diff --git a/lib/AlignedAllocator.h b/lib/AlignedAllocator.h
index 44929ca8..fa001adc 100644
--- a/lib/AlignedAllocator.h
+++ b/lib/AlignedAllocator.h
@@ -141,7 +141,7 @@ public:
 
     if ( bcast != ptr ) {
       std::printf("inconsistent alloc pe %d %lx %lx \n",shmem_my_pe(),bcast,ptr);std::fflush(stdout);
-      BACKTRACEFILE();
+      //      BACKTRACEFILE();
       exit(0);
     }
     assert( bcast == (void *) ptr);
diff --git a/lib/Stencil.h b/lib/Stencil.h
index f22acb6f..221785b4 100644
--- a/lib/Stencil.h
+++ b/lib/Stencil.h
@@ -32,8 +32,6 @@
 
  #include <Grid/stencil/Lebesgue.h>   // subdir aggregate
 
-const int ShmDirectCopy = 1;
-
  //////////////////////////////////////////////////////////////////////////////////////////
  // Must not lose sight that goal is to be able to construct really efficient
  // gather to a point stencil code. CSHIFT is not the best way, so need
@@ -170,13 +168,13 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
     reqs.resize(Packets.size());
     commtime-=usecond();
     for(int i=0;i<Packets.size();i++){
-      if( ShmDirectCopy ) {
 	_grid->StencilSendToRecvFromBegin(reqs[i],
 					  Packets[i].send_buf,
 					  Packets[i].to_rank,
 					  Packets[i].recv_buf,
 					  Packets[i].from_rank,
 					  Packets[i].bytes);
+	/*
       }else{
 	_grid->SendToRecvFromBegin(reqs[i],
 				   Packets[i].send_buf,
@@ -185,17 +183,19 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
 				   Packets[i].from_rank,
 				   Packets[i].bytes);
       }
+	*/
     }
     commtime+=usecond();
   }
   void CommunicateComplete(std::vector<std::vector<CommsRequest_t> > &reqs)
   {
     commtime-=usecond();
+
     for(int i=0;i<Packets.size();i++){
-      if( ShmDirectCopy ) 
+      //      if( ShmDirectCopy ) 
 	_grid->StencilSendToRecvFromComplete(reqs[i]);
-      else 
-	_grid->SendToRecvFromComplete(reqs[i]);
+	//      else 
+	//	_grid->SendToRecvFromComplete(reqs[i]);
     }
     commtime+=usecond();
   }
@@ -253,8 +253,6 @@ PARALLEL_FOR_LOOP
   // Flat vector, change layout for cache friendly.
   Vector<StencilEntry>  _entries;
   
-  inline StencilEntry * GetEntry(int &ptype,int point,int osite) { ptype = _permute_type[point]; return & _entries[point+_npoints*osite]; }
-  
   void PrecomputeByteOffsets(void){
     for(int i=0;i<_entries.size();i++){
       if( _entries[i]._is_local ) {
@@ -265,9 +263,7 @@ PARALLEL_FOR_LOOP
     }
   };
 
-  inline uint64_t Touch(int ent) {
-    //	 _mm_prefetch((char *)&_entries[ent],_MM_HINT_T0);
-  }
+  inline StencilEntry * GetEntry(int &ptype,int point,int osite) { ptype = _permute_type[point]; return & _entries[point+_npoints*osite]; }
   inline uint64_t GetInfo(int &ptype,int &local,int &perm,int point,int ent,uint64_t base) {
     uint64_t cbase = (uint64_t)&u_recv_buf_p[0];
     local = _entries[ent]._is_local;
@@ -685,7 +681,9 @@ PARALLEL_FOR_LOOP
     _grid->StencilBarrier();
     HaloGather(source,compress);
     this->CommunicateBegin(reqs);
+    _grid->StencilBarrier();
     this->CommunicateComplete(reqs);
+    _grid->StencilBarrier();
     CommsMerge(); // spins
   }
   
@@ -823,11 +821,13 @@ PARALLEL_FOR_LOOP
 
 
 	cobj *send_buf = (cobj *)_grid->ShmBufferTranslate(xmit_to_rank,u_recv_buf_p);
-	if ( (ShmDirectCopy==0)||send_buf==NULL ) { 
-	  cobj *send_buf = u_send_buf_p;
+	if ( (send_buf==NULL) ) { 
+	  send_buf = u_send_buf_p;
 	}
-	
+	//	std::cout << " send_bufs  "<<std::hex<< send_buf <<" ubp "<<u_send_buf_p <<std::dec<<std::endl;
 	t_data-=usecond();
+	assert(u_send_buf_p!=NULL);
+	assert(send_buf!=NULL);
 	Gather_plane_simple_table         (face_table[face_idx],rhs,send_buf,compress,u_comm_offset,so);  face_idx++;
 	t_data+=usecond();
 	
