Boosted fermion attempt

Bugfix/irl convergence

Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h

@@ -419,14 +419,15 @@ until convergence
 	}
       }
 
-      if ( Nconv < Nstop )
+      if ( Nconv < Nstop ) {
 	std::cout << GridLogIRL << "Nconv ("<<Nconv<<") < Nstop ("<<Nstop<<")"<<std::endl;
-
+	std::cout << GridLogIRL << "returning Nstop vectors, the last "<< Nstop-Nconv << "of which might meet convergence criterion only approximately" <<std::endl;
+      }
       eval=eval2;
       
       //Keep only converged
-      eval.resize(Nconv);// Nstop?
-      evec.resize(Nconv,grid);// Nstop?
+      eval.resize(Nstop);// was Nconv
+      evec.resize(Nstop,grid);// was Nconv
       basisSortInPlace(evec,eval,reverse);
       
     }

Grid/qcd/action/ActionBase.h

@@ -34,10 +34,24 @@ directory
 
 NAMESPACE_BEGIN(Grid);
 
+///////////////////////////////////
+// Smart configuration base class
+///////////////////////////////////
+template< class Field >
+class ConfigurationBase
+{
+public:
+  ConfigurationBase() {}
+  virtual ~ConfigurationBase() {}
+  virtual void set_Field(Field& U) =0;
+  virtual void smeared_force(Field&) = 0;
+  virtual Field& get_SmearedU() =0;
+  virtual Field &get_U(bool smeared = false) = 0;
+};
+
 template <class GaugeField >
 class Action 
 {
-
 public:
   bool is_smeared = false;
   RealD deriv_norm_sum;
@@ -77,11 +91,39 @@ public:
   void refresh_timer_stop(void)  { refresh_us+=usecond(); }
   void S_timer_start(void)       { S_us-=usecond(); }
   void S_timer_stop(void)        { S_us+=usecond(); }
+  /////////////////////////////
   // Heatbath?
+  /////////////////////////////
   virtual void refresh(const GaugeField& U, GridSerialRNG &sRNG, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
   virtual RealD S(const GaugeField& U) = 0;                             // evaluate the action
   virtual RealD Sinitial(const GaugeField& U) { return this->S(U); } ;  // if the refresh computes the action, can cache it. Alternately refreshAndAction() ?
   virtual void deriv(const GaugeField& U, GaugeField& dSdU) = 0;        // evaluate the action derivative
+
+  /////////////////////////////////////////////////////////////
+  // virtual smeared interface through configuration container
+  /////////////////////////////////////////////////////////////
+  virtual void refresh(ConfigurationBase<GaugeField> & U, GridSerialRNG &sRNG, GridParallelRNG& pRNG)
+  {
+    refresh(U.get_U(is_smeared),sRNG,pRNG);
+  }
+  virtual RealD S(ConfigurationBase<GaugeField>& U)
+  {
+    return S(U.get_U(is_smeared));
+  }
+  virtual RealD Sinitial(ConfigurationBase<GaugeField>& U) 
+  {
+    return Sinitial(U.get_U(is_smeared));
+  }
+  virtual void deriv(ConfigurationBase<GaugeField>& U, GaugeField& dSdU)
+  {
+    deriv(U.get_U(is_smeared),dSdU); 
+    if ( is_smeared ) {
+      U.smeared_force(dSdU);
+    }
+  }
+  ///////////////////////////////
+  // Logging
+  ///////////////////////////////
   virtual std::string action_name()    = 0;                             // return the action name
   virtual std::string LogParameters()  = 0;                             // prints action parameters
   virtual ~Action(){}

Grid/qcd/action/ActionCore.h

@@ -30,6 +30,8 @@ directory
 #ifndef QCD_ACTION_CORE
 #define QCD_ACTION_CORE
 
+#include <Grid/qcd/action/gauge/GaugeImplementations.h>
+
 #include <Grid/qcd/action/ActionBase.h>
 NAMESPACE_CHECK(ActionBase);
 #include <Grid/qcd/action/ActionSet.h>

Grid/qcd/action/fermion/CayleyFermion5D.h

@@ -124,6 +124,11 @@ public:
   RealD                _b;
   RealD                _c;
 
+  // possible boost
+  std::vector<ComplexD> qmu;
+  void set_qmu(std::vector<ComplexD> _qmu) { qmu=_qmu; assert(qmu.size()==Nd);};
+  void addQmu(const FermionField &in, FermionField &out, int dag);
+  
   // Cayley form Moebius (tanh and zolotarev)
   Vector<Coeff_t> omega;
   Vector<Coeff_t> bs;    // S dependent coeffs

Grid/qcd/action/fermion/ContinuedFractionFermion5D.h

@@ -60,6 +60,50 @@ public:
   //      virtual void   Instantiatable(void)=0;
   virtual void   Instantiatable(void) =0;
 
+  void FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary, std::vector<double> twist)
+  {
+    std::cout << "Free Propagator for PartialFraction"<<std::endl;
+    FermionField in_k(in.Grid());
+    FermionField prop_k(in.Grid());
+    
+    FFT theFFT((GridCartesian *) in.Grid());
+
+    //phase for boundary condition
+    ComplexField coor(in.Grid());
+    ComplexField ph(in.Grid());  ph = Zero();
+    FermionField in_buf(in.Grid()); in_buf = Zero();
+    typedef typename Simd::scalar_type Scalar;
+    Scalar ci(0.0,1.0);
+    assert(twist.size() == Nd);//check that twist is Nd
+    assert(boundary.size() == Nd);//check that boundary conditions is Nd
+    int shift = 0;
+    for(unsigned int nu = 0; nu < Nd; nu++)
+      {
+	// Shift coordinate lattice index by 1 to account for 5th dimension.
+	LatticeCoordinate(coor, nu + shift);
+	double boundary_phase = ::acos(real(boundary[nu]));
+	ph = ph + boundary_phase*coor*((1./(in.Grid()->_fdimensions[nu+shift])));
+	//momenta for propagator shifted by twist+boundary
+	twist[nu] = twist[nu] + boundary_phase/((2.0*M_PI));
+      }
+    in_buf = exp(ci*ph*(-1.0))*in;
+
+    theFFT.FFT_all_dim(in_k,in,FFT::forward);
+    this->MomentumSpacePropagatorHw(prop_k,in_k,mass,twist);
+    theFFT.FFT_all_dim(out,prop_k,FFT::backward);
+    
+    //phase for boundary condition
+    out = out * exp(ci*ph);
+  };
+
+  virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass) {
+    std::vector<double> twist(Nd,0.0); //default: periodic boundarys in all directions
+    std::vector<Complex> boundary;
+    for(int i=0;i<Nd;i++) boundary.push_back(1);//default: periodic boundary conditions
+    FreePropagator(in,out,mass,boundary,twist);
+  };
+
+  
   // Efficient support for multigrid coarsening
   virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp);
   virtual void  MdirAll(const FermionField &in, std::vector<FermionField> &out);

Grid/qcd/action/fermion/PartialFractionFermion5D.h

@@ -39,7 +39,7 @@ class PartialFractionFermion5D : public WilsonFermion5D<Impl>
 public:
   INHERIT_IMPL_TYPES(Impl);
 
-  const int part_frac_chroma_convention=1;
+  const int part_frac_chroma_convention=0;
 
   void   Meooe_internal(const FermionField &in, FermionField &out,int dag);
   void   Mooee_internal(const FermionField &in, FermionField &out,int dag);
@@ -83,12 +83,63 @@ public:
 			   GridRedBlackCartesian &FourDimRedBlackGrid,
 			   RealD _mass,RealD M5,const ImplParams &p= ImplParams());
 
+  PartialFractionFermion5D(GaugeField &_Umu,
+			   GridCartesian         &FiveDimGrid,
+			   GridRedBlackCartesian &FiveDimRedBlackGrid,
+			   GridCartesian         &FourDimGrid,
+			   GridRedBlackCartesian &FourDimRedBlackGrid,
+			   RealD _mass,RealD M5,std::vector<RealD> &_qmu,const ImplParams &p= ImplParams());
+
+  void FreePropagator(const FermionField &in,FermionField &out,RealD mass,std::vector<Complex> boundary, std::vector<double> twist)
+  {
+    std::cout << "Free Propagator for PartialFraction"<<std::endl;
+    FermionField in_k(in.Grid());
+    FermionField prop_k(in.Grid());
+    
+    FFT theFFT((GridCartesian *) in.Grid());
+
+    //phase for boundary condition
+    ComplexField coor(in.Grid());
+    ComplexField ph(in.Grid());  ph = Zero();
+    FermionField in_buf(in.Grid()); in_buf = Zero();
+    typedef typename Simd::scalar_type Scalar;
+    Scalar ci(0.0,1.0);
+    assert(twist.size() == Nd);//check that twist is Nd
+    assert(boundary.size() == Nd);//check that boundary conditions is Nd
+    int shift = 0;
+    for(unsigned int nu = 0; nu < Nd; nu++)
+      {
+	// Shift coordinate lattice index by 1 to account for 5th dimension.
+	LatticeCoordinate(coor, nu + shift);
+	double boundary_phase = ::acos(real(boundary[nu]));
+	ph = ph + boundary_phase*coor*((1./(in.Grid()->_fdimensions[nu+shift])));
+	//momenta for propagator shifted by twist+boundary
+	twist[nu] = twist[nu] + boundary_phase/((2.0*M_PI));
+      }
+    in_buf = exp(ci*ph*(-1.0))*in;
+
+    theFFT.FFT_all_dim(in_k,in,FFT::forward);
+    this->MomentumSpacePropagatorHw(prop_k,in_k,mass,twist);
+    theFFT.FFT_all_dim(out,prop_k,FFT::backward);
+    
+    //phase for boundary condition
+    out = out * exp(ci*ph);
+  };
+
+  virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass) {
+    std::vector<double> twist(Nd,0.0); //default: periodic boundarys in all directions
+    std::vector<Complex> boundary;
+    for(int i=0;i<Nd;i++) boundary.push_back(1);//default: periodic boundary conditions
+    FreePropagator(in,out,mass,boundary,twist);
+  };
+  
 protected:
 
   virtual void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD scale);
   virtual void SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata);
 
   // Part frac
+  std::vector<RealD> qmu;
   RealD mass;
   RealD dw_diag;
   RealD R;

Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h

@@ -48,7 +48,8 @@ CayleyFermion5D<Impl>::CayleyFermion5D(GaugeField &_Umu,
 			FourDimGrid,
 			FourDimRedBlackGrid,_M5,p),
   mass_plus(_mass), mass_minus(_mass)
-{ 
+{
+  // qmu defaults to zero size;
 }
 
 ///////////////////////////////////////////////////////////////
@@ -270,6 +271,34 @@ void CayleyFermion5D<Impl>::MeooeDag5D    (const FermionField &psi, FermionField
   M5Ddag(psi,psi,Din,lower,diag,upper);
 }
 
+template<class Impl>
+void CayleyFermion5D<Impl>::addQmu(const FermionField &psi,FermionField &chi, int dag)
+{
+  if ( qmu.size() ) {
+
+    Gamma::Algebra Gmu [] = {
+      Gamma::Algebra::GammaX,
+      Gamma::Algebra::GammaY,
+      Gamma::Algebra::GammaZ,
+      Gamma::Algebra::GammaT
+    };
+    std::vector<ComplexD> coeff(Nd);
+    ComplexD ci(0,1);
+
+    assert(qmu.size()==Nd);
+
+    for(int mu=0;mu<Nd;mu++){
+       coeff[mu] = ci*qmu[mu];
+       if ( dag ) coeff[mu] = conjugate(coeff[mu]);
+    }
+
+    chi = chi + Gamma(Gmu[0])*psi*coeff[0];
+    for(int mu=1;mu<Nd;mu++){
+      chi = chi + Gamma(Gmu[mu])*psi*coeff[mu];
+    }
+  }
+}
+
 template<class Impl>
 void CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
 {
@@ -277,8 +306,12 @@ void CayleyFermion5D<Impl>::M    (const FermionField &psi, FermionField &chi)
   
   // Assemble Din
   Meooe5D(psi,Din);
-  
+
   this->DW(Din,chi,DaggerNo);
+
+  // add i q_mu gamma_mu here
+  addQmu(Din,chi,DaggerNo);
+  
   // ((b D_W + D_w hop terms +1) on s-diag
   axpby(chi,1.0,1.0,chi,psi); 
   
@@ -295,6 +328,9 @@ void CayleyFermion5D<Impl>::Mdag (const FermionField &psi, FermionField &chi)
   FermionField Din(psi.Grid());
   // Apply Dw
   this->DW(psi,Din,DaggerYes); 
+
+  // add -i conj(q_mu) gamma_mu here ... if qmu is real, gammm_5 hermitian, otherwise not.
+  addQmu(psi,Din,DaggerYes);
   
   MeooeDag5D(Din,chi);
   

Grid/qcd/action/fermion/implementation/ContinuedFractionFermion5DImplementation.h

@@ -42,13 +42,13 @@ template<class Impl>
 void ContinuedFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata)
 {
   // How to check Ls matches??
-  //      std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
-  //      std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
+  std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
+  std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
+  std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
+  std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
+  std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
   int Ls = this->Ls;
+  std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
   assert(zdata->db==Ls);// Beta has Ls coeffs
 
   R=(1+this->mass)/(1-this->mass);
@@ -320,7 +320,7 @@ ContinuedFractionFermion5D<Impl>::ContinuedFractionFermion5D(
       int Ls = this->Ls;
       conformable(solution5d.Grid(),this->FermionGrid());
       conformable(exported4d.Grid(),this->GaugeGrid());
-      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
+      ExtractSlice(exported4d, solution5d, Ls-1, 0);
     }
     template<class Impl>
     void ContinuedFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
@@ -330,7 +330,7 @@ ContinuedFractionFermion5D<Impl>::ContinuedFractionFermion5D(
       conformable(input4d.Grid()   ,this->GaugeGrid());
       FermionField tmp(this->FermionGrid());
       tmp=Zero();
-      InsertSlice(input4d, tmp, Ls-1, Ls-1);
+      InsertSlice(input4d, tmp, Ls-1, 0);
       tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
       this->Dminus(tmp,imported5d);
     }

Grid/qcd/action/fermion/implementation/PartialFractionFermion5DImplementation.h

@@ -255,15 +255,76 @@ void   PartialFractionFermion5D<Impl>::M_internal(const FermionField &psi, Fermi
   }
 	
   {
+    // The 'conventional' Cayley overlap operator is
+    //
+    // Dov = (1+m)/2 + (1-m)/2 g5 sgn Hw
+    //
+    //
+    // With massless limit 1/2(1+g5 sgnHw)
+    //
+    // Luscher shows quite neatly that 1+g5 sgn Hw has tree level propagator i qslash +O(a^2)
+    //
+    // However, the conventional normalisation has both a leading order factor of 2 in Zq
+    // at tree level AND a mass dependent (1-m) that are convenient to absorb.
+    //
+    // In WilsonFermion5DImplementation.h, the tree level propagator for Hw is
+    //
+    // num = -i sin kmu gmu
+    //
+    // denom ( sqrt(sk^2 + (2shk^2 - 1)^2
+    //    b_k = sk2 - M5;
+    //     
+    //    w_k = sqrt(sk + b_k*b_k);
+    //
+    //    denom= ( w_k + b_k + mass*mass) ;
+    //
+    //    denom= one/denom;
+    //    out = num*denom;
+    //
+    // Chroma, and Grid define partial fraction via 4d operator
+    //
+    //   Dpf = 2/(1-m) x Dov = (1+m)/(1-m) + g5 sgn Hw
+    //
+    // Now since:
+    //
+    //      (1+m)/(1-m) = (1-m)/(1-m) + 2m/(1-m) = 1 + 2m/(1-m)
+    //
+    // This corresponds to a modified mass parameter
+    //
+    // It has an annoying 
+    //
+    // 
     double R=(1+this->mass)/(1-this->mass);
     //R g5 psi[Ls] + p[0] H
     ag5xpbg5y_ssp(chi,R*scale,psi,p[nblock]*scale/amax,D,Ls-1,Ls-1);
-	
+    
     for(int b=0;b<nblock;b++){
       int s = 2*b+1;
       double pp = p[nblock-1-b];
       axpby_ssp(chi,1.0,chi,-sqrt(amax*pp)*scale*sign,psi,Ls-1,s);
     }
+
+    if ( qmu.size() ) {
+
+      FermionField qslash_psi(psi.Grid());
+      
+      Gamma::Algebra Gmu [] = {
+			 Gamma::Algebra::GammaX,
+			 Gamma::Algebra::GammaY,
+			 Gamma::Algebra::GammaZ,
+			 Gamma::Algebra::GammaT
+      };
+      ComplexD ci(0,1);
+      assert(qmu.size()==Nd);
+      qslash_psi = Gamma(Gmu[0])*psi;
+      for(int mu=1;mu<Nd;mu++){
+	qslash_psi = Gamma(Gmu[mu])*psi;
+      }
+      //      RealD coeff = 1.0;
+      qslash_psi = Gamma(Gamma::Algebra::Gamma5)*qslash_psi*ci ; // i g5 qslash -- 1-m factor???
+      axpby_ssp(chi,1.0,chi,1.0, qslash_psi,Ls-1,Ls-1);
+    }
+    
   }
 
