Consolidation with CDI-PIO (develop)

Generally, this is !11 (merged) rebased onto the develop branch.

This is the first develop-based version accepted by the Buildbot of ICON: https://buildbot.dkrz.de/grid?branch=m300488/cdi-pio-2-integration

src/pio_server.c

@@ -170,12 +170,14 @@ setupClientRanks(void)
 static void
 createClientStreamBuf(size_t streamIdx, const struct clientBufSize *bufSizes, const struct cdiPioConf *conf)
 {
+  /* find and tabulate aggregate size needed for all clients of collector */
   size_t streamBufferSize = 0;
   for (size_t i = 0; i < (size_t) numClients_; ++i)
     {
       streamBufferSize += (rxWin[streamIdx].clientBuf[i].size = bufSizes[i].bufSize);
       rxWin[streamIdx].clientBuf[i].dictSize = bufSizes[i].numDataRecords + bufSizes[i].numRPCRecords;
     }
+  /* set pointer to RMA buffer for client 0 of collector */
   if (conf->batchedRMA)
     rxWin[streamIdx].clientBuf[0].mem = Malloc(streamBufferSize);
   else
@@ -191,16 +193,26 @@ createClientStreamBuf(size_t streamIdx, const struct clientBufSize *bufSizes, co
               for (size_t j = 0; j < numStreams; ++j)
                 if (rxWin[j].getWin != MPI_WIN_NULL)
                   {
-                    rxWin[j].clientBuf[0].mem = sharedClientBuf;
-                    for (size_t i = 1; i < (size_t) numClients_; ++i)
-                      rxWin[j].clientBuf[i].mem = rxWin[j].clientBuf[i - 1].mem + rxWin[j].clientBuf[i - 1].size;
+                    unsigned char *newmem = sharedClientBuf;
+                    for (size_t i = 0; i < (size_t) numClients_; ++i)
+                      {
+                        rxWin[j].clientBuf[i].mem = newmem;
+                        newmem += rxWin[j].clientBuf[i].size;
+                      }
                   }
             }
         }
       rxWin[streamIdx].clientBuf[0].mem = sharedClientBuf;
     }
-  for (size_t i = 1; i < (size_t) numClients_; ++i)
-    rxWin[streamIdx].clientBuf[i].mem = rxWin[streamIdx].clientBuf[i - 1].mem + rxWin[streamIdx].clientBuf[i - 1].size;
+  /* set pointers for other clients */
+  {
+    unsigned char *newmem = rxWin[streamIdx].clientBuf[0].mem;
+    for (size_t i = 1; i < (size_t) numClients_; ++i)
+      {
+        newmem += rxWin[streamIdx].clientBuf[i - 1].size;
+        rxWin[streamIdx].clientBuf[i].mem = newmem;
+      }
+  }
 }
 
 static void