@@ -931,7 +931,8 @@ PARALLEL_FOR_LOOP
 	    _grid->ShiftedRanks(dimension,nbr_proc,xmit_to_rank,recv_from_rank); 
  
 	    scalar_object *shm = (scalar_object *) _grid->ShmBufferTranslate(recv_from_rank,sp);
-	    if ((ShmDirectCopy==0)||(shm==NULL)) { 
+	    //	    if ((ShmDirectCopy==0)||(shm==NULL)) { 
+	    if (shm==NULL) { 
 	      shm = rp;
 	    } 
 	    
diff --git a/lib/communicator/Communicator_base.cc b/lib/communicator/Communicator_base.cc
index 1272b6a2..91e9cf9b 100644
--- a/lib/communicator/Communicator_base.cc
+++ b/lib/communicator/Communicator_base.cc
@@ -69,7 +69,7 @@ int                      CartesianCommunicator::ProcessorCount(void)    { return
 ////////////////////////////////////////////////////////////////////////////////
 // very VERY rarely (Log, serial RNG) we need world without a grid
 ////////////////////////////////////////////////////////////////////////////////
-int  CartesianCommunicator::RankWorld(void) { return WorldRank; };
+int  CartesianCommunicator::RankWorld(void){ return WorldRank; };
 int CartesianCommunicator::Ranks    (void) { return WorldSize; };
 int CartesianCommunicator::Nodes    (void) { return GroupSize; };
 int CartesianCommunicator::Cores    (void) { return ShmSize;   };
@@ -108,22 +108,22 @@ void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsReque
 {
   SendToRecvFromComplete(waitall);
 }
-void StencilBarrier(void){};
+void CartesianCommunicator::StencilBarrier(void){};
 
 commVector<uint8_t> CartesianCommunicator::ShmBufStorageVector;
 
 void *CartesianCommunicator::ShmBufferSelf(void) { return ShmCommBuf; }
+
 void *CartesianCommunicator::ShmBuffer(int rank) {
-  if (rank != ShmRank ) return NULL;
-  else                  return ShmCommBuf;
+  return NULL;
 }
 void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p) { 
-  if (rank != ShmRank ) return NULL;
-  else                  return local_p;
+  return NULL;
 }
 void CartesianCommunicator::ShmInitGeneric(void){
   ShmBufStorageVector.resize(MAX_MPI_SHM_BYTES);
   ShmCommBuf=(void *)&ShmBufStorageVector[0];
+  std::cout << "allocated persistent buffer"<<std::hex << ShmCommBuf << std::dec<<std::endl;
 }
 
 #endif
diff --git a/lib/communicator/Communicator_mpi3.cc b/lib/communicator/Communicator_mpi3.cc
index 00b0ca11..61e5e06b 100644
--- a/lib/communicator/Communicator_mpi3.cc
+++ b/lib/communicator/Communicator_mpi3.cc
@@ -400,7 +400,9 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
 						int from,
 						int bytes)
 {
-#if 1
+#if 0
+  this->StencilBarrier();
+
   MPI_Request xrq;
   MPI_Request rrq;
   
@@ -440,9 +442,6 @@ void CartesianCommunicator::SendToRecvFromBegin(std::vector<CommsRequest_t> &lis
 PARALLEL_FOR_LOOP 
     for(int w=0;w<words;w++) {
       op[w]=ip[w];
-      if ( w == 0 ) { 
-	//	std::cout << " xmit "<< ShmRank <<" -> "<< gdest<<" " <<std::hex<<op[w]<<std::dec<<std::endl;
-      }
     }
 
     bcopy(&_processor,&to_ptr[bytes],sizeof(_processor));
@@ -452,20 +451,15 @@ PARALLEL_FOR_LOOP
     assert(ierr==0);
     list.push_back(xrq);
   }
-  
-  MPI_Win_sync (ShmWindow);   
-  MPI_Barrier  (ShmComm);
-  MPI_Win_sync (ShmWindow);   
 
+  this->StencilBarrier();
+  
   if (small && (gfrom !=MPI_UNDEFINED) ) {
     T *ip = (T *)from_ptr;
     T *op = (T *)recv;
 PARALLEL_FOR_LOOP 
     for(int w=0;w<words;w++) {
       op[w]=ip[w];
-      if ( w == 0 ) { 
-	//	std::cout << " recv "<< ShmRank <<" <- "<< gfrom<<" " <<std::hex<<op[w]<<std::dec<<std::endl;
-      }
     }
     bcopy(&from_ptr[bytes]  ,&tag  ,sizeof(tag));
     bcopy(&from_ptr[bytes+4],&check,sizeof(check));
@@ -477,9 +471,8 @@ PARALLEL_FOR_LOOP
     list.push_back(rrq);
   }
 
-  MPI_Win_sync (ShmWindow);   
-  MPI_Barrier  (ShmComm);
-  MPI_Win_sync (ShmWindow);   
+  this->StencilBarrier();
+
 #else
   MPI_Request xrq;
   MPI_Request rrq;
@@ -528,9 +521,6 @@ void CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_
     list.push_back(rrq);
   }
 
