Merge 32e6d58356 into b461184797

Grid/algorithms/iterative/ImplicitlyRestartedBlockLanczosCoarse.h

@@ -279,11 +279,11 @@ public:
       Qt = Eigen::MatrixXcd::Identity(Nm,Nm);
       diagonalize(eval2,lmd2,lme2,Nu,Nm,Nm,Qt,grid);
       _sort.push(eval2,Nm);
-      //      Glog << "#Ritz value before shift: "<< std::endl;
+      Glog << "#Ritz value before shift: "<< std::endl;
       for(int i=0; i<Nm; ++i){
-	//        std::cout.precision(13);
+	std::cout.precision(13);
-	//        std::cout << "[" << std::setw(4)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
+	std::cout << "[" << std::setw(4)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
-	//        std::cout << "Rval = "<<std::setw(20)<< std::setiosflags(std::ios_base::left)<< eval2[i] << std::endl;
+	std::cout << "Rval = "<<std::setw(20)<< std::setiosflags(std::ios_base::left)<< eval2[i] << std::endl;
       }
       
       //----------------------------------------------------------------------
@@ -298,6 +298,7 @@ public:
         unpackHermitBlockTriDiagMatToEigen(lmd,lme,Nu,Nblock_m,Nm,Nm,BTDM);
 
         for(int ip=Nk; ip<Nm; ++ip){
+	  Glog << " ip "<<ip<<" / "<<Nm<<std::endl;
           shiftedQRDecompEigen(BTDM,Nu,Nm,eval2[ip],Q);
         }
         
@@ -325,7 +326,7 @@ public:
         Qt = Eigen::MatrixXcd::Identity(Nm,Nm);
         diagonalize(eval2,lmd2,lme2,Nu,Nk,Nm,Qt,grid);
         _sort.push(eval2,Nk);
-	//        Glog << "#Ritz value after shift: "<< std::endl;
+	Glog << "#Ritz value after shift: "<< std::endl;
         for(int i=0; i<Nk; ++i){
 	  //          std::cout.precision(13);
 	  //          std::cout << "[" << std::setw(4)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
@@ -467,10 +468,10 @@ public:
   
     // set initial vector
     for (int i=0; i<Nu; ++i) {
-      //      Glog << "norm2(src[" << i << "])= "<< norm2(src[i]) << std::endl;
+      Glog << "norm2(src[" << i << "])= "<< norm2(src[i]) << std::endl;
       evec[i] = src[i];
       orthogonalize(evec[i],evec,i);
-      //      Glog << "norm2(evec[" << i << "])= "<< norm2(evec[i]) << std::endl;
+      Glog << "norm2(evec[" << i << "])= "<< norm2(evec[i]) << std::endl;
     }
 //    exit(-43);
     
@@ -506,11 +507,11 @@ public:
       Qt = Eigen::MatrixXcd::Identity(Nr,Nr);
       diagonalize(eval2,lmd2,lme2,Nu,Nr,Nr,Qt,grid);
       _sort.push(eval2,Nr);
-      //      Glog << "#Ritz value: "<< std::endl;
+      Glog << "#Ritz value: "<< std::endl;
       for(int i=0; i<Nr; ++i){
-	//        std::cout.precision(13);
+	std::cout.precision(13);
-	//        std::cout << "[" << std::setw(4)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
+	std::cout << "[" << std::setw(4)<< std::setiosflags(std::ios_base::right) <<i<<"] ";
-	//        std::cout << "Rval = "<<std::setw(20)<< std::setiosflags(std::ios_base::left)<< eval2[i] << std::endl;
+	std::cout << "Rval = "<<std::setw(20)<< std::setiosflags(std::ios_base::left)<< eval2[i] << std::endl;
       }
       
       // Convergence test
@@ -570,6 +571,7 @@ public:
       Glog << fname + " NOT converged ; Summary :\n";
     } else {
       Glog << fname + " CONVERGED ; Summary :\n";
+      Nstop = Nconv_guess; // Just take them all
       // Sort convered eigenpairs.
       std::vector<Field>  Btmp(Nstop,grid); // waste of space replicating
 