 }
@@ -411,7 +472,7 @@ void  PartialFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,App
       int Ls = this->Ls;
       conformable(solution5d.Grid(),this->FermionGrid());
       conformable(exported4d.Grid(),this->GaugeGrid());
-      ExtractSlice(exported4d, solution5d, Ls-1, Ls-1);
+      ExtractSlice(exported4d, solution5d, Ls-1, 0);
     }
     template<class Impl>
     void PartialFractionFermion5D<Impl>::ImportPhysicalFermionSource(const FermionField &input4d,FermionField &imported5d)
@@ -421,7 +482,8 @@ void  PartialFractionFermion5D<Impl>::SetCoefficientsZolotarev(RealD zolo_hi,App
       conformable(input4d.Grid()   ,this->GaugeGrid());
       FermionField tmp(this->FermionGrid());
       tmp=Zero();
-      InsertSlice(input4d, tmp, Ls-1, Ls-1);
+      std::cout << " importing to slice " << Ls-1 <<std::endl;
+      InsertSlice(input4d, tmp, Ls-1, 0);
       tmp=Gamma(Gamma::Algebra::Gamma5)*tmp;
       this->Dminus(tmp,imported5d);
     }
@@ -442,7 +504,7 @@ PartialFractionFermion5D<Impl>::PartialFractionFermion5D(GaugeField &_Umu,
 
 {
   int Ls = this->Ls;
-
+  qmu.resize(0);
   assert((Ls&0x1)==1); // Odd Ls required
   int nrational=Ls-1;
 
@@ -460,6 +522,22 @@ PartialFractionFermion5D<Impl>::PartialFractionFermion5D(GaugeField &_Umu,
   Approx::zolotarev_free(zdata);
 
 }
+template<class Impl>
+PartialFractionFermion5D<Impl>::PartialFractionFermion5D(GaugeField &_Umu,
+							 GridCartesian         &FiveDimGrid,
+							 GridRedBlackCartesian &FiveDimRedBlackGrid,
+							 GridCartesian         &FourDimGrid,
+							 GridRedBlackCartesian &FourDimRedBlackGrid,
+							 RealD _mass,RealD M5,
+							 std::vector<RealD> &_qmu,
+							 const ImplParams &p)
+  : PartialFractionFermion5D<Impl>(_Umu,
+			     FiveDimGrid,FiveDimRedBlackGrid,
+			     FourDimGrid,FourDimRedBlackGrid,
+			     _mass,M5,p)
+{
+  qmu=_qmu;
+}
 
 NAMESPACE_END(Grid);
 

Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h

@@ -423,7 +423,6 @@ void WilsonKernels<Impl>::DhopDirKernel( StencilImpl &st, DoubledGaugeField &U,S
 #define KERNEL_CALL(A) KERNEL_CALLNB(A); accelerator_barrier();
 
 #define KERNEL_CALL_EXT(A)						\
-  const uint64_t    NN = Nsite*Ls;					\
   const uint64_t    sz = st.surface_list.size();			\
   auto ptr = &st.surface_list[0];					\
   accelerator_forNB( ss, sz, Simd::Nsimd(), {				\

Grid/qcd/hmc/HMC.h

@@ -283,12 +283,13 @@ public:
       std::cout << GridLogHMC << "Total time for trajectory (s): " << (t1-t0)/1e6 << std::endl;
 
       TheIntegrator.print_timer();
-
+      
+      TheIntegrator.Smearer.set_Field(Ucur);
       for (int obs = 0; obs < Observables.size(); obs++) {
       	std::cout << GridLogDebug << "Observables # " << obs << std::endl;
       	std::cout << GridLogDebug << "Observables total " << Observables.size() << std::endl;
       	std::cout << GridLogDebug << "Observables pointer " << Observables[obs] << std::endl;
-        Observables[obs]->TrajectoryComplete(traj + 1, Ucur, sRNG, pRNG);
+        Observables[obs]->TrajectoryComplete(traj + 1, TheIntegrator.Smearer, sRNG, pRNG);
       }
       std::cout << GridLogHMC << ":::::::::::::::::::::::::::::::::::::::::::" << std::endl;
     }

Grid/qcd/hmc/checkpointers/BaseCheckpointer.h

@@ -35,13 +35,16 @@ class CheckpointerParameters : Serializable {
 public:
   GRID_SERIALIZABLE_CLASS_MEMBERS(CheckpointerParameters, 
 				  std::string, config_prefix, 
+				  std::string, smeared_prefix, 
 				  std::string, rng_prefix, 
 				  int, saveInterval, 
+				  bool, saveSmeared, 
 				  std::string, format, );
 
-  CheckpointerParameters(std::string cf = "cfg", std::string rn = "rng",
+  CheckpointerParameters(std::string cf = "cfg", std::string sf="cfg_smr" , std::string rn = "rng",
 			 int savemodulo = 1, const std::string &f = "IEEE64BIG")
     : config_prefix(cf),
+      smeared_prefix(sf),
       rng_prefix(rn),
       saveInterval(savemodulo),
       format(f){};
@@ -61,13 +64,21 @@ template <class Impl>
 class BaseHmcCheckpointer : public HmcObservable<typename Impl::Field> {
 public:
   void build_filenames(int traj, CheckpointerParameters &Params,
-                       std::string &conf_file, std::string &rng_file) {
+                       std::string &conf_file,
+                       std::string &smear_file,
+		       std::string &rng_file) {
     {
       std::ostringstream os;
       os << Params.rng_prefix << "." << traj;
       rng_file = os.str();
     }
 
+    {
+      std::ostringstream os;
+      os << Params.smeared_prefix << "." << traj;
+      smear_file = os.str();
+    }
+
     {
       std::ostringstream os;
       os << Params.config_prefix << "." << traj;
@@ -84,6 +95,11 @@ public:
   }
   virtual void initialize(const CheckpointerParameters &Params) = 0;
 
+  virtual void TrajectoryComplete(int traj,
+                                  typename Impl::Field &U,
+                                  GridSerialRNG &sRNG,
+                                  GridParallelRNG &pRNG) { assert(0); } ; // HMC should pass the smart config with smeared and unsmeared
+  
   virtual void CheckpointRestore(int traj, typename Impl::Field &U,
                                  GridSerialRNG &sRNG,
                                  GridParallelRNG &pRNG) = 0;

Grid/qcd/hmc/checkpointers/BinaryCheckpointer.h

@@ -61,11 +61,14 @@ public:
     fout.close();
   }
 
-  void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
+  void TrajectoryComplete(int traj,
+			  ConfigurationBase<Field> &SmartConfig,
+			  GridSerialRNG &sRNG, GridParallelRNG &pRNG)
+  {
 
     if ((traj % Params.saveInterval) == 0) {
-      std::string config, rng;
-      this->build_filenames(traj, Params, config, rng);
+      std::string config, rng, smr;
+      this->build_filenames(traj, Params, config, smr, rng);
 
       uint32_t nersc_csum;
       uint32_t scidac_csuma;
@@ -74,9 +77,15 @@ public:
       BinarySimpleUnmunger<sobj_double, sobj> munge;
       truncate(rng);
       BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
-      truncate(config);
+      std::cout << GridLogMessage << "Written Binary RNG " << rng
+                << " checksum " << std::hex 
+		<< nersc_csum   <<"/"
+		<< scidac_csuma   <<"/"
+		<< scidac_csumb 
+		<< std::dec << std::endl;
 
-      BinaryIO::writeLatticeObject<vobj, sobj_double>(U, config, munge, 0, Params.format,
+      truncate(config);
+      BinaryIO::writeLatticeObject<vobj, sobj_double>(SmartConfig.get_U(false), config, munge, 0, Params.format,
 						      nersc_csum,scidac_csuma,scidac_csumb);
 
       std::cout << GridLogMessage << "Written Binary Configuration " << config
@@ -85,6 +94,18 @@ public:
 		<< scidac_csuma   <<"/"
 		<< scidac_csumb 
 		<< std::dec << std::endl;
+
+      if ( Params.saveSmeared ) {
+	truncate(smr);
+	BinaryIO::writeLatticeObject<vobj, sobj_double>(SmartConfig.get_U(true), smr, munge, 0, Params.format,
+							nersc_csum,scidac_csuma,scidac_csumb);
+	std::cout << GridLogMessage << "Written Binary Smeared Configuration " << smr
+                << " checksum " << std::hex 
+		<< nersc_csum   <<"/"
+		<< scidac_csuma   <<"/"
+		<< scidac_csumb 
+		<< std::dec << std::endl;
+      }
     }
 
   };

Grid/qcd/hmc/checkpointers/ILDGCheckpointer.h

@@ -69,17 +69,27 @@ public:
     }
   }
 
-  void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
+  void TrajectoryComplete(int traj,
+			  ConfigurationBase<GaugeField> &SmartConfig,
+			  GridSerialRNG &sRNG,
                           GridParallelRNG &pRNG) {
     if ((traj % Params.saveInterval) == 0) {
-      std::string config, rng;
+      std::string config, rng, smr;
       this->build_filenames(traj, Params, config, rng);
-      GridBase *grid = U.Grid();
+      GridBase *grid = SmartConfig.get_U(false).Grid();
       uint32_t nersc_csum,scidac_csuma,scidac_csumb;
       BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
+      std::cout << GridLogMessage << "Written BINARY RNG " << rng
+                << " checksum " << std::hex 
+		<< nersc_csum<<"/"
+		<< scidac_csuma<<"/"
+		<< scidac_csumb
+		<< std::dec << std::endl;
+
+      
       IldgWriter _IldgWriter(grid->IsBoss());
       _IldgWriter.open(config);
-      _IldgWriter.writeConfiguration<GaugeStats>(U, traj, config, config);
+      _IldgWriter.writeConfiguration<GaugeStats>(SmartConfig.get_U(false), traj, config, config);
       _IldgWriter.close();
 
       std::cout << GridLogMessage << "Written ILDG Configuration on " << config
@@ -88,6 +98,21 @@ public:
 		<< scidac_csuma<<"/"
 		<< scidac_csumb
 		<< std::dec << std::endl;
+
+      if ( Params.saveSmeared ) { 
+	IldgWriter _IldgWriter(grid->IsBoss());
+	_IldgWriter.open(smr);
+	_IldgWriter.writeConfiguration<GaugeStats>(SmartConfig.get_U(true), traj, config, config);
+	_IldgWriter.close();
+
+	std::cout << GridLogMessage << "Written ILDG Configuration on " << smr
+                << " checksum " << std::hex 
+		<< nersc_csum<<"/"
+		<< scidac_csuma<<"/"
+		<< scidac_csumb
+		<< std::dec << std::endl;
+      }
+
     }
   };
 

Grid/qcd/hmc/checkpointers/NerscCheckpointer.h

@@ -52,23 +52,29 @@ public:
     Params.format = "IEEE64BIG";  // fixed, overwrite any other choice
   }
 
-  void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
-                          GridParallelRNG &pRNG) {
+  virtual void TrajectoryComplete(int traj,
+                                  ConfigurationBase<GaugeField> &SmartConfig,
+                                  GridSerialRNG &sRNG,
+                                  GridParallelRNG &pRNG)
+  {
     if ((traj % Params.saveInterval) == 0) {
-      std::string config, rng;
-      this->build_filenames(traj, Params, config, rng);
-
+      std::string config, rng, smr;
+      this->build_filenames(traj, Params, config, smr, rng);
+      
       int precision32 = 1;
       int tworow = 0;
       NerscIO::writeRNGState(sRNG, pRNG, rng);
-      NerscIO::writeConfiguration<GaugeStats>(U, config, tworow, precision32);
+      NerscIO::writeConfiguration<GaugeStats>(SmartConfig.get_U(false), config, tworow, precision32);
+      if ( Params.saveSmeared ) {
+	NerscIO::writeConfiguration<GaugeStats>(SmartConfig.get_U(true), smr, tworow, precision32);
+      }
     }
   };
 
   void CheckpointRestore(int traj, GaugeField &U, GridSerialRNG &sRNG,
                          GridParallelRNG &pRNG) {
-    std::string config, rng;
-    this->build_filenames(traj, Params, config, rng);
+    std::string config, rng, smr;
+    this->build_filenames(traj, Params, config, smr, rng );
     this->check_filename(rng);
     this->check_filename(config);
 

Grid/qcd/hmc/checkpointers/ScidacCheckpointer.h

@@ -70,19 +70,37 @@ class ScidacHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
     }
   }
 
-  void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG,
+  void TrajectoryComplete(int traj, 
+			  ConfigurationBase<Field> &SmartConfig,
+			  GridSerialRNG &sRNG,
                           GridParallelRNG &pRNG) {
     if ((traj % Params.saveInterval) == 0) {
-      std::string config, rng;
-      this->build_filenames(traj, Params, config, rng);
-      GridBase *grid = U.Grid();
+      std::string config, rng,smr;
+      this->build_filenames(traj, Params, config, smr, rng);
+      GridBase *grid = SmartConfig.get_U(false).Grid();
       uint32_t nersc_csum,scidac_csuma,scidac_csumb;
       BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
-      ScidacWriter _ScidacWriter(grid->IsBoss());
-      _ScidacWriter.open(config);
-      _ScidacWriter.writeScidacFieldRecord(U, MData);
-      _ScidacWriter.close();
+      std::cout << GridLogMessage << "Written Binary RNG " << rng
+                << " checksum " << std::hex 
+		<< nersc_csum   <<"/"
+		<< scidac_csuma   <<"/"
+		<< scidac_csumb 
+		<< std::dec << std::endl;
 
+
+      {
+	ScidacWriter _ScidacWriter(grid->IsBoss());
+	_ScidacWriter.open(config);
+	_ScidacWriter.writeScidacFieldRecord(SmartConfig.get_U(false), MData);
+	_ScidacWriter.close();
+      }
+      
+      if ( Params.saveSmeared ) {
+	ScidacWriter _ScidacWriter(grid->IsBoss());
+	_ScidacWriter.open(smr);
+	_ScidacWriter.writeScidacFieldRecord(SmartConfig.get_U(true), MData);
+	_ScidacWriter.close();
+      }
       std::cout << GridLogMessage << "Written Scidac Configuration on " << config << std::endl;
     }
   };

Grid/qcd/hmc/integrators/Integrator.h

@@ -66,6 +66,7 @@ public:
 template <class FieldImplementation_, class SmearingPolicy, class RepresentationPolicy>
 class Integrator {
 protected:
+public:
   typedef FieldImplementation_ FieldImplementation;
   typedef typename FieldImplementation::Field MomentaField;  //for readability
   typedef typename FieldImplementation::Field Field;
@@ -96,7 +97,6 @@ protected:
   {
     t_P[level] += ep;
     update_P(P, U, level, ep);
-
     std::cout << GridLogIntegrator << "[" << level << "] P " << " dt " << ep << " : t_P " << t_P[level] << std::endl;
   }
 
@@ -130,28 +130,20 @@ protected:
       Field force(U.Grid());
       conformable(U.Grid(), Mom.Grid());
 
-      Field& Us = Smearer.get_U(as[level].actions.at(a)->is_smeared);
       double start_force = usecond();
 
-      std::cout << GridLogMessage << "AuditForce["<<level<<"]["<<a<<"] before"<<std::endl;
-      
       as[level].actions.at(a)->deriv_timer_start();
-      as[level].actions.at(a)->deriv(Us, force);  // deriv should NOT include Ta
+      as[level].actions.at(a)->deriv(Smearer, force);  // deriv should NOT include Ta
       as[level].actions.at(a)->deriv_timer_stop();
 
-      std::cout << GridLogMessage << "AuditForce["<<level<<"]["<<a<<"] after"<<std::endl;
-
-      std::cout << GridLogIntegrator << "Smearing (on/off): " << as[level].actions.at(a)->is_smeared << std::endl;
       auto name = as[level].actions.at(a)->action_name();
-      if (as[level].actions.at(a)->is_smeared) Smearer.smeared_force(force);
 
       force = FieldImplementation::projectForce(force); // Ta for gauge fields
       double end_force = usecond();
-
-      //      DumpSliceNorm("force ",force,Nd-1);
+      
       MomFilter->applyFilter(force);
+
       std::cout << GridLogIntegrator << " update_P : Level [" << level <<"]["<<a <<"] "<<name<<" dt "<<ep<<  std::endl;
-      DumpSliceNorm("force filtered ",force,Nd-1);
       
       Real force_abs   = std::sqrt(norm2(force)/U.Grid()->gSites()); //average per-site norm.  nb. norm2(latt) = \sum_x norm2(latt[x]) 
       Real impulse_abs = force_abs * ep * HMC_MOMENTUM_DENOMINATOR;    
@@ -377,14 +369,9 @@ public:
 	auto name = as[level].actions.at(actionID)->action_name();
         std::cout << GridLogMessage << "refresh [" << level << "][" << actionID << "] "<<name << std::endl;
 
