diff --git a/Grid/communicator/SharedMemoryMPI.cc b/Grid/communicator/SharedMemoryMPI.cc
index 0ee55a1c..8b27ab7a 100644
--- a/Grid/communicator/SharedMemoryMPI.cc
+++ b/Grid/communicator/SharedMemoryMPI.cc
@@ -448,7 +448,11 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   // Each MPI rank should allocate our own buffer
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
+#ifndef GRID_MPI3_SHM_NONE
   auto err =  cudaMalloc(&ShmCommBuf, bytes);
+#else
+  auto err =  cudaMallocManaged(&ShmCommBuf, bytes);
+#endif
   if ( err !=  cudaSuccess) {
     std::cerr << " SharedMemoryMPI.cc cudaMallocManaged failed for " << bytes<<" bytes " <<cudaGetErrorString(err)<< std::endl;
     exit(EXIT_FAILURE);  
@@ -466,7 +470,8 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
   // Loop over ranks/gpu's on our node
   ///////////////////////////////////////////////////////////////////////////////////////////////////////////
   for(int r=0;r<WorldShmSize;r++){
-    
+
+#ifndef GRID_MPI3_SHM_NONE
     //////////////////////////////////////////////////
     // If it is me, pass around the IPC access key
     //////////////////////////////////////////////////
@@ -506,6 +511,9 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
     // Save a copy of the device buffers
     ///////////////////////////////////////////////////////////////
     WorldShmCommBufs[r] = thisBuf;
+#else
+    WorldShmCommBufs[r] = ShmCommBuf;
+#endif
   }
 
   _ShmAllocBytes=bytes;
diff --git a/tests/Test_stencil.cc b/tests/Test_stencil.cc
index 16a9138a..73fa5d78 100644
--- a/tests/Test_stencil.cc
+++ b/tests/Test_stencil.cc
@@ -97,11 +97,15 @@ int main(int argc, char ** argv) {
 	  ocoor[dir]=(ocoor[dir]+disp)%Fine._rdimensions[dir];
 	}
 
+	std::cout << GridLogMessage << "A" << std::endl;
+
 	SimpleCompressor<vobj> compress;
 	myStencil.HaloExchange(Foo,compress);
 
 	Bar = Cshift(Foo,dir,disp);
 
+	std::cout << GridLogMessage << "B" << std::endl;
+
 	// Implement a stencil code that should agree with cshift!
 	for(int i=0;i<Check.Grid()->oSites();i++){
 	  
@@ -117,8 +121,8 @@ int main(int argc, char ** argv) {
 	    check[i] = foo[SE->_offset];
 	  else { 
 	    check[i] = myStencil.CommBuf()[SE->_offset];
-	    //	    std::cout << " receive "<<i<<" " << Check[i]<<std::endl;
-	    //	    std::cout << " Foo     "<<i<<" " <<   Foo[i]<<std::endl;
+	    	    std::cout << " receive "<<i<<" " << check[i]<<std::endl;
+	    	    std::cout << " Foo     "<<i<<" " <<   foo[i]<<std::endl;
 	  }
 	}