@@ -642,7 +644,7 @@ private:
       //      for (int u=0; u<mrhs; ++u) Glog << " out["<<u<<"] = "<<norm2(out[u])<<std::endl;
       k_start +=mrhs;
     }
-    //    Glog << "LinAlg "<< std::endl;
+    Glog << "LinAlg "<< std::endl;
     
     if (b>0) {
       for (int u=0; u<Nu; ++u) {
@@ -676,7 +678,7 @@ private:
       }
       w_copy[u] = w[u];
     }
-    //    Glog << "LinAlg done"<< std::endl;
+    Glog << "LinAlg done"<< std::endl;
     
     // In block version, the steps 6 and 7 in Lanczos construction is
     // replaced by the QR decomposition of new basis block.
@@ -689,15 +691,15 @@ private:
     }
 
     // re-orthogonalization for numerical stability
-    //    Glog << "Gram Schmidt"<< std::endl;
+    Glog << "Gram Schmidt"<< std::endl;
     orthogonalize(w,Nu,evec,R);
     // QR part
     for (int u=1; u<Nu; ++u) {
       orthogonalize(w[u],w,u);
     }
-    //    Glog << "Gram Schmidt done "<< std::endl;
+    Glog << "Gram Schmidt done "<< std::endl;
     
-    //    Glog << "LinAlg "<< std::endl;
+    Glog << "LinAlg "<< std::endl;
     for (int u=0; u<Nu; ++u) {
       //for (int v=0; v<Nu; ++v) {
       for (int v=u; v<Nu; ++v) {
@@ -714,7 +716,7 @@ private:
 	//        Glog <<" In block "<< b << "," <<" beta[" << u << "," << k-L << "] = " << lme[u][k] << std::endl;
       }
     }
-    //    Glog << "LinAlg done "<< std::endl;
+    Glog << "LinAlg done "<< std::endl;
 
     if (b < Nm/Nu-1) {
       for (int u=0; u<Nu; ++u) {
@@ -779,7 +781,7 @@ private:
     
     for ( int u=0; u<Nu; ++u ) {
       for (int k=0; k<Nk; ++k ) {
-//        Glog << "lmd "<<u<<" "<<k<<" "<<lmd[u][k] -conjugate(lmd[u][k])<<std::endl;
+	//	Glog << "lmd "<<u<<" "<<k<<" "<<lmd[u][k] -conjugate(lmd[u][k])<<std::endl;
         BlockTriDiag(k,u+(k/Nu)*Nu) = lmd[u][k];
       }
     }
@@ -933,7 +935,7 @@ if (1){
          int Nu, int Nb, int Nk, int Nm,
          Eigen::MatrixXcd& M)
   {
-    //Glog << "unpackHermitBlockTriDiagMatToEigen() begin" << '\n'; 
+    Glog << "unpackHermitBlockTriDiagMatToEigen() begin" << '\n'; 
     assert( Nk%Nu == 0 && Nm%Nu == 0 );
     assert( Nk <= Nm );
     M = Eigen::MatrixXcd::Zero(Nk,Nk);
@@ -951,7 +953,7 @@ if (1){
         M(u+(k/Nu)*Nu,k-Nu) = lme[u][k-Nu];
       }
     }
-    //Glog << "unpackHermitBlockTriDiagMatToEigen() end" << endl; 
+    Glog << "unpackHermitBlockTriDiagMatToEigen() end" << std::endl; 
   }
  
 
@@ -961,7 +963,7 @@ if (1){
          int Nu, int Nb, int Nk, int Nm,
          Eigen::MatrixXcd& M)
   {
-    //Glog << "packHermitBlockTriDiagMatfromEigen() begin" << '\n'; 
+    Glog << "packHermitBlockTriDiagMatfromEigen() begin" << '\n'; 
     assert( Nk%Nu == 0 && Nm%Nu == 0 );
     assert( Nk <= Nm );
     
@@ -977,7 +979,7 @@ if (1){
         lme[u][k-Nu] = M(u+(k/Nu)*Nu,k-Nu);
       }
     }
-    //Glog << "packHermitBlockTriDiagMatfromEigen() end" << endl; 
+    Glog << "packHermitBlockTriDiagMatfromEigen() end" <<std::endl; 
   }
 
 
@@ -986,7 +988,7 @@ if (1){
 		            RealD Dsh,
 		            Eigen::MatrixXcd& Qprod)
   {
-    //Glog << "shiftedQRDecompEigen() begin" << '\n'; 
+    Glog << "shiftedQRDecompEigen() begin" << '\n'; 
     Eigen::MatrixXcd Q = Eigen::MatrixXcd::Zero(Nm,Nm);
     Eigen::MatrixXcd R = Eigen::MatrixXcd::Zero(Nm,Nm);
     Eigen::MatrixXcd Mtmp = Eigen::MatrixXcd::Zero(Nm,Nm);
@@ -1002,6 +1004,7 @@ if (1){
                         // lower triangular part used to represent series
                         // of Q sequence.
 