-        Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
-
-	std::cout << GridLogMessage << "AuditRefresh["<<level<<"]["<<actionID<<"] before"<<std::endl;
-
 	as[level].actions.at(actionID)->refresh_timer_start();
-        as[level].actions.at(actionID)->refresh(Us, sRNG, pRNG);
+        as[level].actions.at(actionID)->refresh(Smearer, sRNG, pRNG);
 	as[level].actions.at(actionID)->refresh_timer_stop();
-	std::cout << GridLogMessage << "AuditRefresh["<<level<<"]["<<actionID<<"] after"<<std::endl;
 
       }
 
@@ -425,10 +412,9 @@ public:
 
         // get gauge field from the SmearingPolicy and
         // based on the boolean is_smeared in actionID
-        Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
         std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] action eval " << std::endl;
 	        as[level].actions.at(actionID)->S_timer_start();
-        Hterm = as[level].actions.at(actionID)->S(Us);
+        Hterm = as[level].actions.at(actionID)->S(Smearer);
    	        as[level].actions.at(actionID)->S_timer_stop();
         std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] H = " << Hterm << std::endl;
         H += Hterm;
@@ -469,12 +455,11 @@ public:
       for (int actionID = 0; actionID < as[level].actions.size(); ++actionID) {
         // get gauge field from the SmearingPolicy and
         // based on the boolean is_smeared in actionID
-        Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
         std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] action eval " << std::endl;
-	        as[level].actions.at(actionID)->S_timer_start();
 
-        Hterm = as[level].actions.at(actionID)->Sinitial(Us);
-   	        as[level].actions.at(actionID)->S_timer_stop();
+	as[level].actions.at(actionID)->S_timer_start();
+        Hterm = as[level].actions.at(actionID)->S(Smearer);
+	as[level].actions.at(actionID)->S_timer_stop();
 
         std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] H = " << Hterm << std::endl;
         H += Hterm;

Grid/qcd/observables/hmc_observable.h

@@ -34,6 +34,13 @@ NAMESPACE_BEGIN(Grid);
 template <class Field>
 class HmcObservable {
  public:
+  virtual void TrajectoryComplete(int traj,
+                                  ConfigurationBase<Field> &SmartConfig,
+                                  GridSerialRNG &sRNG,
+                                  GridParallelRNG &pRNG)
+  {
+    TrajectoryComplete(traj,SmartConfig.get_U(false),sRNG,pRNG); // Unsmeared observable
+  };
   virtual void TrajectoryComplete(int traj,
                                   Field &U,
                                   GridSerialRNG &sRNG,

Grid/qcd/observables/plaquette.h

@@ -42,6 +42,18 @@ public:
   // necessary for HmcObservable compatibility
   typedef typename Impl::Field Field;
 
+  virtual void TrajectoryComplete(int traj,
+                                  ConfigurationBase<Field> &SmartConfig,
+                                  GridSerialRNG &sRNG,
+                                  GridParallelRNG &pRNG)
+  {
+    std::cout << GridLogMessage << "+++++++++++++++++++"<<std::endl;
+    std::cout << GridLogMessage << "Unsmeared plaquette"<<std::endl;
+    TrajectoryComplete(traj,SmartConfig.get_U(false),sRNG,pRNG); // Unsmeared observable
+    std::cout << GridLogMessage << "Smeared plaquette"<<std::endl;
+    TrajectoryComplete(traj,SmartConfig.get_U(true),sRNG,pRNG); // Unsmeared observable
+    std::cout << GridLogMessage << "+++++++++++++++++++"<<std::endl;
+  };
   void TrajectoryComplete(int traj,
                           Field &U,
                           GridSerialRNG &sRNG,

Grid/qcd/smearing/GaugeConfiguration.h

@@ -7,23 +7,10 @@
 
 NAMESPACE_BEGIN(Grid);
 
-template< class Impl >
-class ConfigurationBase
-{
-public:
-  INHERIT_FIELD_TYPES(Impl);
-
-  ConfigurationBase() {}
-  virtual ~ConfigurationBase() {}
-  virtual void set_Field(Field& U) =0;
-  virtual void smeared_force(Field&) const = 0;
-  virtual Field& get_SmearedU() =0;
-  virtual Field &get_U(bool smeared = false) = 0;
-};
 
 //trivial class for no smearing
 template< class Impl >
-class NoSmearing : public ConfigurationBase<Impl>
+class NoSmearing : public ConfigurationBase<typename Impl::Field>
 {
 public:
   INHERIT_FIELD_TYPES(Impl);
@@ -32,13 +19,13 @@ public:
 
   NoSmearing(): ThinLinks(NULL) {}
 
-  void set_Field(Field& U) { ThinLinks = &U; }
+  virtual void set_Field(Field& U) { ThinLinks = &U; }
 
-  void smeared_force(Field&) const {}
+  virtual void smeared_force(Field&) {}
 
-  Field& get_SmearedU() { return *ThinLinks; }
+  virtual Field& get_SmearedU() { return *ThinLinks; }
 
-  Field &get_U(bool smeared = false)
+  virtual Field &get_U(bool smeared = false)
   {
     return *ThinLinks;
   }
@@ -56,7 +43,7 @@ public:
   It stores a list of smeared configurations.
 */
 template <class Gimpl>
-class SmearedConfiguration : public ConfigurationBase<Gimpl>
+class SmearedConfiguration : public ConfigurationBase<typename Gimpl::Field>
 {
 public:
   INHERIT_GIMPL_TYPES(Gimpl);
@@ -248,7 +235,7 @@ public:
     : smearingLevels(0), StoutSmearing(nullptr), SmearedSet(), ThinLinks(NULL) {}
 
   // attach the smeared routines to the thin links U and fill the smeared set
-  void set_Field(GaugeField &U)
+  virtual void set_Field(GaugeField &U)
   {
     double start = usecond();
     fill_smearedSet(U);
@@ -258,7 +245,7 @@ public:
   }
 
   //====================================================================
-  void smeared_force(GaugeField &SigmaTilde) const
+  virtual void smeared_force(GaugeField &SigmaTilde) 
   {
     if (smearingLevels > 0)
     {
@@ -285,14 +272,16 @@ public:
       }
       double end = usecond();
       double time = (end - start)/ 1e3;
-      std::cout << GridLogMessage << "Smearing force in " << time << " ms" << std::endl;  
+      std::cout << GridLogMessage << " GaugeConfiguration: Smeared Force chain rule took " << time << " ms" << std::endl;
     }  // if smearingLevels = 0 do nothing
+    SigmaTilde=Gimpl::projectForce(SigmaTilde); // Ta
+      
   }
   //====================================================================
 
-  GaugeField& get_SmearedU() { return SmearedSet[smearingLevels - 1]; }
+  virtual GaugeField& get_SmearedU() { return SmearedSet[smearingLevels - 1]; }
 
-  GaugeField &get_U(bool smeared = false)
+  virtual GaugeField &get_U(bool smeared = false)
   {
     // get the config, thin links by default
     if (smeared)

Grid/qcd/smearing/GaugeConfigurationMasked.h

@@ -131,6 +131,7 @@ public:
     AdjMatrixField  X(grid);
     Complex ci(0,1);
 
+    RealD t0 = usecond();
     Ident = ComplexD(1.0);
     for(int d=0;d<Nd;d++){
       Umu[d] = peekLorentz(U, d);
@@ -161,6 +162,8 @@ public:
     // Assemble the N matrix
     //////////////////////////////////////////////////////////////////
     // Computes ALL the staples -- could compute one only and do it here
+    RealD time;
+    time=-usecond();
     this->StoutSmearing->BaseSmear(C, U);
     Cmu = peekLorentz(C, mu);
 
@@ -169,7 +172,10 @@ public:
     //////////////////////////////////////////////////////////////////
     // Ta so Z lives in Lie algabra
     Zx  = Ta(Cmu * adj(Umu[mu]));
+    time+=usecond();
+    std::cout << GridLogMessage << "Z took "<<time<< " us"<<std::endl;
 
+    time=-usecond();
     // Move Z to the Adjoint Rep == make_adjoint_representation
     ZxAd = Zero();
     for(int b=0;b<8;b++) {
@@ -180,10 +186,13 @@ public:
       cplx = 2.0*trace(ci*tb*Zx); // my convention 1/2 delta ba
       ZxAd = ZxAd + cplx * TRb; // is this right? YES - Guido used Anti herm Ta's and with bloody wrong sign.
     }
+    time+=usecond();
+    std::cout << GridLogMessage << "ZxAd took "<<time<< " us"<<std::endl;
 
     //////////////////////////////////////
     // J(x) = 1 + Sum_k=1..N (-Zac)^k/(k+1)!
     //////////////////////////////////////
+    time=-usecond();
     X=1.0; 
     JxAd = X;
     mZxAd = (-1.0)*ZxAd; 
@@ -193,10 +202,13 @@ public:
       kpfac = kpfac /(k+1);
       JxAd = JxAd + X * kpfac;
     }
+    time+=usecond();
+    std::cout << GridLogMessage << "Jx took "<<time<< " us"<<std::endl;
 
     //////////////////////////////////////
     // dJ(x)/dxe
     //////////////////////////////////////
+    time=-usecond();
     std::vector<AdjMatrixField>  dJdX;    dJdX.resize(8,grid);
     AdjMatrixField tbXn(grid);
     AdjMatrixField sumXtbX(grid);
@@ -220,12 +232,17 @@ public:
       }
       dJdX[b] = -dt2; 
     }
+    time+=usecond();
+    std::cout << GridLogMessage << "dJx took "<<time<< " us"<<std::endl;
     /////////////////////////////////////////////////////////////////
     // Mask Umu for this link
     /////////////////////////////////////////////////////////////////
+    time=-usecond();
     PlaqL = Ident;
     PlaqR = Utmp*adj(Cmu);
     ComputeNxy(PlaqL,PlaqR,NxxAd);
+    time+=usecond();
+    std::cout << GridLogMessage << "ComputeNxy took "<<time<< " us"<<std::endl;
     
     ////////////////////////////
     // Mab
@@ -236,8 +253,12 @@ public:
     /////////////////////////
     // invert the 8x8
     /////////////////////////
+    time=-usecond();
     MpAdInv = Inverse(MpAd);
+    time+=usecond();
+    std::cout << GridLogMessage << "MpAdInv took "<<time<< " us"<<std::endl;
     
+    RealD t3a = usecond();
     /////////////////////////////////////////////////////////////////
     // Nxx Mp^-1
     /////////////////////////////////////////////////////////////////
@@ -283,6 +304,7 @@ public:
     GaugeField Fdet2(grid);
     GaugeLinkField Fdet_pol(grid); // one polarisation
 
+    RealD t4 = usecond();
     for(int nu=0;nu<Nd;nu++){
 
       if (nu!=mu) {
@@ -291,20 +313,29 @@ public:
 	//    |  |
 	//    x==    // nu polarisation -- clockwise
 
+	time=-usecond();
 	PlaqL=Ident;
 
 	PlaqR=(-rho)*Gimpl::CovShiftForward(Umu[nu], nu,
  	       Gimpl::CovShiftForward(Umu[mu], mu,
 	         Gimpl::CovShiftBackward(Umu[nu], nu,
 		   Gimpl::CovShiftIdentityBackward(Utmp, mu))));
+	time+=usecond();
+	std::cout << GridLogMessage << "PlaqLR took "<<time<< " us"<<std::endl;
 
+	time=-usecond();
 	dJdXe_nMpInv_y =   dJdXe_nMpInv;
 	ComputeNxy(PlaqL,PlaqR,Nxy);
 	Fdet1_nu = transpose(Nxy)*dJdXe_nMpInv_y;
+	time+=usecond();
+	std::cout << GridLogMessage << "ComputeNxy (occurs 6x) took "<<time<< " us"<<std::endl;
 
+	time=-usecond();
 	PlaqR=(-1.0)*PlaqR;
 	Compute_MpInvJx_dNxxdSy(PlaqL,PlaqR,MpInvJx,FdetV);
 	Fdet2_nu = FdetV;
+	time+=usecond();
+	std::cout << GridLogMessage << "Compute_MpInvJx_dNxxSy (occurs 6x) took "<<time<< " us"<<std::endl;
 	
 	//    x==
 	//    |  |
@@ -416,14 +447,21 @@ public:
 	
       }
     }
+    RealD t5 = usecond();
 
     Fdet1_mu = Fdet1_mu + transpose(NxxAd)*dJdXe_nMpInv;
 
     InsertForce(Fdet1,Fdet1_mu,mu);
     InsertForce(Fdet2,Fdet2_mu,mu);
 
-    force = Fdet1 + Fdet2;
-    
+    force= (-0.5)*( Fdet1 + Fdet2);
+    RealD t1 = usecond();
+    std::cout << GridLogMessage << " logDetJacobianForce level took "<<t1-t0<<" us "<<std::endl;
+    std::cout << GridLogMessage << " logDetJacobianForce t3-t0 "<<t3a-t0<<" us "<<std::endl;
+    std::cout << GridLogMessage << " logDetJacobianForce t4-t3 dJdXe_nMpInv "<<t4-t3a<<" us "<<std::endl;
+    std::cout << GridLogMessage << " logDetJacobianForce t5-t4 mu nu loop "<<t5-t4<<" us "<<std::endl;
+    std::cout << GridLogMessage << " logDetJacobianForce t1-t5 "<<t1-t5<<" us "<<std::endl;
+    std::cout << GridLogMessage << " logDetJacobianForce level took "<<t1-t0<<" us "<<std::endl;
   }
   RealD logDetJacobianLevel(const GaugeField &U,int smr)
   {
@@ -534,27 +572,82 @@ public:
     RealD ln_det = 0;
     if (this->smearingLevels > 0)
     {
+      double start = usecond();
       for (int ismr = this->smearingLevels - 1; ismr > 0; --ismr) {
 	ln_det+= logDetJacobianLevel(this->get_smeared_conf(ismr-1),ismr);
       }
       ln_det +=logDetJacobianLevel(*(this->ThinLinks),0);
+
+      double end = usecond();
+      double time = (end - start)/ 1e3;
+      std::cout << GridLogMessage << "GaugeConfigurationMasked: logDetJacobian took " << time << " ms" << std::endl;  
     }
     return ln_det;
   }
   void logDetJacobianForce(GaugeField &force)
   {
-    RealD ln_det = 0;
+    force =Zero();
+    GaugeField force_det(force.Grid());
+
     if (this->smearingLevels > 0)
     {
+      double start = usecond();
+
+      GaugeLinkField tmp_mu(force.Grid());
+
       for (int ismr = this->smearingLevels - 1; ismr > 0; --ismr) {
-	ln_det+= logDetJacobianForceLevel(this->get_smeared_conf(ismr-1),force,ismr);
+
+	// remove U in UdSdU...
+	for (int mu = 0; mu < Nd; mu++) {
+	  tmp_mu = adj(peekLorentz(this->get_smeared_conf(ismr), mu)) * peekLorentz(force, mu);
+	  pokeLorentz(force, tmp_mu, mu);
+	}
+	
+      	// Propagate existing force
+        force = this->AnalyticSmearedForce(force, this->get_smeared_conf(ismr - 1), ismr);
+
+	// Add back U in UdSdU...
+	for (int mu = 0; mu < Nd; mu++) {
+	  tmp_mu = peekLorentz(this->get_smeared_conf(ismr - 1), mu) * peekLorentz(force, mu);
+	  pokeLorentz(force, tmp_mu, mu);
+	}
+    	
+	// Get this levels determinant force
+	force_det = Zero();
+	logDetJacobianForceLevel(this->get_smeared_conf(ismr-1),force_det,ismr);
+
+	// Sum the contributions
+	force = force + force_det;
       }
-      ln_det +=logDetJacobianForeceLevel(*(this->ThinLinks),force,0);
-    }
+    
+      // remove U in UdSdU...
+      for (int mu = 0; mu < Nd; mu++) {
+	tmp_mu = adj(peekLorentz(this->get_smeared_conf(0), mu)) * peekLorentz(force, mu);
+	pokeLorentz(force, tmp_mu, mu);
+      }
+
+      force = this->AnalyticSmearedForce(force, *this->ThinLinks,0);
+
+      for (int mu = 0; mu < Nd; mu++) {
+	tmp_mu = peekLorentz(*this->ThinLinks, mu) * peekLorentz(force, mu);
+	pokeLorentz(force, tmp_mu, mu);
+      }
+
+      force_det = Zero();
+
+      logDetJacobianForceLevel(*this->ThinLinks,force_det,0);
+
+      force = force + force_det;
+
+      force=Ta(force); // Ta
+      
+      double end = usecond();
+      double time = (end - start)/ 1e3;
+      std::cout << GridLogMessage << "GaugeConfigurationMasked: lnDetJacobianForce took " << time << " ms" << std::endl;  
+    }  // if smearingLevels = 0 do nothing
   }
 
 private:
-  // Member functions
   //====================================================================
   // Override base clas here to mask it
   virtual void fill_smearedSet(GaugeField &U)
@@ -574,6 +667,7 @@ private:
       GaugeField smeared_B(this->ThinLinks->Grid());
 
       previous_u = *this->ThinLinks;
+      double start = usecond();
       for (int smearLvl = 0; smearLvl < this->smearingLevels; ++smearLvl)
       {
         this->StoutSmearing->smear(smeared_A, previous_u);
@@ -586,8 +680,11 @@ private:
 
