Skip to content

Removed the temporary variable DMRGint_full when transitioning from 2D block parallelism to serial in Hcontainer(develop) #6489

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mohanchen merged 5 commits into from
Sep 11, 2025
Merged

Removed the temporary variable DMRGint_full when transitioning from 2D block parallelism to serial in Hcontainer(develop) #6489

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
7d4fe5a
delete tem Hcontainer to reduce memory usage
zgn-26714 Aug 21, 2025
e6a1983
simplify the compute code
zgn-26714 Sep 6, 2025
3e883d7
Merge branch 'develop' into dev-1
zgn-26714 Sep 9, 2025
6d2c711
change DM2D_tmp to dm2d_tmp, use vector instead of new
zgn-26714 Sep 11, 2025
1f3488e
Merge branch 'dev-1' of https://github.com/zgn-26714/abacus-develop i…
zgn-26714 Sep 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion source/source_lcao/module_gint/gint.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -265,7 +265,7 @@ class Gint {
std::vector<hamilt::HContainer<double>*> DMRGint;

//! tmp tools used in transfer_DM2DtoGrid
hamilt::HContainer<double>* DMRGint_full = nullptr;
hamilt::HContainer<double>* DM2D_tmp = nullptr;

std::vector<hamilt::HContainer<double>> pvdpRx_reduced;
std::vector<hamilt::HContainer<double>> pvdpRy_reduced;
Expand Down
100 changes: 60 additions & 40 deletions source/source_lcao/module_gint/gint_old.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ Gint::~Gint() {
delete this->hRGint_tmp[is];
}
#ifdef __MPI
delete this->DMRGint_full;
delete this->DM2D_tmp;
Comment thread
zgn-26714 marked this conversation as resolved.
Outdated
#endif
}

Expand Down Expand Up @@ -171,10 +171,9 @@ void Gint::initialize_pvpR(const UnitCell& ucell_in, const Grid_Driver* gd, cons
this->hRGint_tmp[is] = new hamilt::HContainer<double>(ucell_in.nat);
}
#ifdef __MPI
if (this->DMRGint_full != nullptr) {
delete this->DMRGint_full;
if (this->DM2D_tmp != nullptr) {
delete this->DM2D_tmp;
}
this->DMRGint_full = new hamilt::HContainer<double>(ucell_in.nat);
#endif
}

Expand Down Expand Up @@ -210,12 +209,6 @@ void Gint::initialize_pvpR(const UnitCell& ucell_in, const Grid_Driver* gd, cons
ModuleBase::Memory::record("Gint::DMRGint",
this->DMRGint[0]->get_memory_size()
* this->DMRGint.size()*nspin);
#ifdef __MPI
this->DMRGint_full->insert_ijrs(this->gridt->get_ijr_info(), ucell_in, npol);
this->DMRGint_full->allocate(nullptr, true);
ModuleBase::Memory::record("Gint::DMRGint_full",
this->DMRGint_full->get_memory_size());
#endif
}
}

