diff --git a/benchmarks/Benchmark_comms.cc b/benchmarks/Benchmark_comms.cc
index de73bc81..969a2a42 100644
--- a/benchmarks/Benchmark_comms.cc
+++ b/benchmarks/Benchmark_comms.cc
@@ -193,6 +193,7 @@ int main (int argc, char ** argv)
     }
   }  
 
+
   Nloop=100;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking concurrent STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
@@ -271,5 +272,90 @@ int main (int argc, char ** argv)
       std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
     }
   }    
+
+
+
+  Nloop=100;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking sequential STENCIL halo exchange in "<<nmu<<" dimensions"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
+
+  for(int lat=4;lat<=maxlat;lat+=2){
+    for(int Ls=1;Ls<=16;Ls*=2){
+
+      std::vector<int> latt_size  ({lat*mpi_layout[0],
+      				    lat*mpi_layout[1],
+      				    lat*mpi_layout[2],
+      				    lat*mpi_layout[3]});
+
+      GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+
+      std::vector<HalfSpinColourVectorD *> xbuf(8);
+      std::vector<HalfSpinColourVectorD *> rbuf(8);
+      Grid.ShmBufferFreeAll();
+      for(int d=0;d<8;d++){
+	xbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+	rbuf[d] = (HalfSpinColourVectorD *)Grid.ShmBufferMalloc(lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD));
+      }
+
+      int ncomm;
+      int bytes=lat*lat*lat*Ls*sizeof(HalfSpinColourVectorD);
+
+      double start=usecond();
+      for(int i=0;i<Nloop;i++){
+
+	std::vector<CartesianCommunicator::CommsRequest_t> requests;
+
+	ncomm=0;
+	for(int mu=0;mu<4;mu++){
+	
+	  if (mpi_layout[mu]>1 ) {
+	  
+	    ncomm++;
+	    int comm_proc=1;
+	    int xmit_to_rank;
+	    int recv_from_rank;
+	    
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.StencilSendToRecvFromBegin(requests,
+					    (void *)&xbuf[mu][0],
+					    xmit_to_rank,
+					    (void *)&rbuf[mu][0],
+					    recv_from_rank,
+					    bytes);
+	    //	    Grid.StencilSendToRecvFromComplete(requests);
+	    //	    requests.resize(0);
+
+	    comm_proc = mpi_layout[mu]-1;
+	  
+	    Grid.ShiftedRanks(mu,comm_proc,xmit_to_rank,recv_from_rank);
+	    Grid.StencilSendToRecvFromBegin(requests,
+					    (void *)&xbuf[mu+4][0],
+					    xmit_to_rank,
+					    (void *)&rbuf[mu+4][0],
+					    recv_from_rank,
+					    bytes);
+	    Grid.StencilSendToRecvFromComplete(requests);
+	    requests.resize(0);
+	  
+	  }
+	}
+	Grid.Barrier();
+
+      }
+      double stop=usecond();
+
+      double dbytes    = bytes;
+      double xbytes    = Nloop*dbytes*2.0*ncomm;
+      double rbytes    = xbytes;
+      double bidibytes = xbytes+rbytes;
+
+      double time = stop-start; // microseconds
+
+      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
+    }
+  }    
+
   Grid_finalize();
 }
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index 5d42a1f9..10e4521b 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -57,7 +57,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   std::vector<int> latt4 = GridDefaultLatt();
-  const int Ls=16;
+  const int Ls=8;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
   GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
@@ -138,7 +138,7 @@ int main (int argc, char ** argv)
 
   int ncall =100;
   if (1) {
-
+    FGrid->Barrier();
     Dw.ZeroCounters();
     double t0=usecond();
     for(int i=0;i<ncall;i++){
@@ -147,6 +147,7 @@ int main (int argc, char ** argv)
       __SSC_STOP;
     }
     double t1=usecond();
+    FGrid->Barrier();
     
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=1344*volume*ncall;
@@ -158,7 +159,7 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage << "mflop/s per rank =  "<< flops/(t1-t0)/NP<<std::endl;
     err = ref-result; 
     std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-    assert (norm2(err)< 1.0e-5 );
+    assert (norm2(err)< 1.0e-4 );
     Dw.Report();
   }
 