         // For debug purposes
         RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(previous_u);
-        std::cout << GridLogMessage << "[SmearedConfigurationMasked] Plaq: " << impl_plaq << std::endl;
+        std::cout << GridLogMessage << "[SmearedConfigurationMasked] smeared Plaq: " << impl_plaq << std::endl;
       }
+      double end = usecond();
+      double time = (end - start)/ 1e3;
+      std::cout << GridLogMessage << "GaugeConfigurationMasked: Link smearing took " << time << " ms" << std::endl;  
     }
   }
   //====================================================================
@@ -603,13 +700,11 @@ private:
     GaugeLinkField iQ(grid), e_iQ(grid);
     GaugeLinkField SigmaKPrime_mu(grid);
     GaugeLinkField GaugeKmu(grid), Cmu(grid);
-
+    
     this->StoutSmearing->BaseSmear(C, GaugeK);
     SigmaK = Zero();
     iLambda = Zero();
 
-    SigmaK = Zero();
-
     SigmaKPrimeA = SigmaKPrime;
     ApplyMask(SigmaKPrimeA,level);
     SigmaKPrimeB = SigmaKPrime - SigmaKPrimeA;
@@ -633,251 +728,85 @@ private:
     // propagate the rest of the force as identity map, just add back
     ////////////////////////////////////////////////////////////////////////////////////
     SigmaK = SigmaK+SigmaKPrimeB;
+
     return SigmaK;
   }
 
-  ////////////////////////////////////////
-  // INHERIT THESE
-  ////////////////////////////////////////
-  
-  /*! @brief Returns smeared configuration at level 'Level' */
-  /*
-  const GaugeField &get_smeared_conf(int Level) const
-  {
-    return SmearedSet[Level];
-  }
-  */
-  
-  // Duplicates code that is in GaugeConfiguration.h
-  // Should inherit or share.
-  //====================================================================
-  /*
-  void set_iLambda(GaugeLinkField& iLambda, GaugeLinkField& e_iQ,
-                   const GaugeLinkField& iQ, const GaugeLinkField& Sigmap,
-                   const GaugeLinkField& GaugeK) const 
-  {
-    GridBase* grid = iQ.Grid();
-    GaugeLinkField iQ2(grid), iQ3(grid), B1(grid), B2(grid), USigmap(grid);
-    GaugeLinkField unity(grid);
-    unity = 1.0;
-
-    LatticeComplex u(grid), w(grid);
-    LatticeComplex f0(grid), f1(grid), f2(grid);
-    LatticeComplex xi0(grid), xi1(grid), tmp(grid);
-    LatticeComplex u2(grid), w2(grid), cosw(grid);
-    LatticeComplex emiu(grid), e2iu(grid), qt(grid), fden(grid);
-    LatticeComplex r01(grid), r11(grid), r21(grid), r02(grid), r12(grid);
-    LatticeComplex r22(grid), tr1(grid), tr2(grid);
-    LatticeComplex b10(grid), b11(grid), b12(grid), b20(grid), b21(grid),
-      b22(grid);
-    LatticeComplex LatticeUnitComplex(grid);
-
-    LatticeUnitComplex = 1.0;
-
-    // Exponential
-    iQ2 = iQ * iQ;
-    iQ3 = iQ * iQ2;
-    StoutSmearing->set_uw(u, w, iQ2, iQ3);
-    StoutSmearing->set_fj(f0, f1, f2, u, w);
-    e_iQ = f0 * unity + timesMinusI(f1) * iQ - f2 * iQ2;
-
-    // Getting B1, B2, Gamma and Lambda
-    // simplify this part, reduntant calculations in set_fj
-    xi0 = StoutSmearing->func_xi0(w);
-    xi1 = StoutSmearing->func_xi1(w);
-    u2 = u * u;
-    w2 = w * w;
-    cosw = cos(w);
-
-    emiu = cos(u) - timesI(sin(u));
-    e2iu = cos(2.0 * u) + timesI(sin(2.0 * u));
-
-    r01 = (2.0 * u + timesI(2.0 * (u2 - w2))) * e2iu +
-      emiu * ((16.0 * u * cosw + 2.0 * u * (3.0 * u2 + w2) * xi0) +
-	      timesI(-8.0 * u2 * cosw + 2.0 * (9.0 * u2 + w2) * xi0));
-
-    r11 = (2.0 * LatticeUnitComplex + timesI(4.0 * u)) * e2iu +
-      emiu * ((-2.0 * cosw + (3.0 * u2 - w2) * xi0) +
-	      timesI((2.0 * u * cosw + 6.0 * u * xi0)));
-
-    r21 =
-      2.0 * timesI(e2iu) + emiu * (-3.0 * u * xi0 + timesI(cosw - 3.0 * xi0));
-
-    r02 = -2.0 * e2iu +
-      emiu * (-8.0 * u2 * xi0 +
-	      timesI(2.0 * u * (cosw + xi0 + 3.0 * u2 * xi1)));
-
-    r12 = emiu * (2.0 * u * xi0 + timesI(-cosw - xi0 + 3.0 * u2 * xi1));
-
-    r22 = emiu * (xi0 - timesI(3.0 * u * xi1));
-
-    fden = LatticeUnitComplex / (2.0 * (9.0 * u2 - w2) * (9.0 * u2 - w2));
-
-    b10 = 2.0 * u * r01 + (3.0 * u2 - w2) * r02 - (30.0 * u2 + 2.0 * w2) * f0;
-    b11 = 2.0 * u * r11 + (3.0 * u2 - w2) * r12 - (30.0 * u2 + 2.0 * w2) * f1;
-    b12 = 2.0 * u * r21 + (3.0 * u2 - w2) * r22 - (30.0 * u2 + 2.0 * w2) * f2;
-
-    b20 = r01 - (3.0 * u) * r02 - (24.0 * u) * f0;
-    b21 = r11 - (3.0 * u) * r12 - (24.0 * u) * f1;
-    b22 = r21 - (3.0 * u) * r22 - (24.0 * u) * f2;
-
-    b10 *= fden;
-    b11 *= fden;
-    b12 *= fden;
-    b20 *= fden;
-    b21 *= fden;
-    b22 *= fden;
-
-    B1 = b10 * unity + timesMinusI(b11) * iQ - b12 * iQ2;
-    B2 = b20 * unity + timesMinusI(b21) * iQ - b22 * iQ2;
-    USigmap = GaugeK * Sigmap;
-
-    tr1 = trace(USigmap * B1);
-    tr2 = trace(USigmap * B2);
-
-    GaugeLinkField QUS = iQ * USigmap;
-    GaugeLinkField USQ = USigmap * iQ;
-
-    GaugeLinkField iGamma = tr1 * iQ - timesI(tr2) * iQ2 +
-      timesI(f1) * USigmap + f2 * QUS + f2 * USQ;
-
-    iLambda = Ta(iGamma);
-  }
-  */
-  //====================================================================
 public:
-  //  GaugeField* ThinLinks; /* Pointer to the thin links configuration -- base class*/ 
-  ////////////////////////
-  // Derived class
-  ////////////////////////
+
   /* Standard constructor */
-  SmearedConfigurationMasked(GridCartesian* _UGrid, unsigned int Nsmear, Smear_Stout<Gimpl>& Stout,bool domask=false)
+  SmearedConfigurationMasked(GridCartesian* _UGrid, unsigned int Nsmear, Smear_Stout<Gimpl>& Stout)
     : SmearedConfiguration<Gimpl>(_UGrid, Nsmear,Stout)
   {
-    if(domask) assert(Nsmear%(2*Nd)==0); // Or multiply by 8??
+    assert(Nsmear%(2*Nd)==0); // Or multiply by 8??
 
+    // was resized in base class
+    assert(this->SmearedSet.size()==Nsmear);
+    
     GridRedBlackCartesian * UrbGrid;
     UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(_UGrid);
     LatticeComplex one(_UGrid); one = ComplexD(1.0,0.0);
     LatticeComplex tmp(_UGrid);
 
     for (unsigned int i = 0; i < this->smearingLevels; ++i) {
-      this->SmearedSet.push_back(*(new GaugeField(_UGrid)));
-      masks.push_back(*(new LatticeLorentzComplex(_UGrid)));
-      if (domask) {
 
-	int mu= (i/2) %Nd;
-	int cb= (i%2);
-	LatticeComplex tmpcb(UrbGrid);
+      masks.push_back(*(new LatticeLorentzComplex(_UGrid)));
+
+      int mu= (i/2) %Nd;
+      int cb= (i%2);
+      LatticeComplex tmpcb(UrbGrid);
 	
-	masks[i]=Zero();
-	////////////////////
-	// Setup the mask
-	////////////////////
-	tmp = Zero();
-	pickCheckerboard(cb,tmpcb,one);
-	setCheckerboard(tmp,tmpcb);
-	PokeIndex<LorentzIndex>(masks[i],tmp, mu);
+      masks[i]=Zero();
+      ////////////////////
+      // Setup the mask
+      ////////////////////
+      tmp = Zero();
+      pickCheckerboard(cb,tmpcb,one);
+      setCheckerboard(tmp,tmpcb);
+      PokeIndex<LorentzIndex>(masks[i],tmp, mu);
 	
-      } else {
-	for(int mu=0;mu<Nd;mu++){
-	  PokeIndex<LorentzIndex>(masks[i],one, mu);
-	}
-      }
     }
     delete UrbGrid;
   }
-
-  //////////////////////////////////////////////////////////////
-  //Base functionality:
-  //////////////////////////////////////////////////////////////
   
-  /*! For just thin links */
-  //  SmearedConfigurationMasked()
-  //    : smearingLevels(0), StoutSmearing(nullptr), SmearedSet(), ThinLinks(NULL), UGrid(NULL), UrbGrid(NULL), masks() {}
-
-  // attach the smeared routines to the thin links U and fill the smeared set
-  /*
-  void set_Field(GaugeField &U)
+  virtual void smeared_force(GaugeField &SigmaTilde) 
   {
-    double start = usecond();
-    fill_smearedSet(U);
-    double end = usecond();
-    double time = (end - start)/ 1e3;
-    std::cout << GridLogMessage << "Smearing in " << time << " ms" << std::endl;  
-  }
-  */
-  
-  //====================================================================
-  /*
-  void smeared_force(GaugeField &SigmaTilde) 
-  {
-    if (smearingLevels > 0)
+    if (this->smearingLevels > 0)
     {
       double start = usecond();
       GaugeField force = SigmaTilde; // actually = U*SigmaTilde
       GaugeLinkField tmp_mu(SigmaTilde.Grid());
 
+      // Remove U from UdSdU
       for (int mu = 0; mu < Nd; mu++)
       {
         // to get just SigmaTilde
-        tmp_mu = adj(peekLorentz(SmearedSet[smearingLevels - 1], mu)) * peekLorentz(force, mu);
+        tmp_mu = adj(peekLorentz(this->SmearedSet[this->smearingLevels - 1], mu)) * peekLorentz(force, mu);
         pokeLorentz(force, tmp_mu, mu);
       }
 
-      for (int ismr = smearingLevels - 1; ismr > 0; --ismr) {
-        force = AnalyticSmearedForce(force, get_smeared_conf(ismr - 1),ismr);
+      for (int ismr = this->smearingLevels - 1; ismr > 0; --ismr) {
+        force = this->AnalyticSmearedForce(force, this->get_smeared_conf(ismr - 1),ismr);
       }
       
-      force = AnalyticSmearedForce(force, *ThinLinks,0);
+      force = this->AnalyticSmearedForce(force, *this->ThinLinks,0);
 
+      // Add U to UdSdU
       for (int mu = 0; mu < Nd; mu++)
       {
-        tmp_mu = peekLorentz(*ThinLinks, mu) * peekLorentz(force, mu);
+        tmp_mu = peekLorentz(*this->ThinLinks, mu) * peekLorentz(force, mu);
         pokeLorentz(SigmaTilde, tmp_mu, mu);
       }
+
+
       double end = usecond();
       double time = (end - start)/ 1e3;
-      std::cout << GridLogMessage << "Smearing force in " << time << " ms" << std::endl;  
+      std::cout << GridLogMessage << " GaugeConfigurationMasked: Smeared Force chain rule took " << time << " ms" << std::endl;
+
     }  // if smearingLevels = 0 do nothing
+    SigmaTilde=Gimpl::projectForce(SigmaTilde); // Ta
   }
-  */
-  //====================================================================
 
-  //  GaugeField& get_SmearedU() { return SmearedSet[smearingLevels - 1]; }
-  //  GaugeField& get_SmearedU(int n) { return this->SmearedSet[n]; }
-
-  /*
-  GaugeField &get_U(bool smeared = false)
-  {
-    // get the config, thin links by default
-    if (smeared)
-    {
-      if (smearingLevels)
-      {
-        RealD impl_plaq =
-	  WilsonLoops<Gimpl>::avgPlaquette(SmearedSet[smearingLevels - 1]);
-        std::cout << GridLogDebug << "getting Usmr Plaq: " << impl_plaq
-                  << std::endl;
-        return get_SmearedU();
-      }
-      else
-      {
-        RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
-        std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
-                  << std::endl;
-        return *ThinLinks;
-      }
-    }
-    else
-    {
-      RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
-      std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
-                << std::endl;
-      return *ThinLinks;
-    }
-  }
-  */
 };
 
 NAMESPACE_END(Grid);

Grid/qcd/smearing/JacobianAction.h

@@ -0,0 +1,87 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Source file: ./lib/qcd/action/gauge/JacobianAction.h
+
+Copyright (C) 2015
+
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+			   /*  END LEGAL */
+#pragma once
+
+NAMESPACE_BEGIN(Grid);
+
+////////////////////////////////////////////////////////////////////////
+// Jacobian Action .. 
+////////////////////////////////////////////////////////////////////////
+template <class Gimpl>
+class JacobianAction : public Action<typename Gimpl::GaugeField> {
+public:  
+  INHERIT_GIMPL_TYPES(Gimpl);
+
+  SmearedConfigurationMasked<Gimpl> * smearer;
+  /////////////////////////// constructors
+  explicit JacobianAction(SmearedConfigurationMasked<Gimpl> * _smearer ) { smearer=_smearer;};
+
+  virtual std::string action_name() {return "JacobianAction";}
+
+  virtual std::string LogParameters(){
+    std::stringstream sstream;
+    sstream << GridLogMessage << "[JacobianAction] " << std::endl;
+    return sstream.str();
+  }
+
+  //////////////////////////////////
+  // Usual cases are not used
+  //////////////////////////////////
+  virtual void refresh(const GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG){ assert(0);};
+  virtual RealD S(const GaugeField &U) { assert(0); }
+  virtual void deriv(const GaugeField &U, GaugeField &dSdU) { assert(0);  }
+
+  //////////////////////////////////
+  // Functions of smart configs only
+  //////////////////////////////////
+  virtual void refresh(ConfigurationBase<GaugeField> & U, GridSerialRNG &sRNG, GridParallelRNG& pRNG)
+  {
+    return;
+  }
+  virtual RealD S(ConfigurationBase<GaugeField>& U)
+  {
+    // det M = e^{ - ( - logDetM) }
+    assert( &U == smearer );
+    return -smearer->logDetJacobian();
+  }
+  virtual RealD Sinitial(ConfigurationBase<GaugeField>& U) 
+  {
+    return S(U);
+  }
+  virtual void deriv(ConfigurationBase<GaugeField>& U, GaugeField& dSdU)
+  {
+    assert( &U == smearer );
+    smearer->logDetJacobianForce(dSdU);
+  }
+
+private:
+ };
+
+NAMESPACE_END(Grid);
+

Grid/qcd/utils/SUn.h

@@ -34,6 +34,61 @@ directory
 
 NAMESPACE_BEGIN(Grid);
 
