diff --git a/Grid/communicator/Communicator_mpi3.cc b/Grid/communicator/Communicator_mpi3.cc
index b667d32e..38b9f9c6 100644
--- a/Grid/communicator/Communicator_mpi3.cc
+++ b/Grid/communicator/Communicator_mpi3.cc
@@ -746,26 +746,34 @@ double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsReques
 }
 void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsRequest_t> &list,int dir)
 {
-  //  int nreq=list.size();
+  acceleratorCopySynchronise(); // Complete all pending copy transfers D2D
 
-  //  if (nreq==0) return;
-  //  std::vector<MPI_Status> status(nreq);
-  //  std::vector<MPI_Request> MpiRequests(nreq);
+  std::vector<MPI_Status> status;
+  std::vector<MPI_Request> MpiRequests;
+    
+  for(int r=0;r<list.size();r++){
+    // Must check each Send buf is clear to reuse
+    if ( list[r].PacketType == InterNodeXmitISend ) MpiRequests.push_back(list[r].req);
+    //    if ( list[r].PacketType == InterNodeRecv ) MpiRequests.push_back(list[r].req); // Already "Test" passed
+  }
 
-  //  for(int r=0;r<nreq;r++){
-  //    MpiRequests[r] = list[r].req;
-  //  }
+  int nreq=MpiRequests.size();
+
+  std::cout << GridLogMessage << " StencilSendToRevFromComplete "<<nreq<<" Mpi Requests"<<std::endl;
+
+
+  if (nreq>0) {
+    status.resize(MpiRequests.size());
+    int ierr = MPI_Waitall(MpiRequests.size(),&MpiRequests[0],&status[0]); // Sends are guaranteed in order. No harm in not completing.
+    assert(ierr==0);
+  }
   
-  //  int ierr = MPI_Waitall(nreq,&MpiRequests[0],&status[0]); // Sends are guaranteed in order. No harm in not completing.
-  //  assert(ierr==0);
-
   //  for(int r=0;r<nreq;r++){
   //    if ( list[r].PacketType==InterNodeRecv ) {
   //      acceleratorCopyToDeviceAsynch(list[r].host_buf,list[r].device_buf,list[r].bytes);
   //    }
   //  }
   
-  acceleratorCopySynchronise(); // Complete all pending copy transfers D2D
   
   list.resize(0);               // Delete the list
   this->HostBufferFreeAll();    // Clean up the buffer allocs
diff --git a/Grid/threads/Accelerator.h b/Grid/threads/Accelerator.h
index b5aaccb4..28c3aa0a 100644
--- a/Grid/threads/Accelerator.h
+++ b/Grid/threads/Accelerator.h
@@ -676,7 +676,7 @@ inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
 
 template<class T> void acceleratorPut(T& dev,const T&host)
 {
-  acceleratorCopyToDevice(&host,&dev,sizeof(T));
+  acceleratorCopyToDevice((void *)&host,&dev,sizeof(T));
 }
 template<class T> T acceleratorGet(T& dev)
 {
diff --git a/Grid/util/Init.cc b/Grid/util/Init.cc
index 1424667e..feb44645 100644
--- a/Grid/util/Init.cc
+++ b/Grid/util/Init.cc
@@ -509,7 +509,14 @@ void Grid_init(int *argc,char ***argv)
 		  Grid_default_latt,
 		  Grid_default_mpi);
 
-
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--flightrecorder") ){
+    std::cout << GridLogMessage <<" Enabling flight recorder " <<std::endl;
+    FlightRecorder::SetLoggingMode(FlightRecorder::LoggingModeRecord);
+    FlightRecorder::PrintEntireLog = 1;
+    FlightRecorder::ChecksumComms  = 1;
+    FlightRecorder::ChecksumCommsSend=1;
+  }
+  
   if( GridCmdOptionExists(*argv,*argv+*argc,"--decomposition") ){
     std::cout<<GridLogMessage<<"Grid Default Decomposition patterns\n";
     std::cout<<GridLogMessage<<"\tOpenMP threads : "<<GridThread::GetThreads()<<std::endl;
@@ -651,3 +658,4 @@ void Grid_debug_handler_init(void)
 }
 
 NAMESPACE_END(Grid);
+