@@ -193,6 +194,7 @@ int main (int argc, char ** argv)
       pokeSite(tmp,ssrc,site);
     }}}}}
     std::cout<<GridLogMessage<< "src norms "<< norm2(src)<<" " <<norm2(ssrc)<<std::endl;
+    FGrid->Barrier();
     double t0=usecond();
     sDw.ZeroCounters();
     for(int i=0;i<ncall;i++){
@@ -201,6 +203,7 @@ int main (int argc, char ** argv)
       __SSC_STOP;
     }
     double t1=usecond();
+    FGrid->Barrier();
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=1344*volume*ncall;
 
@@ -211,12 +214,12 @@ int main (int argc, char ** argv)
   
     if(0){
       for(int i=0;i< PerformanceCounter::NumTypes(); i++ ){
-  sDw.Dhop(ssrc,sresult,0);
-  PerformanceCounter Counter(i);
-  Counter.Start();
-  sDw.Dhop(ssrc,sresult,0);
-  Counter.Stop();
-  Counter.Report();
+	sDw.Dhop(ssrc,sresult,0);
+	PerformanceCounter Counter(i);
+	Counter.Start();
+	sDw.Dhop(ssrc,sresult,0);
+	Counter.Stop();
+	Counter.Report();
       }
     }
 
@@ -240,7 +243,7 @@ int main (int argc, char ** argv)
       }
     }}}}}
     std::cout<<GridLogMessage<<" difference between normal and simd is "<<sum<<std::endl;
-    assert (sum< 1.0e-5 );
+    assert (sum< 1.0e-4 );
 
 
     if (1) {
@@ -271,6 +274,7 @@ int main (int argc, char ** argv)
       if ( WilsonKernelsStatic::Opt == WilsonKernelsStatic::OptInlineAsm ) std::cout << GridLogMessage<< "* Using Asm Nc=3   WilsonKernels" <<std::endl;
       std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
 
+      FGrid->Barrier();
       sDw.ZeroCounters();
       sDw.stat.init("DhopEO");
       double t0=usecond();
@@ -278,6 +282,7 @@ int main (int argc, char ** argv)
         sDw.DhopEO(ssrc_o, sr_e, DaggerNo);
       }
       double t1=usecond();
+      FGrid->Barrier();
       sDw.stat.print();
 
       double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
@@ -301,7 +306,7 @@ int main (int argc, char ** argv)
 
       error+= norm2(ssrc_o);
       std::cout<<GridLogMessage << "sO norm diff   "<< norm2(ssrc_o)<< "  vec nrm"<<norm2(sr_o) <<std::endl;
-      if(error>1.0e-5) { 
+      if(error>1.0e-4) { 
 	setCheckerboard(ssrc,ssrc_o);
 	setCheckerboard(ssrc,ssrc_e);
 	std::cout<< ssrc << std::endl;
@@ -337,7 +342,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   err = ref-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  assert(norm2(err)<1.0e-5);
+  assert(norm2(err)<1.0e-4);
   LatticeFermion src_e (FrbGrid);
   LatticeFermion src_o (FrbGrid);
   LatticeFermion r_e   (FrbGrid);
@@ -363,11 +368,13 @@ int main (int argc, char ** argv)
   std::cout << GridLogMessage<< "*********************************************************" <<std::endl;
   {
     Dw.ZeroCounters();
+    FGrid->Barrier();
     double t0=usecond();
     for(int i=0;i<ncall;i++){
       Dw.DhopEO(src_o,r_e,DaggerNo);
     }
     double t1=usecond();
+    FGrid->Barrier();
     
     double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
     double flops=(1344.0*volume*ncall)/2;
@@ -389,14 +396,14 @@ int main (int argc, char ** argv)
 
   err = r_eo-result; 
   std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
-  assert(norm2(err)<1.0e-5);
+  assert(norm2(err)<1.0e-4);
 
   pickCheckerboard(Even,src_e,err);
   pickCheckerboard(Odd,src_o,err);
   std::cout<<GridLogMessage << "norm diff even  "<< norm2(src_e)<<std::endl;
   std::cout<<GridLogMessage << "norm diff odd   "<< norm2(src_o)<<std::endl;
-  assert(norm2(src_e)<1.0e-5);
-  assert(norm2(src_o)<1.0e-5);
+  assert(norm2(src_e)<1.0e-4);
+  assert(norm2(src_o)<1.0e-4);
 