+template<int N, class Vec>
+Lattice<iScalar<iScalar<iScalar<Vec> > > > Determinant(const Lattice<iScalar<iScalar<iMatrix<Vec, N> > > > &Umu)
+{
+  GridBase *grid=Umu.Grid();
+  auto lvol = grid->lSites();
+  Lattice<iScalar<iScalar<iScalar<Vec> > > > ret(grid);
+  typedef typename Vec::scalar_type scalar;
+  autoView(Umu_v,Umu,CpuRead);
+  autoView(ret_v,ret,CpuWrite);
+  thread_for(site,lvol,{
+    Eigen::MatrixXcd EigenU = Eigen::MatrixXcd::Zero(N,N);
+    Coordinate lcoor;
+    grid->LocalIndexToLocalCoor(site, lcoor);
+    iScalar<iScalar<iMatrix<scalar, N> > > Us;
+    peekLocalSite(Us, Umu_v, lcoor);
+    for(int i=0;i<N;i++){
+      for(int j=0;j<N;j++){
+	scalar tmp= Us()()(i,j);
+	ComplexD ztmp(real(tmp),imag(tmp));
+	EigenU(i,j)=ztmp;
+      }}
+    ComplexD detD  = EigenU.determinant();
+    typename Vec::scalar_type det(detD.real(),detD.imag());
+    pokeLocalSite(det,ret_v,lcoor);
+  });
+  return ret;
+}
+
+template<int N, class Vec>
+static void ProjectSUn(Lattice<iScalar<iScalar<iMatrix<Vec, N> > > > &Umu)
+{
+  Umu      = ProjectOnGroup(Umu);
+  auto det = Determinant(Umu);
+
+  det = conjugate(det);
+
+  for(int i=0;i<N;i++){
+    auto element = PeekIndex<ColourIndex>(Umu,N-1,i);
+    element = element * det;
+    PokeIndex<ColourIndex>(Umu,element,Nc-1,i);
+  }
+}
+template<int N,class Vec>
+static void ProjectSUn(Lattice<iVector<iScalar<iMatrix<Vec, N> >,Nd> > &U)
+{
+  GridBase *grid=U.Grid();
+  // Reunitarise
+  for(int mu=0;mu<Nd;mu++){
+    auto Umu = PeekIndex<LorentzIndex>(U,mu);
+    Umu      = ProjectOnGroup(Umu);
+    ProjectSUn(Umu);
+    PokeIndex<LorentzIndex>(U,Umu,mu);
+  }
+}
+
 template <int ncolour>
 class SU {
 public:
@@ -741,8 +796,14 @@ public:
     typedef Lattice<vMatrixType> LatticeMatrixType;
 
     LatticeMatrixType Umu(out.Grid());
+    LatticeMatrixType tmp(out.Grid());
     for (int mu = 0; mu < Nd; mu++) {
-      LieRandomize(pRNG, Umu, 1.0);
+      //      LieRandomize(pRNG, Umu, 1.0);
+      //      PokeIndex<LorentzIndex>(out, Umu, mu);
+      gaussian(pRNG,Umu);
+      tmp = Ta(Umu);
+      taExp(tmp,Umu);
+      ProjectSUn(Umu);
       PokeIndex<LorentzIndex>(out, Umu, mu);
     }
   }
@@ -798,30 +859,6 @@ public:
   }
 };
 
-template<int N>
-LatticeComplexD Determinant(const Lattice<iScalar<iScalar<iMatrix<vComplexD, N> > > > &Umu)
-{
-  GridBase *grid=Umu.Grid();
-  auto lvol = grid->lSites();
-  LatticeComplexD ret(grid);
-
-  autoView(Umu_v,Umu,CpuRead);
-  autoView(ret_v,ret,CpuWrite);
-  thread_for(site,lvol,{
-    Eigen::MatrixXcd EigenU = Eigen::MatrixXcd::Zero(N,N);
-    Coordinate lcoor;
-    grid->LocalIndexToLocalCoor(site, lcoor);
-    iScalar<iScalar<iMatrix<ComplexD, N> > > Us;
-    peekLocalSite(Us, Umu_v, lcoor);
-    for(int i=0;i<N;i++){
-      for(int j=0;j<N;j++){
-	EigenU(i,j) = Us()()(i,j);
-      }}
-    ComplexD det = EigenU.determinant();
-    pokeLocalSite(det,ret_v,lcoor);
-  });
-  return ret;
-}
 template<int N>
 Lattice<iScalar<iScalar<iMatrix<vComplexD, N> > > > Inverse(const Lattice<iScalar<iScalar<iMatrix<vComplexD, N> > > > &Umu)
 {
@@ -851,32 +888,6 @@ Lattice<iScalar<iScalar<iMatrix<vComplexD, N> > > > Inverse(const Lattice<iScala
   });
   return ret;
 }
-template<int N>
-static void ProjectSUn(Lattice<iScalar<iScalar<iMatrix<vComplexD, N> > > > &Umu)
-{
-  Umu      = ProjectOnGroup(Umu);
-  auto det = Determinant(Umu);
-
-  det = conjugate(det);
-
-  for(int i=0;i<N;i++){
-    auto element = PeekIndex<ColourIndex>(Umu,N-1,i);
-    element = element * det;
-    PokeIndex<ColourIndex>(Umu,element,Nc-1,i);
-  }
-}
-template<int N>
-static void ProjectSUn(Lattice<iVector<iScalar<iMatrix<vComplexD, N> >,Nd> > &U)
-{
-  GridBase *grid=U.Grid();
-  // Reunitarise
-  for(int mu=0;mu<Nd;mu++){
-    auto Umu = PeekIndex<LorentzIndex>(U,mu);
-    Umu      = ProjectOnGroup(Umu);
-    ProjectSUn(Umu);
-    PokeIndex<LorentzIndex>(U,Umu,mu);
-  }
-}
 // Explicit specialisation for SU(3).
 // Explicit specialisation for SU(3).
 static void

Grid/stencil/Stencil.h

@@ -705,7 +705,7 @@ public:
 	}
       }
     }
-    std::cout << "BuildSurfaceList size is "<<surface_list.size()<<std::endl;
+    std::cout << GridLogDebug << "BuildSurfaceList size is "<<surface_list.size()<<std::endl;
   }
   /// Introduce a block structure and switch off comms on boundaries
   void DirichletBlock(const Coordinate &dirichlet_block)

Grid/tensors/Tensor_exp.h

@@ -55,7 +55,7 @@ template<class vtype, int N> accelerator_inline iVector<vtype, N> Exponentiate(c
 
 
 // Specialisation: Cayley-Hamilton exponential for SU(3)
-#ifndef GRID_ACCELERATED
+#if 0
 template<class vtype, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0>::type * =nullptr> 
 accelerator_inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, RealD alpha  , Integer Nexp = DEFAULT_MAT_EXP )
 {

HMC/FTHMC2p1f.cc

@@ -0,0 +1,224 @@
+/*************************************************************************************
+
+Grid physics library, www.github.com/paboyle/Grid
+
+Copyright (C) 2023
+
+Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation; either version 2 of the License, or
+(at your option) any later version.
+
+This program is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along
+with this program; if not, write to the Free Software Foundation, Inc.,
+51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+See the full license in the file "LICENSE" in the top level distribution
+directory
+*************************************************************************************/
+/*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/qcd/smearing/GaugeConfigurationMasked.h>
+#include <Grid/qcd/smearing/JacobianAction.h>
+
+using namespace Grid;
+
+int main(int argc, char **argv)
+{
+  std::cout << std::setprecision(12);
+  
+  Grid_init(&argc, &argv);
+  int threads = GridThread::GetThreads();
+  // here make a routine to print all the relevant information on the run
+  std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl;
+
+   // Typedefs to simplify notation
+  typedef WilsonImplR FermionImplPolicy;
+  typedef MobiusFermionD FermionAction;
+  typedef typename FermionAction::FermionField FermionField;
+
+  typedef Grid::XmlReader       Serialiser;
+
+  //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
+  IntegratorParameters MD;
+  //  typedef GenericHMCRunner<LeapFrog> HMCWrapper;
+  //  MD.name    = std::string("Leap Frog");
+  //  typedef GenericHMCRunner<ForceGradient> HMCWrapper;
+  //  MD.name    = std::string("Force Gradient");
+  typedef GenericHMCRunner<MinimumNorm2> HMCWrapper;
+  MD.name    = std::string("MinimumNorm2");
+  MD.MDsteps = 12;
+  MD.trajL   = 1.0;
+
+  HMCparameters HMCparams;
+  HMCparams.StartTrajectory  = 0;
+  HMCparams.Trajectories     = 200;
+  HMCparams.NoMetropolisUntil=  20;
+  // "[HotStart, ColdStart, TepidStart, CheckpointStart]\n";
+  HMCparams.StartingType     =std::string("HotStart");
+  HMCparams.MD = MD;
+  HMCWrapper TheHMC(HMCparams);
+
+  // Grid from the command line arguments --grid and --mpi
+  TheHMC.Resources.AddFourDimGrid("gauge"); // use default simd lanes decomposition
+
+  CheckpointerParameters CPparams;
+  CPparams.config_prefix = "ckpoint_EODWF_lat";
+  CPparams.smeared_prefix = "ckpoint_EODWF_lat_smr";
+  CPparams.rng_prefix    = "ckpoint_EODWF_rng";
+  CPparams.saveInterval  = 1;
+  CPparams.saveSmeared   = true;
+  CPparams.format        = "IEEE64BIG";
+  TheHMC.Resources.LoadNerscCheckpointer(CPparams);
+
+  RNGModuleParameters RNGpar;
+  RNGpar.serial_seeds = "1 2 3 4 5";
+  RNGpar.parallel_seeds = "6 7 8 9 10";
+  TheHMC.Resources.SetRNGSeeds(RNGpar);
+
+  // Construct observables
+  // here there is too much indirection
+  typedef PlaquetteMod<HMCWrapper::ImplPolicy> PlaqObs;
+  TheHMC.Resources.AddObservable<PlaqObs>();
+  //////////////////////////////////////////////
+
+  const int Ls      = 16;
+  Real beta         = 2.13;
+  Real light_mass   = 0.01;
+  Real strange_mass = 0.04;
+  Real pv_mass      = 1.0;
+  RealD M5  = 1.8;
+  RealD b   = 1.0; // Scale factor two
+  RealD c   = 0.0;
+
+  OneFlavourRationalParams OFRp;
+  OFRp.lo       = 1.0e-2;
+  OFRp.hi       = 64;
+  OFRp.MaxIter  = 10000;
+  OFRp.tolerance= 1.0e-10;
+  OFRp.degree   = 14;
+  OFRp.precision= 40;
+
+  std::vector<Real> hasenbusch({ 0.1 });
+
+  auto GridPtr   = TheHMC.Resources.GetCartesian();
+  auto GridRBPtr = TheHMC.Resources.GetRBCartesian();
+  auto FGrid     = SpaceTimeGrid::makeFiveDimGrid(Ls,GridPtr);
+  auto FrbGrid   = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,GridPtr);
+
+  IwasakiGaugeActionR GaugeAction(beta);
+
+  // temporarily need a gauge field
+  LatticeGaugeField U(GridPtr);
+  LatticeGaugeField Uhot(GridPtr);
+
+  // These lines are unecessary if BC are all periodic
+  std::vector<Complex> boundary = {1,1,1,-1};
+  FermionAction::ImplParams Params(boundary);
+
+  double StoppingCondition = 1e-10;
+  double MaxCGIterations = 30000;
+  ConjugateGradient<FermionField>  CG(StoppingCondition,MaxCGIterations);
+
+  bool ApplySmearing = true;
+  
+  ////////////////////////////////////
+  // Collect actions
+  ////////////////////////////////////
+  ActionLevel<HMCWrapper::Field> Level1(1);
+  ActionLevel<HMCWrapper::Field> Level2(2);
+  ActionLevel<HMCWrapper::Field> Level3(4);
+
+  ////////////////////////////////////
+  // Strange action
+  ////////////////////////////////////
+
+  MobiusEOFAFermionD Strange_Op_L (U , *FGrid , *FrbGrid , *GridPtr , *GridRBPtr , strange_mass, strange_mass, pv_mass, 0.0, -1, M5, b, c);
+  MobiusEOFAFermionD Strange_Op_R (U , *FGrid , *FrbGrid , *GridPtr , *GridRBPtr , pv_mass, strange_mass,      pv_mass, -1.0, 1, M5, b, c);
+  ExactOneFlavourRatioPseudoFermionAction<FermionImplPolicy> 
+    EOFA(Strange_Op_L, Strange_Op_R, 
+	 CG,
+	 CG, CG,
+	 CG, CG, 
+	 OFRp, false);
+
+  EOFA.is_smeared = ApplySmearing;
+  Level1.push_back(&EOFA);
+
+  ////////////////////////////////////
+  // up down action
+  ////////////////////////////////////
+  std::vector<Real> light_den;
+  std::vector<Real> light_num;
+
+  int n_hasenbusch = hasenbusch.size();
+  light_den.push_back(light_mass);
+  for(int h=0;h<n_hasenbusch;h++){
+    light_den.push_back(hasenbusch[h]);
+    light_num.push_back(hasenbusch[h]);
+  }
+  light_num.push_back(pv_mass);
+
+  std::vector<FermionAction *> Numerators;
+  std::vector<FermionAction *> Denominators;
+  std::vector<TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy> *> Quotients;
+
+  for(int h=0;h<n_hasenbusch+1;h++){
+    std::cout << GridLogMessage << " 2f quotient Action  "<< light_num[h] << " / " << light_den[h]<< std::endl;
+    Numerators.push_back  (new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_num[h],M5,b,c, Params));
+    Denominators.push_back(new FermionAction(U,*FGrid,*FrbGrid,*GridPtr,*GridRBPtr,light_den[h],M5,b,c, Params));
+    Quotients.push_back   (new TwoFlavourEvenOddRatioPseudoFermionAction<FermionImplPolicy>(*Numerators[h],*Denominators[h],CG,CG));
+  }
+
+  for(int h=0;h<n_hasenbusch+1;h++){
+    Quotients[h]->is_smeared = ApplySmearing;
+    Level1.push_back(Quotients[h]);
+  }
+
+  /////////////////////////////////////////////////////////////
+  // lnDetJacobianAction
+  /////////////////////////////////////////////////////////////
+  double rho = 0.1;  // smearing parameter
+  int Nsmear = 1;    // number of smearing levels - must be multiple of 2Nd
+  int Nstep  = 8*Nsmear;    // number of smearing levels - must be multiple of 2Nd
+  Smear_Stout<HMCWrapper::ImplPolicy> Stout(rho);
+  SmearedConfigurationMasked<HMCWrapper::ImplPolicy> SmearingPolicy(GridPtr, Nstep, Stout);
+  JacobianAction<HMCWrapper::ImplPolicy> Jacobian(&SmearingPolicy);
+  if( ApplySmearing ) Level2.push_back(&Jacobian);
+  std::cout << GridLogMessage << " Built the Jacobian "<< std::endl;
+
+
+  /////////////////////////////////////////////////////////////
+  // Gauge action
+  /////////////////////////////////////////////////////////////
+  //  GaugeAction.is_smeared = ApplySmearing;
+  GaugeAction.is_smeared = true;
+  Level3.push_back(&GaugeAction);
+
+  std::cout << GridLogMessage << " ************************************************"<< std::endl;
+  std::cout << GridLogMessage << " Action complete -- NO FERMIONS FOR NOW -- FIXME"<< std::endl;
+  std::cout << GridLogMessage << " ************************************************"<< std::endl;
+  std::cout << GridLogMessage <<  std::endl;
+  std::cout << GridLogMessage <<  std::endl;
+
+
+  std::cout << GridLogMessage << " Running the FT HMC "<< std::endl;
+
+  TheHMC.TheAction.push_back(Level1);
+  TheHMC.TheAction.push_back(Level2);
+  TheHMC.TheAction.push_back(Level3);
+
+  TheHMC.Run(SmearingPolicy); // for smearing
+
+  Grid_finalize();
+} // main
+
+
+

HMC/Mobius2p1f_EOFA_96I_hmc.cc

@@ -146,6 +146,8 @@ NAMESPACE_END(Grid);
 int main(int argc, char **argv) {
   using namespace Grid;
 
+  std::cout << " Grid Initialise "<<std::endl;
+  
   Grid_init(&argc, &argv);
 
   CartesianCommunicator::BarrierWorld();
@@ -170,24 +172,24 @@ int main(int argc, char **argv) {
   IntegratorParameters MD;
   //  typedef GenericHMCRunner<LeapFrog> HMCWrapper;
   //  MD.name    = std::string("Leap Frog");
-  typedef GenericHMCRunner<ForceGradient> HMCWrapper;
-  MD.name    = std::string("Force Gradient");
-  //typedef GenericHMCRunner<MinimumNorm2> HMCWrapper;
-  // MD.name    = std::string("MinimumNorm2");
+  //  typedef GenericHMCRunner<ForceGradient> HMCWrapper;
+  //  MD.name    = std::string("Force Gradient");
+  typedef GenericHMCRunner<MinimumNorm2> HMCWrapper;
+  MD.name    = std::string("MinimumNorm2");
   // TrajL = 2
   // 4/2 => 0.6 dH
   // 3/3 => 0.8 dH .. depth 3, slower
   //MD.MDsteps =  4;
-  MD.MDsteps =  12;
+  MD.MDsteps =  14;
   MD.trajL   = 0.5;
 
   HMCparameters HMCparams;
   HMCparams.StartTrajectory  = 1077;
-  HMCparams.Trajectories     = 1;
+  HMCparams.Trajectories     = 20;
   HMCparams.NoMetropolisUntil=  0;
   // "[HotStart, ColdStart, TepidStart, CheckpointStart]\n";
-  //  HMCparams.StartingType     =std::string("ColdStart");
-  HMCparams.StartingType     =std::string("CheckpointStart");
+  HMCparams.StartingType     =std::string("ColdStart");
+  //  HMCparams.StartingType     =std::string("CheckpointStart");
   HMCparams.MD = MD;
   HMCWrapper TheHMC(HMCparams);
 