+    Glog << "shiftedQRDecompEigen() Housholder & QR" << '\n'; 
     // equivalent operation of Qprod *= Q
     //M = Eigen::MatrixXcd::Zero(Nm,Nm);
     
@@ -1022,6 +1025,7 @@ if (1){
     
     Mtmp = Eigen::MatrixXcd::Zero(Nm,Nm);
 
+    Glog << "shiftedQRDecompEigen() Mtmp create" << '\n'; 
     for (int i=0; i<Nm; ++i) {
       for (int j=0; j<Nm-(Nu+1); ++j) {
         for (int k=0; k<Nu+1+j; ++k) {
@@ -1029,6 +1033,7 @@ if (1){
         }
       }
     }
+    Glog << "shiftedQRDecompEigen() Mtmp loop1" << '\n'; 
     for (int i=0; i<Nm; ++i) {
       for (int j=Nm-(Nu+1); j<Nm; ++j) {
         for (int k=0; k<Nm; ++k) {
@@ -1036,6 +1041,7 @@ if (1){
         }
       }
     }
+    Glog << "shiftedQRDecompEigen() Mtmp loop2" << '\n'; 
     
     //static int ntimes = 2;
     //for (int j=0; j<Nm-(ntimes*Nu); ++j) {
@@ -1061,11 +1067,13 @@ if (1){
         Mtmp(j,i) = conj(Mtmp(i,j));
       }
     }
+    Glog << "shiftedQRDecompEigen() Mtmp loop3" << '\n'; 
 
     for (int i=0; i<Nm; ++i) {
       Mtmp(i,i) = real(Mtmp(i,i)) + Dsh;
     }
     
+    Glog << "shiftedQRDecompEigen() Mtmp loop4" << '\n'; 
     M = Mtmp;
 
     //M = Q.adjoint()*(M*Q);
@@ -1077,7 +1085,7 @@ if (1){
     //  }
     //}
     
-    //Glog << "shiftedQRDecompEigen() end" << endl; 
+    Glog << "shiftedQRDecompEigen() end" <<std::endl; 
   }
 
   void exampleQRDecompEigen(void)

Grid/qcd/action/pseudofermion/EvenOddSchurDifferentiable.h

@@ -86,8 +86,13 @@ public:
     assert(ForceE.Checkerboard()==Even);
     assert(ForceO.Checkerboard()==Odd);
 
+#if defined(GRID_CUDA) || defined(GRID_HIP)  || defined(GRID_SYCL)
+    acceleratorSetCheckerboard(Force,ForceE);
+    acceleratorSetCheckerboard(Force,ForceO);
+#else
     setCheckerboard(Force,ForceE); 
     setCheckerboard(Force,ForceO);
+#endif
     Force=-Force;
 
     delete forcecb;
@@ -130,8 +135,13 @@ public:
     assert(ForceE.Checkerboard()==Even);
     assert(ForceO.Checkerboard()==Odd);
 
+#if defined(GRID_CUDA) || defined(GRID_HIP)  || defined(GRID_SYCL)
+    acceleratorSetCheckerboard(Force,ForceE);
+    acceleratorSetCheckerboard(Force,ForceO);
+#else
     setCheckerboard(Force,ForceE); 
     setCheckerboard(Force,ForceO);
+#endif
     Force=-Force;
 
     delete forcecb;

systems/Frontier/sourceme.sh

@@ -3,7 +3,7 @@ spack load c-lime
 module load emacs 
 module load PrgEnv-gnu
 module load rocm
-module load cray-mpich/8.1.23
+module load cray-mpich
 module load gmp
 module load cray-fftw
 module load craype-accel-amd-gfx90a