Expand All @@ -231,9 +224,7 @@ void Gint::reset_DMRGint(const int& nspin)
{
for (auto& d : this->DMRGint) { d->allocate(nullptr, false); }
#ifdef __MPI
delete this->DMRGint_full;
this->DMRGint_full = new hamilt::HContainer<double>(*this->hRGint);
this->DMRGint_full->allocate(nullptr, false);
delete this->DM2D_tmp;
#endif
}
}
Expand Down Expand Up @@ -262,37 +253,66 @@ void Gint::transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> DM2D) {
} else // NSPIN=4 case
{
#ifdef __MPI
hamilt::transferParallels2Serials(*DM2D[0], this->DMRGint_full);
#else
this->DMRGint_full = DM2D[0];
#endif
std::vector<double*> tmp_pointer(4, nullptr);
for (int iap = 0; iap < this->DMRGint_full->size_atom_pairs(); ++iap) {
auto& ap = this->DMRGint_full->get_atom_pair(iap);
int iat1 = ap.get_atom_i();
int iat2 = ap.get_atom_j();
for (int ir = 0; ir < ap.get_R_size(); ++ir) {
const ModuleBase::Vector3<int> r_index = ap.get_R_index(ir);
for (int is = 0; is < 4; is++) {
tmp_pointer[is] = this->DMRGint[is]
->find_matrix(iat1, iat2, r_index)
->get_pointer();
}
double* data_full = ap.get_pointer(ir);
for (int irow = 0; irow < ap.get_row_size(); irow += 2) {
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer[0])++ = data_full[icol];
*(tmp_pointer[1])++ = data_full[icol + 1];
}
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer[2])++ = data_full[icol];
*(tmp_pointer[3])++ = data_full[icol + 1];
int mg = DM2D[0]->get_paraV()->get_global_row_size()/2;
int ng = DM2D[0]->get_paraV()->get_global_col_size()/2;
int nb = DM2D[0]->get_paraV()->get_block_size()/2;
int blacs_ctxt = DM2D[0]->get_paraV()->blacs_ctxt;
int *iat2iwt = new int[ucell->nat];
for (int iat = 0; iat < ucell->nat; iat++) {
iat2iwt[iat] = ucell->get_iat2iwt()[iat]/2;
}
Parallel_Orbitals *pv = new Parallel_Orbitals();
pv->set(mg, ng, nb, blacs_ctxt);
pv->set_atomic_trace(iat2iwt, ucell->nat, mg);
auto ijr_info = DM2D[0]->get_ijr_info();
this-> DM2D_tmp = new hamilt::HContainer<double>(pv, nullptr, &ijr_info);
ModuleBase::Memory::record("Gint::DM2D_tmp", this->DM2D_tmp->get_memory_size());
for (int is = 0; is < 4; is++){
for (int iap = 0; iap < DM2D[0]->size_atom_pairs(); ++iap) {
auto& ap = DM2D[0]->get_atom_pair(iap);
int iat1 = ap.get_atom_i();
int iat2 = ap.get_atom_j();
for (int ir = 0; ir < ap.get_R_size(); ++ir) {
const ModuleBase::Vector3<int> r_index = ap.get_R_index(ir);
double* tmp_pointer = this -> DM2D_tmp -> find_matrix(iat1, iat2, r_index)->get_pointer();
double* data_full = ap.get_pointer(ir);
for (int irow = 0; irow < ap.get_row_size(); irow += 2) {
switch (is) {//todo: It can be written more compactly
case 0:
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol];
}
data_full += ap.get_col_size() * 2;
break;
case 1:
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol + 1];
}
data_full += ap.get_col_size() * 2;
break;
case 2:
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol];
}
data_full += ap.get_col_size();
break;
case 3:
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol + 1];
}
data_full += ap.get_col_size();
break;
}
}
data_full += ap.get_col_size();
}
}
hamilt::transferParallels2Serials( *(this->DM2D_tmp), this->DMRGint[is]);
}
#else
//this->DMRGint_full = DM2D[0];
#endif
}
ModuleBase::timer::tick("Gint", "transfer_DMR");
}
91 changes: 57 additions & 34 deletions source/source_lcao/module_gint/temp_gint/gint_common.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,44 +163,67 @@ void transfer_dm_2d_to_gint(
} else // NSPIN=4 case
{
#ifdef __MPI
const int npol = 2;
HContainer<T> dm_full = gint_info.get_hr<T>(npol);
hamilt::transferParallels2Serials(*dm[0], &dm_full);
#else
HContainer<T>& dm_full = *(dm[0]);
#endif
std::vector<T*> tmp_pointer(4, nullptr);
for (int iap = 0; iap < dm_full.size_atom_pairs(); iap++)
{