@@ -223,7 +225,7 @@ int main(int argc, char **argv) {
   Real pv_mass      = 1.0;
   //  std::vector<Real> hasenbusch({ 0.01, 0.045, 0.108, 0.25, 0.51 , pv_mass });
   //  std::vector<Real> hasenbusch({ light_mass, 0.01, 0.045, 0.108, 0.25, 0.51 , pv_mass });
-  std::vector<Real> hasenbusch({ 0.005, 0.0145, 0.045, 0.108, 0.25, 0.51 , pv_mass }); // Updated
+  std::vector<Real> hasenbusch({ 0.005, 0.0145, 0.045, 0.108, 0.25, 0.51 }); // Updated
   //  std::vector<Real> hasenbusch({ light_mass, 0.0145, 0.045, 0.108, 0.25, 0.51 , 0.75 , pv_mass });
 
   auto GridPtr   = TheHMC.Resources.GetCartesian();
@@ -275,10 +277,10 @@ int main(int argc, char **argv) {
 
   //  double StoppingCondition = 1e-14;
   //  double MDStoppingCondition = 1e-9;
-  double StoppingCondition = 1e-8;
-  double MDStoppingCondition = 1e-7;
-  double MDStoppingConditionLoose = 1e-7;
-  double MDStoppingConditionStrange = 1e-7;
+  double StoppingCondition = 1e-9;
+  double MDStoppingCondition = 1e-8;
+  double MDStoppingConditionLoose = 1e-8;
+  double MDStoppingConditionStrange = 1e-8;
   double MaxCGIterations = 300000;
   ConjugateGradient<FermionField>  CG(StoppingCondition,MaxCGIterations);
   ConjugateGradient<FermionField>  MDCG(MDStoppingCondition,MaxCGIterations);

grid-config.in

@@ -60,7 +60,7 @@ while test $# -gt 0; do
     ;;
     
     --cxxflags)
-      echo @GRID_CXXFLAGS@
+      echo @GRID_CXXFLAGS@ -I@prefix@/include
     ;;
     
     --cxx)
@@ -72,11 +72,11 @@ while test $# -gt 0; do
     ;;
     
     --ldflags)
-      echo @GRID_LDFLAGS@
+      echo @GRID_LDFLAGS@ -L@prefix@/lib
     ;;
     
     --libs)
-      echo @GRID_LIBS@
+      echo @GRID_LIBS@ -lGrid
     ;;
     
     --summary)

systems/Lumi/benchmarks/bench2.slurm

@@ -0,0 +1,44 @@
+#!/bin/bash -l
+#SBATCH --job-name=bench_lehner
+#SBATCH --partition=small-g
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=8
+#SBATCH --cpus-per-task=7
+#SBATCH --gpus-per-node=8
+#SBATCH --time=00:10:00
+#SBATCH --account=project_465000546
+#SBATCH --gpu-bind=none
+#SBATCH --exclusive
+#SBATCH --mem=0
+
+CPU_BIND="map_cpu:48,56,32,40,16,24,1,8"
+echo $CPU_BIND
+
+cat << EOF > select_gpu
+#!/bin/bash
+export GPU_MAP=(0 1 2 3 4 5 6 7)
+export GPU=\${GPU_MAP[\$SLURM_LOCALID]}
+export HIP_VISIBLE_DEVICES=\$GPU
+unset ROCR_VISIBLE_DEVICES
+echo RANK \$SLURM_LOCALID using GPU \$GPU    
+exec \$*
+EOF
+
+chmod +x ./select_gpu
+
+root=/scratch/project_465000546/boylepet/Grid/systems/Lumi
+source ${root}/sourceme.sh
+
+export OMP_NUM_THREADS=7
+export MPICH_GPU_SUPPORT_ENABLED=1
+export MPICH_SMP_SINGLE_COPY_MODE=XPMEM
+
+for vol in 16.16.16.64 32.32.32.64  32.32.32.128
+do
+srun --cpu-bind=${CPU_BIND} ./select_gpu ./Benchmark_dwf_fp32 --mpi 2.2.2.2 --accelerator-threads 8 --comms-overlap --shm 2048 --shm-mpi 0 --grid $vol  > log.shm0.ov.$vol
+#srun --cpu-bind=${CPU_BIND} ./select_gpu ./Benchmark_dwf_fp32 --mpi 2.2.2.2 --accelerator-threads 8 --comms-overlap --shm 2048 --shm-mpi 1 --grid $vol  > log.shm1.ov.$vol
+
+srun --cpu-bind=${CPU_BIND} ./select_gpu ./Benchmark_dwf_fp32 --mpi 2.2.2.2 --accelerator-threads 8 --comms-sequential --shm 2048 --shm-mpi 0 --grid $vol  > log.shm0.seq.$vol
+#srun --cpu-bind=${CPU_BIND} ./select_gpu ./Benchmark_dwf_fp32 --mpi 2.2.2.2 --accelerator-threads 8 --comms-sequential --shm 2048 --shm-mpi 1 --grid $vol > log.shm1.seq.$vol
+done
+

systems/Lumi/config-command

@@ -3,30 +3,28 @@ spack load gmp
 spack load mpfr
 CLIME=`spack find --paths c-lime | grep c-lime| cut -c 15-`
 GMP=`spack find --paths gmp | grep gmp | cut -c 12-`
-MPFR=`spack find --paths mpfr | grep mpfr | cut -c 12-`
-echo clime $CLIME
-echo gmp $GMP
-echo mpfr $MPFR
+MPFR=`spack find --paths mpfr | grep mpfr | cut -c 13-`
+echo clime X$CLIME
+echo gmp X$GMP
+echo mpfr X$MPFR
 
-../../configure --enable-comms=mpi-auto \
+../../configure \
+--enable-comms=mpi-auto \
 --with-lime=$CLIME \
 --enable-unified=no \
 --enable-shm=nvlink \
---enable-tracing=timer \
 --enable-accelerator=hip \
 --enable-gen-simd-width=64 \
 --enable-simd=GPU \
---disable-accelerator-cshift \
---with-gmp=$OLCF_GMP_ROOT \
+--enable-accelerator-cshift \
+--with-gmp=$GMP \
+--with-mpfr=$MPFR \
 --with-fftw=$FFTW_DIR/.. \
---with-mpfr=/opt/cray/pe/gcc/mpfr/3.1.4/ \
 --disable-fermion-reps \
 --disable-gparity \
 CXX=hipcc MPICXX=mpicxx \
-CXXFLAGS="-fPIC -I{$ROCM_PATH}/include/ -std=c++14 -I${MPICH_DIR}/include -L/lib64 --amdgpu-target=gfx90a" \
- LDFLAGS="-L/lib64 -L/opt/rocm-5.2.0/lib/ -L${MPICH_DIR}/lib -lmpi -L${CRAY_MPICH_ROOTDIR}/gtl/lib -lmpi_gtl_hsa -lamdhip64 --amdgpu-target=gfx90a "
+  CXXFLAGS="-fPIC --offload-arch=gfx90a -I/opt/rocm/include/ -std=c++14 -I/opt/cray/pe/mpich/8.1.23/ofi/gnu/9.1/include" \
+  LDFLAGS="-L/opt/cray/pe/mpich/8.1.23/ofi/gnu/9.1/lib -lmpi -L/opt/cray/pe/mpich/8.1.23/gtl/lib -lmpi_gtl_hsa -lamdhip64 -fopenmp" 
 
 
-#--enable-simd=GPU-RRII \
-
 

systems/Lumi/sourceme.sh

@@ -1 +1,5 @@
-module load CrayEnv LUMI/22.12 partition/G  cray-fftw/3.3.10.1
+source ~/spack/share/spack/setup-env.sh
+module load CrayEnv LUMI/22.12 partition/G  cray-fftw/3.3.10.1 rocm
+spack load c-lime
+spack load gmp
+spack load mpfr

systems/Sunspot/benchmarks/bench.pbs

@@ -0,0 +1,46 @@
+#!/bin/bash
+
+#PBS -l select=1:system=sunspot,place=scatter
+#PBS -A LatticeQCD_aesp_CNDA
+#PBS -l walltime=01:00:00
+#PBS -N dwf
+#PBS -k doe
+
+HDIR=/home/paboyle/
+module use /soft/testing/modulefiles/
+module load intel-UMD23.05.25593.11/23.05.25593.11
+module load tools/pti-gpu  
+export LD_LIBRARY_PATH=$HDIR/tools/lib64:$LD_LIBRARY_PATH
+export PATH=$HDIR/tools/bin:$PATH
+
+export TZ='/usr/share/zoneinfo/US/Central'
+export OMP_PROC_BIND=spread
+export OMP_NUM_THREADS=3
+unset OMP_PLACES
+
+cd $PBS_O_WORKDIR
+
+qsub jobscript.pbs
+
+echo Jobid: $PBS_JOBID
+echo Running on host `hostname`
+echo Running on nodes `cat $PBS_NODEFILE`
+
+echo NODES
+cat $PBS_NODEFILE
+NNODES=`wc -l < $PBS_NODEFILE`
+NRANKS=12         # Number of MPI ranks per node
+NDEPTH=4          # Number of hardware threads per rank, spacing between MPI ranks on a node
+NTHREADS=$OMP_NUM_THREADS # Number of OMP threads per rank, given to OMP_NUM_THREADS
+
+NTOTRANKS=$(( NNODES * NRANKS ))
+
+echo "NUM_NODES=${NNODES}  TOTAL_RANKS=${NTOTRANKS}  RANKS_PER_NODE=${NRANKS}  THREADS_PER_RANK=${OMP_NUM_THREADS}"
+echo "OMP_PROC_BIND=$OMP_PROC_BIND OMP_PLACES=$OMP_PLACES"
+
+    
+CMD="mpiexec -np ${NTOTRANKS} -ppn ${NRANKS} -d ${NDEPTH} --cpu-bind=depth -envall \
+	     ./gpu_tile_compact.sh \
+	./Benchmark_dwf_fp32 --mpi 1.1.2.6 --grid 16.32.64.192 --comms-overlap \
+	--shm-mpi 0 --shm 2048 --device-mem 32000 --accelerator-threads 32"
+

systems/Sunspot/benchmarks/gpu_tile_compact.sh

@@ -0,0 +1,52 @@
+#!/bin/bash
+
+display_help() {
+  echo " Will map gpu tile to rank in compact and then round-robin fashion"
+  echo " Usage (only work for one node of ATS/PVC):"
+  echo "   mpiexec --np N gpu_tile_compact.sh ./a.out"
+  echo
+  echo " Example 3 GPU of 2 Tiles with 7 Ranks:"
+  echo "   0 Rank 0.0"
+  echo "   1 Rank 0.1"
+  echo "   2 Rank 1.0"
+  echo "   3 Rank 1.1"
+  echo "   4 Rank 2.0"
+  echo "   5 Rank 2.1"
+  echo "   6 Rank 0.0"
+  echo
+  echo " Hacked together by apl@anl.gov, please contact if bug found"
+  exit 1
+}
+
+#This give the exact GPU count i915 knows about and I use udev to only enumerate the devices with physical presence.
+#works? num_gpu=$(/usr/bin/udevadm info /sys/module/i915/drivers/pci\:i915/* |& grep -v Unknown | grep -c "P: /devices")
+num_gpu=6
+num_tile=2
+
+if [ "$#" -eq 0 ] || [ "$1" == "--help" ] || [ "$1" == "-h" ] || [ "$num_gpu" = 0 ]; then
+  display_help
+fi
+
+gpu_id=$(( (PALS_LOCAL_RANKID / num_tile ) % num_gpu ))
+tile_id=$((PALS_LOCAL_RANKID % num_tile))
+
+unset EnableWalkerPartition
+export EnableImplicitScaling=0
+export ZE_ENABLE_PCI_ID_DEVICE_ORDER=1
+export ZE_AFFINITY_MASK=$gpu_id.$tile_id
+export ONEAPI_DEVICE_FILTER=gpu,level_zero
+export SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0
+export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE=0:2
+export SYCL_PI_LEVEL_ZERO_USE_COPY_ENGINE_FOR_D2D_COPY=1
+#export SYCL_PI_LEVEL_ZERO_USM_RESIDENT=1
+
+echo "rank $PALS_RANKID ; local rank $PALS_LOCAL_RANKID ; ZE_AFFINITY_MASK=$ZE_AFFINITY_MASK"
+
+if [ $PALS_LOCAL_RANKID = 0 ]
+then
+    onetrace --chrome-device-timeline "$@"
+#    "$@"
+else
+"$@"
+fi

systems/Sunspot/config-command

@@ -0,0 +1,16 @@
+TOOLS=$HOME/tools
+../../configure \
+	--enable-simd=GPU \
+	--enable-gen-simd-width=64 \
+	--enable-comms=mpi-auto \
+	--enable-accelerator-cshift \
+	--disable-gparity \
+	--disable-fermion-reps \
+	--enable-shm=nvlink \
+	--enable-accelerator=sycl \
+	--enable-unified=no \
+	MPICXX=mpicxx \
+	CXX=icpx \
+	LDFLAGS="-fiopenmp -fsycl -fsycl-device-code-split=per_kernel -fsycl-device-lib=all -lze_loader -lapmidg -L$TOOLS/lib64/" \
+	CXXFLAGS="-fiopenmp -fsycl-unnamed-lambda -fsycl -I$INSTALL/include -Wno-tautological-compare -I$HOME/ -I$TOOLS/include"
+

systems/mac-arm/config-command-mpi

@@ -1,2 +1,4 @@
 BREW=/opt/local/
-MPICXX=mpicxx CXX=c++-12 ../../configure --enable-simd=GEN --enable-comms=mpi-auto --enable-unified=yes --prefix $HOME/QCD/GridInstallOpt --with-lime=/Users/peterboyle/QCD/SciDAC/install/ --with-openssl=$BREW
+CXX=mpicxx-openmpi-mp ../../configure --enable-simd=GEN --enable-comms=mpi --enable-unified=yes --prefix $HOME/QCD/GridInstall --with-lime=/Users/peterboyle/QCD/SciDAC/install/ --with-openssl=$BREW --disable-fermion-reps --disable-gparity --disable-debug
+
+