tests/debug/Test_8888.cc

@@ -0,0 +1,118 @@
+/*************************************************************************************
+
+    Grid physics library, www.github.com/paboyle/Grid 
+
+    Source file: ./tests/Test_general_coarse_hdcg.cc
+
+    Copyright (C) 2023
+
+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/algorithms/iterative/ImplicitlyRestartedBlockLanczos.h>
+#include <Grid/algorithms/iterative/ImplicitlyRestartedBlockLanczosCoarse.h>
+#include <Grid/algorithms/iterative/AdefMrhs.h>
+
+using namespace std;
+using namespace Grid;
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  const int Ls=8;
+  const int nbasis = 40;
+  const int cb = 0 ;
+  RealD mass=0.01;
+  RealD M5=1.8;
+  RealD b=1.0;
+  RealD c=0.0;
+
+  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);
+
+  ///////////////////////// RNGs /////////////////////////////////
+  std::vector<int> seeds4({1,2,3,4});
+  std::vector<int> seeds5({5,6,7,8});
+  std::vector<int> cseeds({5,6,7,8});
+
+  GridParallelRNG          RNG5(FGrid);   RNG5.SeedFixedIntegers(seeds5);
+  GridParallelRNG          RNG4(UGrid);   RNG4.SeedFixedIntegers(seeds4);
+
+  ///////////////////////// Configuration /////////////////////////////////
+  LatticeGaugeField Umu(UGrid);
+
+  FieldMetaData header;
+  std::string file("ckpoint_EODWF_lat.125");
+  NerscIO::readConfiguration(Umu,header,file);
+
+  //////////////////////// Fermion action //////////////////////////////////
+  MobiusFermionD Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c);
+
+  MdagMLinearOperator<MobiusFermionD, LatticeFermion> HermOp(Ddwf);
+
+  
+  std::cout << "**************************************"<<std::endl;
+  std::cout << "         Fine Power method            "<<std::endl;
+  std::cout << "**************************************"<<std::endl;
+
+  LatticeFermionD pm_src(FGrid);
+  pm_src = ComplexD(1.0);
+  PowerMethod<LatticeFermionD>       fPM;
+  fPM(HermOp,pm_src);
+
+  
+  std::cout << "**************************************"<<std::endl;
+  std::cout << "         Fine Lanczos  (poly, low)    "<<std::endl;
+  std::cout << "**************************************"<<std::endl;
+  
+  int Nk=80;
+  int Nm=Nk*3;
+  int Nstop=8;
+  int Nconv_test_interval=1;
+  
+  //  Chebyshev<LatticeFermionD>      IRLChebyLo(0.2,64.0,201);  // 1 iter
+  Chebyshev<LatticeFermionD>      IRLChebyLo(0.0,55.0,101);  // 1 iter
+  FunctionHermOp<LatticeFermionD>    PolyOp(IRLChebyLo,HermOp);
+  PlainHermOp<LatticeFermionD>          Op(HermOp);
+
+  ImplicitlyRestartedLanczos IRL(PolyOp,
+				 Op,
+				 Nk, // sought vecs
+				 Nk, // sought vecs
+				 Nm, // spare vecs
+				 1.0e-8,
+				 10 // Max iterations
+				 );
+
+  int Nconv;
+  std::vector<RealD>            eval(Nm);
+  std::vector<LatticeFermionD>     evec(Nm,FGrid);
+  LatticeFermionD     irl_src(FGrid);
+
+  IRL.calc(eval,evec,irl_src,Nconv);
+
+  Grid_finalize();
+  return 0;
+}

tests/debug/Test_general_coarse.cc

@@ -244,7 +244,7 @@ int main (int argc, char ** argv)
 
   GridCartesian *CoarseMrhs = new GridCartesian(rhLatt,rhSimd,rhMpi); 
 
-  
+#if 0  
   MultiGeneralCoarsenedMatrix mrhs(LittleDiracOp,CoarseMrhs);
   typedef decltype(mrhs) MultiGeneralCoarsenedMatrix_t;
   
@@ -308,6 +308,7 @@ int main (int argc, char ** argv)
     mrhsCG(MrhsCoarseOp,rh_src,rh_res);
   }
 
+#endif
   std::cout<<GridLogMessage<<std::endl;
   std::cout<<GridLogMessage<<std::endl;
   std::cout<<GridLogMessage<<"*******************************************"<<std::endl;

tests/debug/Test_general_coarse_hdcg_phys48_mixed.cc

@@ -145,7 +145,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   const int Ls=24;
-  const int nbasis = 60;
+  const int nbasis = 62;
   const int cb = 0 ;
   RealD mass=0.00078;
   RealD M5=1.8;
@@ -160,7 +160,7 @@ int main (int argc, char ** argv)
   GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
 
