cstd[11] program: }
+    {./example_code/shmem_local_complete_example.c}
+    {\FUNC{shmem\_local\_complete} allows reusing the \VAR{source} buffer
+    without waiting for the completion and global visibility on target process
+    \VAR{tpe}}
+\end{apiexamples}
+
+\end{apidefinition}
+

example_code/shmem_local_complete_example.c

@@ -0,0 +1,40 @@
+#include <shmem.h>
+#include <stdio.h>
+
+#define SIZE 10
+
+int main(void) {
+  int count = 0;
+  shmem_init();
+  static int64_t sig_addr = 0;
+  int *dest = shmem_malloc (SIZE * sizeof(int));
+  int *source = shmem_malloc (SIZE * sizeof(int));
+  int mype = shmem_my_pe();
+  int npes = shmem_n_pes();
+
+  for (int i = 0; i < SIZE; i++) {
+    source[i] = count;
+    dest[i] = 0;
+  }
+
+  if (mype == 0) {
+    for (int tpe = 1; tpe < npes; tpe++) {
+      shmem_put_signal_nbi(dest, source, SIZE, &sig_addr, 1, SHMEM_SIGNAL_SET, tpe);
+      shmem_local_complete();
+      count++;
+      for (int i = 0; i < SIZE; i++) {
+        source[i] = count;
+      }
+    }
+  } else {
+    shmem_wait_until(&sig_addr, SHMEM_CMP_EQ, 1);
+    for (int i = 0; i < SIZE; i++) {
+      if (dest[i] != mype) {
+        count++;
+      }
+    }
+    if (count) fprintf(stderr, "Program Error\n");
+  }
+  shmem_finalize();
+  return 0;
+}

main_spec.tex

@@ -486,9 +486,18 @@ \subsubsection{Synchronization and Communication Ordering in OpenSHMEM}
 \input{content/synchronization_model.tex}
 
 
-
-
-
+\color{teal}
+\subsection{Local Completion Operations}\label{sec:lcomplete}
+This section specifies the OpenSHMEM support for \OPR{local-complete} 
+operation. The \OPR{local-complete} operation provide a method for reusing 
+the \VAR{source} buffers associated with the process initiating the 
+non-blocking remote memory access routines and signaling operations as 
+specified in Section~\ref{subsec:shmem_put_nbi} and
+Section~\ref{subsec:shmem_put_signal_nbi} respectively. 
+
+\subsubsection{\textbf{SHMEM\_LOCAL\_COMPLETE}}\label{subsec:shmem_local_complete}
+\input{content/shmem_local_complete.tex}
+\color{black}
 
 \subsection{Distributed Locking Routines}
 The following section discusses \openshmem locks as a mechanism to provide