-
-  StencilBarrier();
-
 }
 
 
diff --git a/lib/communicator/Communicator_none.cc b/lib/communicator/Communicator_none.cc
index d32fe4fa..f4e4c575 100644
--- a/lib/communicator/Communicator_none.cc
+++ b/lib/communicator/Communicator_none.cc
@@ -47,8 +47,8 @@ void CartesianCommunicator::Init(int *argc, char *** arv)
   WorldSize = 1;
   ShmRank=0;
   ShmSize=1;
-  GroupRank=_WorldRank;
-  GroupSize=_WorldSize;
+  GroupRank=WorldRank;
+  GroupSize=WorldSize;
   Slave    =0;
   ShmInitGeneric();
 }
diff --git a/lib/communicator/Communicator_shmem.cc b/lib/communicator/Communicator_shmem.cc
index 544b37c7..d5f3ed76 100644
--- a/lib/communicator/Communicator_shmem.cc
+++ b/lib/communicator/Communicator_shmem.cc
@@ -44,13 +44,16 @@ namespace Grid {
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 // Info that is setup once and indept of cartesian layout
 ///////////////////////////////////////////////////////////////////////////////////////////////////
-int CartesianCommunicator::ShmRank;
-int CartesianCommunicator::ShmSize;
-int CartesianCommunicator::GroupRank;
-int CartesianCommunicator::GroupSize;
-int CartesianCommunicator::WorldRank;
-int CartesianCommunicator::WorldSize;
-int CartesianCommunicator::Slave;
+
+typedef struct HandShake_t { 
+  uint64_t seq_local;
+  uint64_t seq_remote;
+} HandShake;
+
+
+static Vector< HandShake > XConnections;
+static Vector< HandShake > RConnections;
+
 
 void CartesianCommunicator::Init(int *argc, char ***argv) {
   shmem_init();
@@ -62,37 +65,17 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
     RConnections[pe].seq_local = 0;
     RConnections[pe].seq_remote= 0;
   }
+  WorldSize = shmem_n_pes();
+  WorldRank = shmem_my_pe();
+  ShmRank=0;
+  ShmSize=1;
+  GroupRank=WorldRank;
+  GroupSize=WorldSize;
+  Slave    =0;
   shmem_barrier_all();
   ShmInitGeneric();
 }
 
-
-// Should error check all MPI calls.
-void CartesianCommunicator::Init(int *argc, char ***argv) {
-  int flag;
-  MPI_Initialized(&flag); // needed to coexist with other libs apparently
-  if ( !flag ) {
-    MPI_Init(argc,argv);
-    MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world);
-    MPI_Comm_rank(communicator_world,&_WorldRank);
-    MPI_Comm_size(communicator_world,&_WorldSize);
-    _ShmRank=0;
-    _ShmSize=1;
-    _GroupRank=_WorldRank;
-    _GroupSize=_WorldSize;
-    _Slave    =0;
-  }
-}
-
-
-typedef struct HandShake_t { 
-  uint64_t seq_local;
-  uint64_t seq_remote;
-} HandShake;
-
-static Vector< HandShake > XConnections;
-static Vector< HandShake > RConnections;
-
 CartesianCommunicator::CartesianCommunicator(const std::vector<int> &processors)
 {
   _ndimension = processors.size();
@@ -261,12 +244,9 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
 
   if ( _processor == sender ) {
 
-    printf("Sender SHMEM pt2pt %d -> %d\n",sender,receiver);
     // Check he has posted a receive
     while(SendSeq->seq_remote == SendSeq->seq_local);
 
-    printf("Sender receive %d posted\n",sender,receiver);
-
     // Advance our send count
     seq = ++(SendSeq->seq_local);
     
@@ -275,26 +255,19 @@ void CartesianCommunicator::SendRecvPacket(void *xmit,
     shmem_putmem(recv,xmit,bytes,receiver);
     shmem_fence();
 
-    printf("Sender sent payload %d\n",seq);
     //Notify him we're done
     shmem_putmem((void *)&(RecvSeq->seq_remote),&seq,sizeof(seq),receiver);
     shmem_fence();
-    printf("Sender ringing door bell  %d\n",seq);
   }
   if ( _processor == receiver ) {
 
-    printf("Receiver SHMEM pt2pt %d->%d\n",sender,receiver);
     // Post a receive
     seq = ++(RecvSeq->seq_local);
     shmem_putmem((void *)&(SendSeq->seq_remote),&seq,sizeof(seq),sender);
 
-    printf("Receiver Opening letter box %d\n",seq);
-
-    
     // Now wait until he has advanced our reception counter
     while(RecvSeq->seq_remote != RecvSeq->seq_local);
 
-    printf("Receiver Got the mail %d\n",seq);
   }
 }