portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-24 12:45:56 +01:00
Code Releases Activity

Checking in before pulling develop

Browse Source
This commit is contained in:
Chulwoo Jung 2020-10-01 15:35:03 -04:00
parent f91e3af97f
commit 8ee26f9112
2 changed files with 22 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
Grid/algorithms/iterative/ImplicitlyRestartedBlockLanczos.h
View File

@ -447,7 +447,7 @@ for( int i =0;i<total;i++){
cudaStat = cudaMallocManaged((void **)&w_acc, Nu*sites*12*sizeof(CUDA_COMPLEX)); cudaStat = cudaMallocManaged((void **)&w_acc, Nu*sites*12*sizeof(CUDA_COMPLEX));
// Glog << "w_acc= "<<w_acc << " "<< cudaStat << std::endl; // Glog << "w_acc= "<<w_acc << " "<< cudaStat << std::endl;
cudaStat = cudaMallocManaged((void **)&evec_acc, Nevec_acc*sites*12*sizeof(CUDA_COMPLEX)); cudaStat = cudaMallocManaged((void **)&evec_acc, Nevec_acc*sites*12*sizeof(CUDA_COMPLEX));
// Glog << "evec_acc= "<<evec_acc << " "<< cudaStat << std::endl; // Glog << "evec_acc= "<<evec_acc << " "<< cudaStat << std::endl;
cudaStat = cudaMallocManaged((void **)&c_acc, Nu*Nevec_acc*sizeof(CUDA_COMPLEX)); cudaStat = cudaMallocManaged((void **)&c_acc, Nu*Nevec_acc*sizeof(CUDA_COMPLEX));
// Glog << "c_acc= "<<c_acc << " "<< cudaStat << std::endl; // Glog << "c_acc= "<<c_acc << " "<< cudaStat << std::endl;
@ -721,7 +721,9 @@ for( int i =0;i<total;i++){
Eigen::MatrixXcd Q = Eigen::MatrixXcd::Zero(Nm,Nm); Eigen::MatrixXcd Q = Eigen::MatrixXcd::Zero(Nm,Nm);
std::vector<int> Iconv(Nm); std::vector<int> Iconv(Nm);
std::vector<Field> B(Nm,grid); // waste of space replicating int Ntest=Nu;
// std::vector<Field> B(Nm,grid); // waste of space replicating
std::vector<Field> B(1,grid); // waste of space replicating
std::vector<Field> f(Nu,grid); std::vector<Field> f(Nu,grid);
std::vector<Field> f_copy(Nu,grid); std::vector<Field> f_copy(Nu,grid);
@ -782,22 +784,22 @@ for( int i =0;i<total;i++){
// Convergence test // Convergence test
Glog <<" #Convergence test: "<<std::endl; Glog <<" #Convergence test: "<<std::endl;
Nconv = 0; Nconv = 0;
for(int k = 0; k<Nr; ++k) B[k]=0.0;
for(int j = 0; j<Nr; j+=Nconv_test_interval){ for(int j = 0; j<Nr; j+=Nconv_test_interval){
B[0]=0.0;
if ( j/Nconv_test_interval == Nconv ) { if ( j/Nconv_test_interval == Nconv ) {
Glog <<" #rotation for next check point evec" Glog <<" #rotation for next check point evec"
<< std::setw(4)<< std::setiosflags(std::ios_base::right) << std::setw(4)<< std::setiosflags(std::ios_base::right)
<< "["<< j <<"]" <<std::endl; << "["<< j <<"]" <<std::endl;
for(int k = 0; k<Nr; ++k){ for(int k = 0; k<Nr; ++k){
B[j].Checkerboard() = evec[k].Checkerboard(); B[0].Checkerboard() = evec[k].Checkerboard();
B[j] += evec[k]*Qt(k,j); B[0] += evec[k]*Qt(k,j);
} }
_Linop.HermOp(B[j],v); _Linop.HermOp(B[0],v);
RealD vnum = real(innerProduct(B[j],v)); // HermOp. RealD vnum = real(innerProduct(B[0],v)); // HermOp.
RealD vden = norm2(B[j]); RealD vden = norm2(B[0]);
eval2[j] = vnum/vden; eval2[j] = vnum/vden;
v -= eval2[j]*B[j]; v -= eval2[j]*B[0];
RealD vv = norm2(v); RealD vv = norm2(v);
resid[j] = vv; resid[j] = vv;
@ -836,11 +838,20 @@ for( int i =0;i<total;i++){
} else { } else {
Glog << fname + " CONVERGED ; Summary :\n"; Glog << fname + " CONVERGED ; Summary :\n";
// Sort convered eigenpairs. // Sort convered eigenpairs.
std::vector<Field> Btmp(Nconv,grid); // waste of space replicating
for(int i=0; i<Nconv; ++i) Btmp[i]=0;
for(int i=0; i<Nconv; ++i)
for(int k = 0; k<Nr; ++k){
Btmp[i].Checkerboard() = evec[k].Checkerboard();
Btmp[i] += evec[k]*Qt(k,Iconv[i]);
}
eval.resize(Nconv); eval.resize(Nconv);
evec.resize(Nconv,grid); evec.resize(Nconv,grid);
for(int i=0; i<Nconv; ++i){ for(int i=0; i<Nconv; ++i){
eval[i] = eval2[Iconv[i]]; eval[i] = eval2[Iconv[i]];
evec[i] = B[Iconv[i]]; evec[i] = Btmp[i];
} }
_sort.push(eval,evec,Nconv); _sort.push(eval,evec,Nconv);
} }

2
tests/lanczos/Test_dwf_block_lanczos.cc
View File

@ -359,7 +359,7 @@ int main (int argc, char ** argv)
JP.MaxIter, JP.MaxIter,
IRBLdiagonaliseWithEigen); IRBLdiagonaliseWithEigen);
// IRBLdiagonaliseWithLAPACK); // IRBLdiagonaliseWithLAPACK);
IRBL.split_test=1; IRBL.split_test=0;
std::vector<RealD> eval(JP.Nm); std::vector<RealD> eval(JP.Nm);