   // Construct a coarsened grid with 4^4 cell
-  Coordinate Block({4,4,4,4});
+  Coordinate Block({4,4,6,4});
   Coordinate clatt = GridDefaultLatt();
   for(int d=0;d<clatt.size();d++){
     clatt[d] = clatt[d]/Block[d];

tests/debug/Test_general_coarse_hdcg_phys96_mixed.cc

@@ -160,7 +160,8 @@ int main (int argc, char ** argv)
   GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
 
   // Construct a coarsened grid with 4^4 cell
-  Coordinate Block({4,4,6,6});
+  //  Coordinate Block({4,4,6,4});
+  Coordinate Block({4,4,4,4});
   Coordinate clatt = GridDefaultLatt();
   for(int d=0;d<clatt.size();d++){
     clatt[d] = clatt[d]/Block[d];
@@ -217,7 +218,7 @@ int main (int argc, char ** argv)
   std::string evec_file("/lustre/orion/phy157/proj-shared/phy157_dwf/paboyle/evecs.scidac");
   std::string eval_file("/lustre/orion/phy157/proj-shared/phy157_dwf/paboyle/eval.xml");
   bool load_agg=true;
-  bool load_refine=false;
+  bool load_refine=true;
   bool load_mat=false;
   bool load_evec=false;
 
@@ -276,17 +277,25 @@ int main (int argc, char ** argv)
   std::cout << "**************************************"<<std::endl;
 
   typedef HermitianLinearOperator<MultiGeneralCoarsenedMatrix_t,CoarseVector> MrhsHermMatrix;
-  Chebyshev<CoarseVector>      IRLCheby(0.0012,42.0,301);  // 1 iter
+  //  Chebyshev<CoarseVector>      IRLCheby(0.0012,42.0,301);  // 4.4.6.4
+  //  Chebyshev<CoarseVector>      IRLCheby(0.0012,42.0,501);  // for 4.4.4.4 blocking 350 evs
+  //  Chebyshev<CoarseVector>      IRLCheby(0.0014,42.0,501);  // for 4.4.4.4 blocking 700 evs
+  //  Chebyshev<CoarseVector>      IRLCheby(0.002,42.0,501);  // for 4.4.4.4 blocking 1226 evs
+  //  Chebyshev<CoarseVector>      IRLCheby(0.0025,42.0,501);  // for 4.4.4.4 blocking 1059 evs
+							  //							  3e-4,2);
+  Chebyshev<CoarseVector>      IRLCheby(0.0018,42.0,301);  // for 4.4.4.4 blocking  // 790 evs
+  
   MrhsHermMatrix MrhsCoarseOp     (mrhs);
 
   CoarseVector pm_src(CoarseMrhs);
   pm_src = ComplexD(1.0);
   PowerMethod<CoarseVector>       cPM;   cPM(MrhsCoarseOp,pm_src);
 
-  int Nk=nrhs*30;
+  //  int Nk=nrhs*30; // 4.4.6.4
   //  int Nk=nrhs*80;
-  int Nm=Nk*4;
+  int Nk=nrhs*60; // 720
-  int Nstop=Nk;
+  int Nm=Nk*4;    // 2880 ; generally finishes at 1440
+  int Nstop=512;
   int Nconv_test_interval=1;
   
   ImplicitlyRestartedBlockLanczosCoarse<CoarseVector> IRL(MrhsCoarseOp,
@@ -299,7 +308,7 @@ int main (int argc, char ** argv)
 							  nrhs,
 							  Nk,
 							  Nm,
-							  1e-4,20);
+							  3e-4,2);
 
   std::vector<RealD>            eval(Nm);
   std::vector<CoarseVector>     evec(Nm,Coarse5d);
@@ -331,7 +340,7 @@ int main (int argc, char ** argv)
   // Extra HDCG parameters
   //////////////////////////
   int maxit=3000;
-  ConjugateGradient<CoarseVector>  CG(5.0e-2,maxit,false);
+  ConjugateGradient<CoarseVector>  CG(7.5e-2,maxit,false);
   RealD lo=2.0;
   int ord = 7;