Add routine for intermediate I/O RDMA progress.

src/cdi.h

@@ -239,6 +239,7 @@ void     pioFinalize           ( void );
 MPI_Comm pioInit(MPI_Comm commSuper, int nProcsIO, int IOMode,
                  int *pioNamespace, float partInflate);
 void     pioWriteTimestep();
+void     cdiPioRDMAProgress();
 
 void     streamWriteVarPart    (int streamID, int varID,
                                 const void *data, int nmiss,

src/cdi.inc

@@ -399,6 +399,9 @@
 !                     pioWriteTimestep
       EXTERNAL        pioWriteTimestep
 
+!                     cdiPioRDMAProgress
+      EXTERNAL        cdiPioRDMAProgress
+
 !                     streamWriteVarPart
 !                                    (INTEGER         streamID,
 !                                     INTEGER         varID,

src/cdiFortran.c

@@ -87,6 +87,7 @@ static int pioInit_fwrap(int commSuper, int nProcsIO, int IOMode, int * pioNames
 }
 FCALLSCFUN5 (INT, pioInit_fwrap, PIOINIT, pioinit, INT, INT, INT, PINT, FLOAT)
 FCALLSCSUB0 (pioWriteTimestep, PIOWRITETIMESTEP, piowritetimestep)
+FCALLSCSUB0 (cdiPioRDMAProgress, CDIPIORDMAPROGRESS, cdipiordmaprogress)
 static void streamWriteVarPart_fwrap(int streamID, int varID, const void * data, int nmiss, void * partDesc)
 {
   streamWriteVarPart( streamID,  varID,  data,  nmiss, (*(Xt_idxlist *)partDesc));

src/pio_interface.c

@@ -573,6 +573,30 @@ collWait(int collID)
     }
 }
 
+static inline void
+collProbe(int collID)
+{
+  if (txWin[collID].postSet)
+    {
+      int flag;
+      xmpi(MPI_Win_test(txWin[collID].win, &flag));
+      if (flag)
+        {
+          txWin[collID].postSet = 0;
+          modelWinFlushBuffer(collID);
+        }
+    }
+}
+
+void
+cdiPioRDMAProgress()
+{
+  int nProcsColl = commInqNProcsColl();
+  for (int collID = 0; collID < nProcsColl; collID++)
+    collProbe(collID);
+}
+
+
 void
 pioBufferPartData(int streamID, int varID, const double *data,
                   int nmiss, Xt_idxlist partDesc)