   Grid_finalize();
 }
diff --git a/benchmarks/Benchmark_dwf_sweep.cc b/benchmarks/Benchmark_dwf_sweep.cc
index a3803b87..21db823d 100644
--- a/benchmarks/Benchmark_dwf_sweep.cc
+++ b/benchmarks/Benchmark_dwf_sweep.cc
@@ -69,8 +69,8 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "Volume \t\t\tProcs \t Dw \t eoDw \t sDw \t eosDw (Mflop/s)  "<<std::endl;
   std::cout<<GridLogMessage << "=========================================================================="<<std::endl;
 
-  int Lmax=32;
-  int dmin=0;
+  int Lmax=16;
+  int dmin=2;
   if ( getenv("LMAX") ) Lmax=atoi(getenv("LMAX"));
   if ( getenv("DMIN") ) dmin=atoi(getenv("DMIN"));
   for (int L=8;L<=Lmax;L*=2){
diff --git a/lib/Init.cc b/lib/Init.cc
index b5ef0a95..d6d6b9f8 100644
--- a/lib/Init.cc
+++ b/lib/Init.cc
@@ -369,7 +369,7 @@ void Grid_init(int *argc,char ***argv)
   
 void Grid_finalize(void)
 {
-#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3)
+#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPI3) 
   MPI_Finalize();
   Grid_unquiesce_nodes();
 #endif
diff --git a/lib/Log.cc b/lib/Log.cc
index 733bfc58..7521657b 100644
--- a/lib/Log.cc
+++ b/lib/Log.cc
@@ -93,7 +93,7 @@ void GridLogConfigure(std::vector<std::string> &logstreams) {
 ////////////////////////////////////////////////////////////
 void Grid_quiesce_nodes(void) {
   int me = 0;
-#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3)
+#if defined(GRID_COMMS_MPI) || defined(GRID_COMMS_MPI3) || defined(GRID_COMMS_MPI3L)
   MPI_Comm_rank(MPI_COMM_WORLD, &me);
 #endif
 #ifdef GRID_COMMS_SHMEM
diff --git a/lib/algorithms/iterative/ConjugateGradient.h b/lib/algorithms/iterative/ConjugateGradient.h
index f340eb38..cf3872c8 100644
--- a/lib/algorithms/iterative/ConjugateGradient.h
+++ b/lib/algorithms/iterative/ConjugateGradient.h
@@ -154,7 +154,7 @@ class ConjugateGradient : public OperatorFunction<Field> {
                   << LinalgTimer.Elapsed();
         std::cout << std::endl;
 
-        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 1000.0);
+        if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);
 
         return;
       }
diff --git a/lib/qcd/hmc/HmcRunner.h b/lib/qcd/hmc/HmcRunner.h
index a31ba784..53b127cf 100644
--- a/lib/qcd/hmc/HmcRunner.h
+++ b/lib/qcd/hmc/HmcRunner.h
@@ -116,7 +116,7 @@ class NerscHmcRunnerTemplate {
     NoSmearing<Gimpl> SmearingPolicy;
     typedef MinimumNorm2<GaugeField, NoSmearing<Gimpl>, RepresentationsPolicy >
         IntegratorType;  // change here to change the algorithm
-    IntegratorParameters MDpar(20, 1.0);
+    IntegratorParameters MDpar(40, 1.0);
     IntegratorType MDynamics(UGrid, MDpar, TheAction, SmearingPolicy);
 
     // Checkpoint strategy
diff --git a/tests/hmc/Test_hmc_WilsonFermionGauge.cc b/tests/hmc/Test_hmc_WilsonFermionGauge.cc
index 9dcf6343..351d1e68 100644
--- a/tests/hmc/Test_hmc_WilsonFermionGauge.cc
+++ b/tests/hmc/Test_hmc_WilsonFermionGauge.cc
@@ -68,7 +68,7 @@ class HmcRunner : public NerscHmcRunner {
     TwoFlavourPseudoFermionAction<ImplPolicy> Nf2(FermOp, CG, CG);
 
     // Set smearing (true/false), default: false
-    Nf2.is_smeared = true;
+    Nf2.is_smeared = false;
 
     // Collect actions
     ActionLevel<LatticeGaugeField> Level1(1);