auto& ap = dm_full.get_atom_pair(iap);
const int iat1 = ap.get_atom_i();
const int iat2 = ap.get_atom_j();
for (int ir = 0; ir < ap.get_R_size(); ir++)
{
const ModuleBase::Vector3<int> r_index = ap.get_R_index(ir);
for (int is = 0; is < 4; is++)
{
tmp_pointer[is] =
dm_gint[is].find_matrix(iat1, iat2, r_index)->get_pointer();
}
T* data_full = ap.get_pointer(ir);
for (int irow = 0; irow < ap.get_row_size(); irow += 2)
{
for (int icol = 0; icol < ap.get_col_size(); icol += 2)
{
*(tmp_pointer[0])++ = data_full[icol];
*(tmp_pointer[1])++ = data_full[icol + 1];
}
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2)
{
*(tmp_pointer[2])++ = data_full[icol];
*(tmp_pointer[3])++ = data_full[icol + 1];
int mg = dm[0]->get_paraV()->get_global_row_size()/2;
int ng = dm[0]->get_paraV()->get_global_col_size()/2;
int nb = dm[0]->get_paraV()->get_block_size()/2;
int blacs_ctxt = dm[0]->get_paraV()->blacs_ctxt;
const UnitCell* ucell = gint_info.get_ucell();
int *iat2iwt = new int[ucell->nat];
Comment thread
zgn-26714 marked this conversation as resolved.
Outdated
for (int iat = 0; iat < ucell->nat; iat++) {
iat2iwt[iat] = ucell->get_iat2iwt()[iat]/2;
}
Parallel_Orbitals *pv = new Parallel_Orbitals();
pv->set(mg, ng, nb, blacs_ctxt);
pv->set_atomic_trace(iat2iwt, ucell->nat, mg);
auto ijr_info = dm[0]->get_ijr_info();
HContainer<T>* DM2D_tmp = new hamilt::HContainer<T>(pv, nullptr, &ijr_info);
//ModuleBase::Memory::record("Gint::DM2D_tmp", this->DM2D_tmp->get_memory_size());
for (int is = 0; is < 4; is++){
for (int iap = 0; iap < dm[0]->size_atom_pairs(); ++iap) {
auto& ap = dm[0]->get_atom_pair(iap);
int iat1 = ap.get_atom_i();
int iat2 = ap.get_atom_j();
for (int ir = 0; ir < ap.get_R_size(); ++ir) {
const ModuleBase::Vector3<int> r_index = ap.get_R_index(ir);
T* tmp_pointer = DM2D_tmp -> find_matrix(iat1, iat2, r_index)->get_pointer();
T* data_full = ap.get_pointer(ir);
for (int irow = 0; irow < ap.get_row_size(); irow += 2) {
switch (is) {//todo: It can be written more compactly
case 0:
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol];
}
data_full += ap.get_col_size() * 2;
break;
case 1:
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol + 1];
}
data_full += ap.get_col_size() * 2;
break;
case 2:
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol];
}
data_full += ap.get_col_size();
break;
case 3:
data_full += ap.get_col_size();
for (int icol = 0; icol < ap.get_col_size(); icol += 2) {
*(tmp_pointer)++ = data_full[icol + 1];
}
data_full += ap.get_col_size();
break;
}
}
data_full += ap.get_col_size();
}
}
hamilt::transferParallels2Serials( *DM2D_tmp, &dm_gint[is]);
}
#else
//HContainer<T>& dm_full = *(dm[0]);
#endif
}
ModuleBase::timer::tick("Gint", "transfer_dm_2d_to_gint");
}
Expand Down
1 change: 1 addition & 0 deletions source/source_lcao/module_gint/temp_gint/gint_info.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ class GintInfo
const std::vector<int>& get_trace_lo() const{ return trace_lo_; }
int get_lgd() const { return lgd_; }
int get_nat() const { return ucell_->nat; } // return the number of atoms in the unitcell
const UnitCell* get_ucell() const { return ucell_; }
int get_local_mgrid_num() const { return localcell_info_->get_mgrids_num(); }
double get_mgrid_volume() const { return meshgrid_info_->get_volume(); }

Expand Down
4 changes: 2 additions & 2 deletions source/source_lcao/module_lr/utils/gint_move.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,8 +60,8 @@ Gint& Gint::operator=(Gint&& rhs)
this->pvdpRz_reduced = std::move(rhs.pvdpRz_reduced);
this->DMRGint = std::move(rhs.DMRGint);
this->hRGint_tmp = std::move(rhs.hRGint_tmp);
this->DMRGint_full = rhs.DMRGint_full;
rhs.DMRGint_full = nullptr;
this->DM2D_tmp = rhs.DM2D_tmp;
rhs.DM2D_tmp = nullptr;

return *this;
}
Expand Down
Loading