tests/core/Test_fft_pf.cc

@@ -0,0 +1,307 @@
+    /*************************************************************************************
+
+    grid` physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_cshift.cc
+
+    Copyright (C) 2015
+
+Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
+Author: Peter Boyle <paboyle@ph.ed.ac.uk>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+
+using namespace Grid;
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  Coordinate latt_size   = GridDefaultLatt();
+  Coordinate simd_layout( { vComplexD::Nsimd(),1,1,1});
+  Coordinate mpi_layout  = GridDefaultMpi();
+
+  int vol = 1;
+  for(int d=0;d<latt_size.size();d++){
+    vol = vol * latt_size[d];
+  }
+  GridCartesian         GRID(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian RBGRID(&GRID);
+
+  ComplexD ci(0.0,1.0);
+
+  std::vector<int> seeds({1,2,3,4});
+  GridSerialRNG          sRNG;  sRNG.SeedFixedIntegers(seeds); // naughty seeding
+  GridParallelRNG          pRNG(&GRID);
+  pRNG.SeedFixedIntegers(seeds);
+
+  LatticeGaugeFieldD Umu(&GRID);
+
+  SU<Nc>::ColdConfiguration(pRNG,Umu); // Unit gauge
+
+  ////////////////////////////////////////////////////
+  // PF prop
+  ////////////////////////////////////////////////////
+  LatticeFermionD    src(&GRID);
+
+  gaussian(pRNG,src);
+#if 1
+    Coordinate point(4,0);
+    src=Zero();
+    SpinColourVectorD ferm; gaussian(sRNG,ferm);
+    pokeSite(ferm,src,point);
+#endif
+  
+  {
+    std::cout<<"****************************************"<<std::endl;
+    std::cout << "Testing PartialFraction Hw kernel Mom space 4d propagator \n";
+    std::cout<<"****************************************"<<std::endl;
+
+    //    LatticeFermionD    src(&GRID); gaussian(pRNG,src);
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    diff(&GRID);
+
+    const int Ls=48+1;
+    GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,&GRID);
+    GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,&GRID);
+
+    RealD mass=0.1;
+    RealD M5  =0.8;
+    OverlapWilsonPartialFractionZolotarevFermionD Dov(Umu,*FGrid,*FrbGrid,GRID,RBGRID,mass,M5,0.001,8.0);
+
+    // Momentum space prop
+    std::cout << " Solving by FFT and Feynman rules" <<std::endl;
+    bool fiveD = false; //calculate 4d free propagator
+
+    std::cout << " Free propagator " <<std::endl;
+    Dov.FreePropagator(src,ref,mass) ;
+    std::cout << " Free propagator norm "<< norm2(ref) <<std::endl;
+
+    Gamma G5(Gamma::Algebra::Gamma5);
+
+    LatticeFermionD    src5(FGrid); src5=Zero();
+    LatticeFermionD    tmp5(FGrid); 
+    LatticeFermionD    result5(FGrid); result5=Zero();
+    LatticeFermionD    result4(&GRID); 
+    const int sdir=0;
+
+    ////////////////////////////////////////////////////////////////////////
+    // Import
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Free propagator Import "<< norm2(src) <<std::endl;
+    Dov.ImportPhysicalFermionSource  (src,src5);
+    std::cout << " Free propagator Imported "<< norm2(src5) <<std::endl;
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Conjugate gradient on normal equations system
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Solving by Conjugate Gradient (CGNE)" <<std::endl;
+    Dov.Mdag(src5,tmp5);
+    src5=tmp5;
+    MdagMLinearOperator<OverlapWilsonPartialFractionZolotarevFermionD,LatticeFermionD> HermOp(Dov);
+    ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
+    CG(HermOp,src5,result5);
+    ////////////////////////////////////////////////////////////////////////
+    // Domain wall physical field propagator
+    ////////////////////////////////////////////////////////////////////////
+    Dov.ExportPhysicalFermionSolution(result5,result4);
+
+    // From DWF4d.pdf :
+    //
+    // Dov_pf = 2/(1-m) D_cayley_ovlap  [ Page 43 ]
+    // Dinv_cayley_ovlap = 2/(1-m) Dinv_pf 
+    // Dinv_cayley_surface =1/(1-m) ( Dinv_cayley_ovlap - 1 ) =>  2/(1-m)^2 Dinv_pf - 1/(1-m) * src   [ Eq.2.67 ]
+
+    RealD scale = 2.0/(1.0-mass)/(1.0-mass);
+    result4 = result4 * scale;
+    result4 = result4 - src*(1.0/(1.0-mass)); // Subtract contact term
+    DumpSliceNorm("Src",src);
+    DumpSliceNorm("Grid",result4);
+    DumpSliceNorm("Fourier",ref);
+
+    std::cout << "Dov result4 "<<norm2(result4)<<std::endl;
+    std::cout << "Dov ref     "<<norm2(ref)<<std::endl;
+
+    diff = result4- ref;
+    DumpSliceNorm("diff ",diff);
+    
+  }
+  
+  ////////////////////////////////////////////////////
+  // Dwf prop
+  ////////////////////////////////////////////////////
+  {
+    std::cout<<"****************************************"<<std::endl;
+    std::cout << "Testing Dov(Hw) Mom space 4d propagator \n";
+    std::cout<<"****************************************"<<std::endl;
+
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    diff(&GRID);
+
+    const int Ls=48;
+    GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,&GRID);
+    GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,&GRID);
+
+    RealD mass=0.1;
+    RealD M5  =0.8;
+
+    OverlapWilsonCayleyTanhFermionD Dov(Umu,*FGrid,*FrbGrid,GRID,RBGRID,mass,M5,1.0);
+
+    // Momentum space prop
+    std::cout << " Solving by FFT and Feynman rules" <<std::endl;
+    Dov.FreePropagator(src,ref,mass) ;
+
+    Gamma G5(Gamma::Algebra::Gamma5);
+
+    LatticeFermionD    src5(FGrid); src5=Zero();
+    LatticeFermionD    tmp5(FGrid); 
+    LatticeFermionD    result5(FGrid); result5=Zero();
+    LatticeFermionD    result4(&GRID); 
+    const int sdir=0;
+
+    ////////////////////////////////////////////////////////////////////////
+    // Domain wall physical field source; need D_minus
+    ////////////////////////////////////////////////////////////////////////
+    /*
+	chi_5[0]   = chiralProjectPlus(chi);
+	chi_5[Ls-1]= chiralProjectMinus(chi);
+    */      
+    tmp =   (src + G5*src)*0.5;      InsertSlice(tmp,src5,   0,sdir);
+    tmp =   (src - G5*src)*0.5;      InsertSlice(tmp,src5,Ls-1,sdir);
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Conjugate gradient on normal equations system
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Solving by Conjugate Gradient (CGNE)" <<std::endl;
+    Dov.Dminus(src5,tmp5);
+    src5=tmp5;
+    Dov.Mdag(src5,tmp5);
+    src5=tmp5;
+    MdagMLinearOperator<OverlapWilsonCayleyTanhFermionD,LatticeFermionD> HermOp(Dov);
+    ConjugateGradient<LatticeFermionD> CG(1.0e-16,10000);
+    CG(HermOp,src5,result5);
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Domain wall physical field propagator
+    ////////////////////////////////////////////////////////////////////////
+    /*
+      psi  = chiralProjectMinus(psi_5[0]);
+      psi += chiralProjectPlus(psi_5[Ls-1]);
+    */
+    ExtractSlice(tmp,result5,0   ,sdir);   result4 =         (tmp-G5*tmp)*0.5;
+    ExtractSlice(tmp,result5,Ls-1,sdir);   result4 = result4+(tmp+G5*tmp)*0.5;
+    
+    std::cout << " Taking difference" <<std::endl;
+    std::cout << "Dov result4 "<<norm2(result4)<<std::endl;
+    std::cout << "Dov ref     "<<norm2(ref)<<std::endl;
+    DumpSliceNorm("Grid",result4);
+    DumpSliceNorm("Fourier",ref);
+    diff = ref - result4;
+    std::cout << "result - ref     "<<norm2(diff)<<std::endl;
+    
+    DumpSliceNorm("diff",diff);
+
+  }
+
+  
+  {
+    std::cout<<"****************************************"<<std::endl;
+    std::cout << "Testing PartialFraction Hw kernel Mom space 4d propagator with q\n";
+    std::cout<<"****************************************"<<std::endl;
+
+    //    LatticeFermionD    src(&GRID); gaussian(pRNG,src);
+    LatticeFermionD    tmp(&GRID);
+    LatticeFermionD    ref(&GRID);
+    LatticeFermionD    diff(&GRID);
+
+    const int Ls=48+1;
+    GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,&GRID);
+    GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,&GRID);
+
+    RealD mass=0.1;
+    RealD M5  =0.8;
+    OverlapWilsonPartialFractionZolotarevFermionD Dov(Umu,*FGrid,*FrbGrid,GRID,RBGRID,mass,M5,0.001,8.0);
+
+    // Momentum space prop
+    std::cout << " Solving by FFT and Feynman rules" <<std::endl;
+    bool fiveD = false; //calculate 4d free propagator
+
+    std::cout << " Free propagator " <<std::endl;
+    Dov.FreePropagator(src,ref,mass) ;
+    std::cout << " Free propagator norm "<< norm2(ref) <<std::endl;
+
+    Gamma G5(Gamma::Algebra::Gamma5);
+
+    LatticeFermionD    src5(FGrid); src5=Zero();
+    LatticeFermionD    tmp5(FGrid); 
+    LatticeFermionD    result5(FGrid); result5=Zero();
+    LatticeFermionD    result4(&GRID); 
+    const int sdir=0;
+
+    ////////////////////////////////////////////////////////////////////////
+    // Import
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Free propagator Import "<< norm2(src) <<std::endl;
+    Dov.ImportPhysicalFermionSource  (src,src5);
+    std::cout << " Free propagator Imported "<< norm2(src5) <<std::endl;
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Conjugate gradient on normal equations system
+    ////////////////////////////////////////////////////////////////////////
+    std::cout << " Solving by Conjugate Gradient (CGNE)" <<std::endl;
+    Dov.Mdag(src5,tmp5);
+    src5=tmp5;
+    MdagMLinearOperator<OverlapWilsonPartialFractionZolotarevFermionD,LatticeFermionD> HermOp(Dov);
+    ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
+    CG(HermOp,src5,result5);
+    ////////////////////////////////////////////////////////////////////////
+    // Domain wall physical field propagator
+    ////////////////////////////////////////////////////////////////////////
+    Dov.ExportPhysicalFermionSolution(result5,result4);
+
+    // From DWF4d.pdf :
+    //
+    // Dov_pf = 2/(1-m) D_cayley_ovlap  [ Page 43 ]
+    // Dinv_cayley_ovlap = 2/(1-m) Dinv_pf 
+    // Dinv_cayley_surface =1/(1-m) ( Dinv_cayley_ovlap - 1 ) =>  2/(1-m)^2 Dinv_pf - 1/(1-m) * src   [ Eq.2.67 ]
+
+    RealD scale = 2.0/(1.0-mass)/(1.0-mass);
+    result4 = result4 * scale;
+    result4 = result4 - src*(1.0/(1.0-mass)); // Subtract contact term
+    DumpSliceNorm("Src",src);
+    DumpSliceNorm("Grid",result4);
+    DumpSliceNorm("Fourier",ref);
+
+    std::cout << "Dov result4 "<<norm2(result4)<<std::endl;
+    std::cout << "Dov ref     "<<norm2(ref)<<std::endl;
+
+    diff = result4- ref;
+    DumpSliceNorm("diff ",diff);
+    
+  }
+
+  
+  Grid_finalize();
+}

tests/forces/Test_fthmc.cc

@@ -0,0 +1,219 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_fthmc.cc
+
+    Copyright (C) 2022
+
+Author: Peter Boyle <pboyle@bnl.gov>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+    See the full license in the file "LICENSE" in the top level distribution directory
+    *************************************************************************************/
+    /*  END LEGAL */
+#include <Grid/Grid.h>
+#include <Grid/qcd/smearing/GaugeConfigurationMasked.h>
+#include <Grid/qcd/smearing/JacobianAction.h>
+
+using namespace std;
+using namespace Grid;
+
+typedef MobiusFermionD FermionAction;
+typedef WilsonImplD FimplD;
+typedef WilsonImplD FermionImplPolicy;
+
+template<class Gimpl>
+void ForceTest(Action<LatticeGaugeField> &action,ConfigurationBase<LatticeGaugeField> & smU,MomentumFilterBase<LatticeGaugeField> &Filter)
+{
+  LatticeGaugeField U = smU.get_U(false); // unsmeared config
+  GridBase *UGrid = U.Grid();
+
+  std::vector<int> seeds({1,2,3,5});
+  GridSerialRNG            sRNG;         sRNG.SeedFixedIntegers(seeds);
+  GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds);
+
+  LatticeColourMatrix Pmu(UGrid); 
+  LatticeGaugeField P(UGrid); 
+  LatticeGaugeField UdSdU(UGrid); 
+
+  std::cout << GridLogMessage << "*********************************************************"<<std::endl;
+  std::cout << GridLogMessage << " Force test for "<<action.action_name()<<std::endl;
+  std::cout << GridLogMessage << "*********************************************************"<<std::endl;
+  
+  RealD eps=0.01;
+
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  std::cout << GridLogMessage << " Refresh "<<action.action_name()<<std::endl;
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  
+  Gimpl::generate_momenta(P,sRNG,RNG4);
+  //  Filter.applyFilter(P);
+
+  action.refresh(smU,sRNG,RNG4);
+
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  std::cout << GridLogMessage << " Action "<<action.action_name()<<std::endl;
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+
+  RealD S1 = action.S(smU);
+
+  Gimpl::update_field(P,U,eps);
+  smU.set_Field(U);
+
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  std::cout << GridLogMessage << " Derivative "<<action.action_name()<<std::endl;
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  action.deriv(smU,UdSdU);
+  UdSdU = Ta(UdSdU);
+  //  Filter.applyFilter(UdSdU);
+
+  DumpSliceNorm("Force",UdSdU,Nd-1);
+  
+  Gimpl::update_field(P,U,eps);
+  smU.set_Field(U);
+
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  std::cout << GridLogMessage << " Action "<<action.action_name()<<std::endl;
+  std::cout << GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  
+  RealD S2 = action.S(smU);
+
+  // Use the derivative
+  LatticeComplex dS(UGrid); dS = Zero();
+  for(int mu=0;mu<Nd;mu++){
+    auto UdSdUmu = PeekIndex<LorentzIndex>(UdSdU,mu);
+    Pmu= PeekIndex<LorentzIndex>(P,mu);
+    dS = dS - trace(Pmu*UdSdUmu)*eps*2.0*HMC_MOMENTUM_DENOMINATOR;
+  }
+  ComplexD dSpred    = sum(dS);
+  RealD diff =  S2-S1-dSpred.real();
+
+  std::cout<< GridLogMessage << "+++++++++++++++++++++++++++++++++++++++++++++++++++++++++"<<std::endl;
+  std::cout<< GridLogMessage << "S1 : "<< S1    <<std::endl;
+  std::cout<< GridLogMessage << "S2 : "<< S2    <<std::endl;
+  std::cout<< GridLogMessage << "dS : "<< S2-S1 <<std::endl;
+  std::cout<< GridLogMessage << "dSpred : "<< dSpred.real() <<std::endl;
+  std::cout<< GridLogMessage << "diff : "<< diff<<std::endl;
+  std::cout<< GridLogMessage << "*********************************************************"<<std::endl;
+  //  assert(diff<1.0);
+  std::cout<< GridLogMessage << "Done" <<std::endl;
+  std::cout << GridLogMessage << "*********************************************************"<<std::endl;
+}
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::cout << std::setprecision(14);
+  Coordinate latt_size   = GridDefaultLatt();
+  Coordinate mpi_layout  = GridDefaultMpi();
+  Coordinate simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  Coordinate shm;
+  GlobalSharedMemory::GetShmDims(mpi_layout,shm);
+
+  const int Ls=12;
+  const int Nt = latt_size[3];
+  GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
+  GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
+  GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+  GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
+
+  
+  ///////////////////// Gauge Field and Gauge Forces ////////////////////////////
+  LatticeGaugeField U(UGrid);
+
+#if  0
+  FieldMetaData header;
+  std::string file("./ckpoint_lat.2000");
+  NerscIO::readConfiguration(U,header,file);
+#else
+  std::vector<int> seeds({1,2,3,4,5,6,7,8});
+  GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds);
+  SU<Nc>::HotConfiguration(RNG4,U);
+#endif
+
+  
+  WilsonGaugeActionR  PlaqAction(6.0);
+  IwasakiGaugeActionR RectAction(2.13);
+  PlaqAction.is_smeared = true;  
+  RectAction.is_smeared = true;  
+
+  ////////////////////////////////////
+  // Fermion Action
+  ////////////////////////////////////
+  RealD mass=0.01; 
+  RealD pvmass=1.0; 
+  RealD M5=1.8; 
+  RealD b=1.5;
+  RealD c=0.5;
+  
+  // Double versions
+  std::vector<Complex> boundary = {1,1,1,-1};
+  FermionAction::ImplParams Params(boundary);
+  FermionAction DdwfPeriodic(U,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c,Params);
+  FermionAction PVPeriodic  (U,*FGrid,*FrbGrid,*UGrid,*UrbGrid,pvmass,M5,b,c,Params);
+
+  double StoppingCondition = 1.0e-8;
+  double MaxCGIterations = 50000;
+  ConjugateGradient<LatticeFermion>  CG(StoppingCondition,MaxCGIterations);
+
+  TwoFlavourRatioPseudoFermionAction<FimplD> Nf2(PVPeriodic, DdwfPeriodic,CG,CG);
+  Nf2.is_smeared = true;  
+  
+  ////////////////////////////////////////////////
+  // Plaquette only FTHMC smearer
+  ////////////////////////////////////////////////
+  double rho = 0.1;
+  Smear_Stout<PeriodicGimplR> Smearer(rho);
+  SmearedConfigurationMasked<PeriodicGimplR> SmartConfig(UGrid,2*Nd,Smearer);
+  SmearedConfiguration<PeriodicGimplR> StoutConfig(UGrid,1,Smearer);
+
+  JacobianAction<PeriodicGimplR> Jacobian(&SmartConfig);
+  
+  ////////////////////////////////////////////////
+  // Run some tests
+  ////////////////////////////////////////////////
+  MomentumFilterNone<LatticeGaugeField> FilterNone;
+
+  std::cout << " *********  FIELD TRANSFORM SMEARING ***** "<<std::endl;
+
+  SmartConfig.set_Field(U);
+  ForceTest<GimplTypesR>(PlaqAction,SmartConfig,FilterNone);
+
+  SmartConfig.set_Field(U);
+  ForceTest<GimplTypesR>(RectAction,SmartConfig,FilterNone);
+
+  SmartConfig.set_Field(U);
+  ForceTest<GimplTypesR>(Jacobian,SmartConfig,FilterNone);
+
+  SmartConfig.set_Field(U);
+  ForceTest<GimplTypesR>(Nf2,SmartConfig,FilterNone);
+
+  std::cout << " *********    STOUT SMEARING ***** "<<std::endl;
+
+  StoutConfig.set_Field(U);
+  ForceTest<GimplTypesR>(PlaqAction,StoutConfig,FilterNone);
+
+  StoutConfig.set_Field(U);
+  ForceTest<GimplTypesR>(RectAction,StoutConfig,FilterNone);
+  
+  StoutConfig.set_Field(U);
+  ForceTest<GimplTypesR>(Nf2,StoutConfig,FilterNone);
+  
+
+  Grid_finalize();
+}

tests/qdpxx/Test_qdpxx_munprec.cc

@@ -1,7 +1,6 @@
     /*************************************************************************************
 
     Grid physics library, www.github.com/paboyle/Grid 
-
     Source file: ./tests/qdpxx/Test_qdpxx_munprec.cc
 
