portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-03 18:55:56 +01:00
Code Releases Activity

Basis rotate stack passig to GPU reduction

Browse Source
This commit is contained in:
Peter Boyle 2020-04-30 12:31:07 -04:00
parent 5011753f4f
commit 9b2d2d0fc3
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Grid/algorithms/iterative/ImplicitlyRestartedLanczos.h
View File

@ -108,7 +108,6 @@ void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, i
} }
// Block the loop to keep storage footprint down // Block the loop to keep storage footprint down
vobj zz=Zero();
for(uint64_t s=0;s<oSites;s+=siteBlock){ for(uint64_t s=0;s<oSites;s+=siteBlock){
// remaining work in this block // remaining work in this block
@ -116,7 +115,8 @@ void basisRotate(std::vector<Field> &basis,Eigen::MatrixXd& Qt,int j0, int j1, i
// zero out the accumulators // zero out the accumulators
accelerator_for(ss,siteBlock*nrot,vobj::Nsimd(),{ accelerator_for(ss,siteBlock*nrot,vobj::Nsimd(),{
auto z=coalescedRead(zz); auto z=coalescedRead(Bp[ss]);
z=Zero();
coalescedWrite(Bp[ss],z); coalescedWrite(Bp[ss],z);
}); });
@ -158,12 +158,12 @@ void basisRotateJ(Field &result,std::vector<Field> &basis,Eigen::MatrixXd& Qt,in
for(int k=0;k<basis.size();k++){ for(int k=0;k<basis.size();k++){
basis_v[k] = basis[k].View(); basis_v[k] = basis[k].View();
} }
vobj zz=Zero();
Vector<double> Qt_jv(Nm); Vector<double> Qt_jv(Nm);
double * Qt_j = & Qt_jv[0]; double * Qt_j = & Qt_jv[0];
for(int k=0;k<Nm;++k) Qt_j[k]=Qt(j,k); for(int k=0;k<Nm;++k) Qt_j[k]=Qt(j,k);
accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{ accelerator_for(ss, grid->oSites(),vobj::Nsimd(),{
auto B=coalescedRead(zz); auto B=coalescedRead(basis_v[k0][ss]);
B=Zero();
for(int k=k0; k<k1; ++k){ for(int k=k0; k<k1; ++k){
B +=Qt_j[k] * coalescedRead(basis_v[k][ss]); B +=Qt_j[k] * coalescedRead(basis_v[k][ss]);
} }