     Copyright (C) 2015
@@ -26,13 +25,17 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
     See the full license in the file "LICENSE" in the top level distribution directory
     *************************************************************************************/
     /*  END LEGAL */
+#include <chroma.h>
+#include <actions/ferm/invert/syssolver_linop_cg_array.h>
+#include <actions/ferm/invert/syssolver_linop_aggregate.h>
+
 #include <Grid/Grid.h>
 
 int    Ls=8;
 double M5=1.6;
 double mq=0.01;
-double zolo_lo = 0.1;
-double zolo_hi = 2.0;
+double zolo_lo = 0.01;
+double zolo_hi = 7.0;
 double mobius_scale=2.0;
 
 enum ChromaAction {
@@ -55,11 +58,6 @@ enum ChromaAction {
 void calc_grid      (ChromaAction action,Grid::LatticeGaugeField & lat, Grid::LatticeFermion &src, Grid::LatticeFermion &res,int dag);
 void calc_chroma    (ChromaAction action,Grid::LatticeGaugeField & lat, Grid::LatticeFermion &src, Grid::LatticeFermion &res,int dag);
 
-#include <chroma.h>
-#include <actions/ferm/invert/syssolver_linop_cg_array.h>
-#include <actions/ferm/invert/syssolver_linop_aggregate.h>
-
-
 
 namespace Chroma { 
 
@@ -81,7 +79,7 @@ public:
 
     std::vector<int> x(4);
     QDP::multi1d<int> cx(4);
-    std::vector<int> gd= gr.Grid()->GlobalDimensions();
+    Grid::Coordinate gd = gr.Grid()->GlobalDimensions();
 
     for (x[0]=0;x[0]<gd[0];x[0]++){
     for (x[1]=0;x[1]<gd[1];x[1]++){
@@ -124,7 +122,7 @@ public:
 
     std::vector<int> x(5);
     QDP::multi1d<int> cx(4);
-    std::vector<int> gd= gr.Grid()->GlobalDimensions();
+    Grid::Coordinate gd= gr.Grid()->GlobalDimensions();
 
     for (x[0]=0;x[0]<gd[0];x[0]++){
     for (x[1]=0;x[1]<gd[1];x[1]++){
@@ -166,7 +164,7 @@ public:
 
     std::vector<int> x(5);
     QDP::multi1d<int> cx(4);
-    std::vector<int> gd= gr.Grid()->GlobalDimensions();
+    Grid::Coordinate gd= gr.Grid()->GlobalDimensions();
 
     for (x[0]=0;x[0]<gd[0];x[0]++){
     for (x[1]=0;x[1]<gd[1];x[1]++){
@@ -304,7 +302,30 @@ public:
      //     param.approximation_type=COEFF_TYPE_TANH_UNSCALED;
      //     param.approximation_type=COEFF_TYPE_TANH;
      param.tuning_strategy_xml=
-"<TuningStrategy><Name>OVEXT_CONSTANT_STRATEGY</Name></TuningStrategy>\n";
+"<TuningStrategy><Name>OVEXT_CONSTANT_STRATEGY</Name><TuningConstant>1.0</TuningConstant></TuningStrategy>\n";
+     UnprecOvExtFermActArray S_f(cfs,param);
+     Handle< FermState<T4,U,U> > fs( S_f.createState(u) );
+     Handle< LinearOperatorArray<T4> > M(S_f.linOp(fs));
+     return M;
+   }
+   if ( parms == HwPartFracTanh ) {
+     if ( Ls%2 == 0 ) { 
+       printf("Ls is not odd\n");
+       exit(-1);
+     }
+     UnprecOvExtFermActArrayParams param;
+     param.OverMass=M5; 
+     param.Mass=_mq;
+     param.RatPolyDeg = Ls;
+     param.ApproxMin =eps_lo;
+     param.ApproxMax =eps_hi;
+     param.b5 =1.0;
+     param.c5 =1.0;
+     //     param.approximation_type=COEFF_TYPE_ZOLOTAREV;
+     param.approximation_type=COEFF_TYPE_TANH_UNSCALED;
+     //param.approximation_type=COEFF_TYPE_TANH;
+     param.tuning_strategy_xml=
+       "<TuningStrategy><Name>OVEXT_CONSTANT_STRATEGY</Name><TuningConstant>1.0</TuningConstant></TuningStrategy>\n";
      UnprecOvExtFermActArray S_f(cfs,param);
      Handle< FermState<T4,U,U> > fs( S_f.createState(u) );
      Handle< LinearOperatorArray<T4> > M(S_f.linOp(fs));
@@ -316,7 +337,35 @@ public:
      param.ApproxMin=eps_lo;
      param.ApproxMax=eps_hi;
      param.approximation_type=COEFF_TYPE_ZOLOTAREV;
-     param.RatPolyDeg=Ls;
+     param.RatPolyDeg=Ls-1;
+     // The following is why I think Chroma made some directional errors:
+     param.AuxFermAct= std::string(
+"<AuxFermAct>\n"
+"  <FermAct>UNPRECONDITIONED_WILSON</FermAct>\n"
+"  <Mass>-1.8</Mass>\n"
+"  <b5>1</b5>\n"
+"  <c5>0</c5>\n"
+"  <MaxCG>1000</MaxCG>\n"
+"  <RsdCG>1.0e-9</RsdCG>\n"
+"  <FermionBC>\n"
+"      <FermBC>SIMPLE_FERMBC</FermBC>\n"
+"      <boundary>1 1 1 1</boundary>\n"
+"   </FermionBC> \n"
+"</AuxFermAct>"
+);
+     param.AuxFermActGrp= std::string("");
+     UnprecOvlapContFrac5DFermActArray S_f(fbc,param);
+     Handle< FermState<T4,U,U> > fs( S_f.createState(u) );
+     Handle< LinearOperatorArray<T4> > M(S_f.linOp(fs));
+     return  M;
+   }
+   if ( parms == HwContFracTanh ) {
+     UnprecOvlapContFrac5DFermActParams param;
+     param.Mass=_mq; // How is M5 set? Wilson mass In AuxFermAct
+     param.ApproxMin=eps_lo;
+     param.ApproxMax=eps_hi;
+     param.approximation_type=COEFF_TYPE_TANH_UNSCALED;
+     param.RatPolyDeg=Ls-1;
      // The following is why I think Chroma made some directional errors:
      param.AuxFermAct= std::string(
 "<AuxFermAct>\n"
@@ -378,7 +427,14 @@ int main (int argc,char **argv )
    * Setup QDP
    *********************************************************/
   Chroma::initialize(&argc,&argv);
-  Chroma::WilsonTypeFermActs4DEnv::registerAll(); 
+  //  Chroma::WilsonTypeFermActs4DEnv::registerAll(); 
+  Chroma::WilsonTypeFermActsEnv::registerAll(); 
+  //bool linkageHack(void)
+  //{
+  //  bool foo = true;
+  // Inline Measurements
+  //  InlineAggregateEnv::registerAll();
+  //  GaugeInitEnv::registerAll();
 
   /********************************************************
    * Setup Grid
@@ -388,26 +444,34 @@ int main (int argc,char **argv )
                                                                        Grid::GridDefaultSimd(Grid::Nd,Grid::vComplex::Nsimd()),
                                                                        Grid::GridDefaultMpi());
   
-  std::vector<int> gd = UGrid->GlobalDimensions();
+  Grid::Coordinate gd = UGrid->GlobalDimensions();
   QDP::multi1d<int> nrow(QDP::Nd);
   for(int mu=0;mu<4;mu++) nrow[mu] = gd[mu];
 
   QDP::Layout::setLattSize(nrow);
   QDP::Layout::create();
 
-  Grid::GridCartesian         * FGrid   = Grid::SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
-  Grid::LatticeGaugeField lat(UGrid);
-  Grid::LatticeFermion    src(FGrid);
-  Grid::LatticeFermion    res_chroma(FGrid);
-  Grid::LatticeFermion    res_grid  (FGrid);
-  
   std::vector<ChromaAction> ActionList({
 		 HtCayleyTanh, // Plain old DWF.
 		 HmCayleyTanh,
 		 HwCayleyTanh,
 		 HtCayleyZolo, // Plain old DWF.
 		 HmCayleyZolo,
-		 HwCayleyZolo
+		 HwCayleyZolo,
+		 HwPartFracZolo,
+		 HwContFracZolo,
+		 HwContFracTanh
+  });
+  std::vector<int> LsList({
+      8,//HtCayleyTanh, // Plain old DWF.
+      8,//HmCayleyTanh,
+      8,//HwCayleyTanh,
+      8,//HtCayleyZolo, // Plain old DWF.
+      8,//HmCayleyZolo,
+      8,//HwCayleyZolo,
+      9,//HwPartFracZolo
+      9, //HwContFracZolo
+      9 //HwContFracTanh
   });
   std::vector<std::string> ActionName({
         "HtCayleyTanh",
@@ -415,10 +479,19 @@ int main (int argc,char **argv )
 	"HwCayleyTanh",
 	"HtCayleyZolo",
 	"HmCayleyZolo",
-        "HwCayleyZolo"
+        "HwCayleyZolo",
+	"HwPartFracZolo",
+	"HwContFracZolo",
+	"HwContFracTanh"
   });
 
   for(int i=0;i<ActionList.size();i++) {
+    Ls = LsList[i];
+    Grid::GridCartesian      * FGrid   = Grid::SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
+    Grid::LatticeGaugeField lat(UGrid);
+    Grid::LatticeFermion    src(FGrid);
+    Grid::LatticeFermion    res_chroma(FGrid);
+    Grid::LatticeFermion    res_grid  (FGrid);
     std::cout << "*****************************"<<std::endl;
     std::cout << "Action "<<ActionName[i]<<std::endl;
     std::cout << "*****************************"<<std::endl;
@@ -439,6 +512,7 @@ int main (int argc,char **argv )
       
       std::cout << "Norm of difference "<<Grid::norm2(res_chroma)<<std::endl;
     }
+    delete FGrid;
   }
 
   std::cout << "Finished test "<<std::endl;
@@ -502,7 +576,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
   Grid::gaussian(RNG5,src);
   Grid::gaussian(RNG5,res);
 
-  Grid::SU<Nc>::HotConfiguration(RNG4,Umu);
+  Grid::SU<Grid::Nc>::HotConfiguration(RNG4,Umu);
 
   /*
   Grid::LatticeColourMatrix U(UGrid);
@@ -519,7 +593,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
   if ( action == HtCayleyTanh ) { 
 
-    Grid::DomainWallFermionR Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5);
+    Grid::DomainWallFermionD Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5);
 
     std::cout << Grid::GridLogMessage <<" Calling domain wall multiply "<<std::endl;
 
@@ -535,7 +609,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
     Grid::Real _b = 0.5*(mobius_scale +1.0);
     Grid::Real _c = 0.5*(mobius_scale -1.0);
-    Grid::MobiusZolotarevFermionR D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,_b,_c,zolo_lo,zolo_hi);
+    Grid::MobiusZolotarevFermionD D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,_b,_c,zolo_lo,zolo_hi);
 
     std::cout << Grid::GridLogMessage <<" Calling mobius zolo multiply "<<std::endl;
 
@@ -549,7 +623,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
   if ( action == HtCayleyZolo ) {
 
-    Grid::ShamirZolotarevFermionR D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
+    Grid::ShamirZolotarevFermionD D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
 
     std::cout << Grid::GridLogMessage <<" Calling shamir zolo multiply "<<std::endl;
 
@@ -561,6 +635,60 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
     return;
   }
 
+  if ( action == HwPartFracTanh ) {
+
+    Grid::OverlapWilsonPartialFractionTanhFermionD Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,1.0);
+
+    std::cout << Grid::GridLogMessage <<" Calling part frac tanh multiply "<<std::endl;
+
+    if ( dag ) 
+      Dov.Mdag(src,res);  
+    else 
+      Dov.M(src,res);  
+
+    return;
+  }
+
+  if ( action == HwContFracTanh ) {
+
+    Grid::OverlapWilsonContFracTanhFermionD Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,1.0);
+
+    std::cout << Grid::GridLogMessage <<" Calling cont frac tanh multiply "<<std::endl;
+
+    if ( dag ) 
+      Dov.Mdag(src,res);  
+    else 
+      Dov.M(src,res);  
+
+    return;
+  }
+  if ( action == HwContFracZolo ) {
+
+    Grid::OverlapWilsonContFracZolotarevFermionD Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
+
+    std::cout << Grid::GridLogMessage <<" Calling cont frac zolo multiply "<<std::endl;
+
+    if ( dag ) 
+      Dov.Mdag(src,res);  
+    else 
+      Dov.M(src,res);  
+
+    return;
+  }
+
+  if ( action == HwPartFracZolo ) {
+
+    Grid::OverlapWilsonPartialFractionZolotarevFermionD Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
+    std::cout << Grid::GridLogMessage <<" Calling part frac zolotarev multiply "<<std::endl;
+
+    if ( dag ) 
+      Dov.Mdag(src,res);  
+    else 
+      Dov.M(src,res);  
+
+    return;
+  }
+  
   /*
   if ( action == HmCayleyTanh ) {
     Grid::Real _b = 0.5*(mobius_scale +1.0);
@@ -581,7 +709,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
   if ( action == HmCayleyTanh ) {
 
-    Grid::ScaledShamirFermionR D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,mobius_scale);
+    Grid::ScaledShamirFermionD D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,mobius_scale);
 
     std::cout << Grid::GridLogMessage <<" Calling scaled shamir multiply "<<std::endl;
 
@@ -595,7 +723,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
   if ( action == HwCayleyTanh ) {
 
-    Grid::OverlapWilsonCayleyTanhFermionR D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,1.0);
+    Grid::OverlapWilsonCayleyTanhFermionD D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,1.0);
 
     if ( dag ) 
       D.Mdag(src,res);  
@@ -607,7 +735,7 @@ void calc_grid(ChromaAction action,Grid::LatticeGaugeField & Umu, Grid::LatticeF
 
   if ( action == HwCayleyZolo ) {
 
-    Grid::OverlapWilsonCayleyZolotarevFermionR D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
+    Grid::OverlapWilsonCayleyZolotarevFermionD D(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,_mass,_M5,zolo_lo,zolo_hi);
 
     if ( dag ) 
       D.Mdag(src,res);  

tests/solver/Test_dwf_cg_prec.cc

@@ -1,4 +1,4 @@
-/*************************************************************************************
+*************************************************************************************
 
 Grid physics library, www.github.com/paboyle/Grid
 
@@ -67,7 +67,13 @@ int main(int argc, char** argv) {
   result = Zero();
   LatticeGaugeField Umu(UGrid);
 
+#if 0
+  FieldMetaData header;
+  std::string file("ckpoint_lat.4000");
+  NerscIO::readConfiguration(Umu,header,file);
+#else  
   SU<Nc>::HotConfiguration(RNG4, Umu);
+#endif
 
   std::cout << GridLogMessage << "Lattice dimensions: " << GridDefaultLatt()
             << "   Ls: " << Ls << std::endl;

tests/solver/Test_dwf_cg_unprec.cc

@@ -54,15 +54,30 @@ int main (int argc, char ** argv)
   GridCartesian         * FGrid   = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
   GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
 
+  std::vector<ComplexD> qmu;
+  qmu.push_back(ComplexD(0.1,0.0));
+  qmu.push_back(ComplexD(0.0,0.0));
+  qmu.push_back(ComplexD(0.0,0.0));
+  qmu.push_back(ComplexD(0.0,0.01));
+  
+
   std::vector<int> seeds4({1,2,3,4});
   std::vector<int> seeds5({5,6,7,8});
   GridParallelRNG          RNG5(FGrid);  RNG5.SeedFixedIntegers(seeds5);
   GridParallelRNG          RNG4(UGrid);  RNG4.SeedFixedIntegers(seeds4);
 
+  LatticeFermion    tmp(FGrid);
   LatticeFermion    src(FGrid); random(RNG5,src);
   LatticeFermion result(FGrid); result=Zero();
-  LatticeGaugeField Umu(UGrid); SU<Nc>::HotConfiguration(RNG4,Umu);
-
+  LatticeGaugeField Umu(UGrid); 
+#if 0
+  FieldMetaData header;
+  std::string file("ckpoint_lat.4000");
+  NerscIO::readConfiguration(Umu,header,file);
+#else  
+  SU<Nc>::HotConfiguration(RNG4,Umu);
+#endif
+  
   std::vector<LatticeColourMatrix> U(4,UGrid);
   for(int mu=0;mu<Nd;mu++){
     U[mu] = PeekIndex<LorentzIndex>(Umu,mu);
@@ -71,8 +86,15 @@ int main (int argc, char ** argv)
   RealD mass=0.1;
   RealD M5=1.8;
   DomainWallFermionD Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
+  Ddwf.qmu = qmu;
 
+  Ddwf.M(src,tmp);
+  std::cout << " |M src|^2 "<<norm2(tmp)<<std::endl;
   MdagMLinearOperator<DomainWallFermionD,LatticeFermion> HermOp(Ddwf);
+  HermOp.HermOp(src,tmp);
+
+  std::cout << " <src|MdagM| src> "<<innerProduct(src,tmp)<<std::endl;
+  
   ConjugateGradient<LatticeFermion> CG(1.0e-6,10000);
   CG(HermOp,src,result);