Removed the temporary variable hr_Gint_full_ when transitioning from 2D block parallelism to serial in Hcontainer (develop) (#6510)

* delete tem Hcontainer to reduce memory usage

* simplify the compute code

* change DM2D_tmp to dm2d_tmp, use vector instead of new

* delete hr_gint_full_ to reduce memory

* add serial code

* fix bug

* fix a not declare bug

* fix serial bug

* Fix the issue of not resetting to zero in the loop

* resolve confilct

* resolve confilct

* Change the names to comply with the conventions

Author: zgn-26714

source/source_estate/elecstate_lcao.cpp

@@ -35,7 +35,7 @@ void ElecStateLCAO<std::complex<double>>::psiToRho(const psi::Psi<std::complex<d
 
     ModuleBase::GlobalFunc::NOTE("Calculate the charge on real space grid!");
 #ifdef __OLD_GINT
-    this->gint_k->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer DM2D to DM_grid in gint
+    this->gint_k->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer dm2d to DM_grid in gint
     Gint_inout inout(this->charge->rho, Gint_Tools::job_type::rho, PARAM.inp.nspin);
     this->gint_k->cal_gint(&inout);
 #else
@@ -72,7 +72,7 @@ void ElecStateLCAO<double>::psiToRho(const psi::Psi<double>& psi)
     ModuleBase::GlobalFunc::NOTE("Calculate the charge on real space grid!");
 
 #ifdef __OLD_GINT 
-    this->gint_gamma->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer DM2D to DM_grid in gint
+    this->gint_gamma->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer dm2d to DM_grid in gint
     Gint_inout inout(this->charge->rho, Gint_Tools::job_type::rho, PARAM.inp.nspin);
     this->gint_gamma->cal_gint(&inout);
 #else
@@ -140,7 +140,7 @@ void ElecStateLCAO<double>::dmToRho(std::vector<double*> pexsi_DM, std::vector<d
 
     ModuleBase::GlobalFunc::NOTE("Calculate the charge on real space grid!");
 #ifdef __OLD_GINT
-    this->gint_gamma->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer DM2D to DM_grid in gint
+    this->gint_gamma->transfer_DM2DtoGrid(this->DM->get_DMR_vector()); // transfer dm2d to DM_grid in gint
     Gint_inout inout(this->charge->rho, Gint_Tools::job_type::rho, PARAM.inp.nspin);
     this->gint_gamma->cal_gint(&inout);
 #else

source/source_lcao/module_gint/gint.h

@@ -26,7 +26,7 @@ class Gint {
 
     hamilt::HContainer<double>* get_hRGint() const { return hRGint; }
 
-    std::vector<hamilt::HContainer<double>*> get_DMRGint() const { return DMRGint; }
+    std::vector<hamilt::HContainer<double>*> get_DMRGint() const { return dmr_gint; }
 
     int get_ncxyz() const { return ncxyz; }
 
@@ -58,14 +58,14 @@ class Gint {
     void initialize_pvpR(const UnitCell& unitcell, const Grid_Driver* gd, const int& nspin);
 
     /**
-     * @brief resize DMRGint to nspin and reallocate the memory
+     * @brief resize dmr_gint to nspin and reallocate the memory
      */
     void reset_DMRGint(const int& nspin);
 
     /**
      * @brief transfer DMR (2D para) to DMR (Grid para) in elecstate_lcao.cpp
      */
-    void transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> DM2D);
+    void transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> dm2d);
 
     const Grid_Technique* gridt = nullptr;
     const UnitCell* ucell;
@@ -256,13 +256,13 @@ class Gint {
     hamilt::HContainer<double>* hRGint = nullptr; 
 
     //! size of vec is 4, only used when nspin = 4
-    std::vector<hamilt::HContainer<double>*> hRGint_tmp; 
+    std::vector<hamilt::HContainer<double>*> hr_gint_tmp; 
 
     //! stores Hamiltonian in sparse format
     hamilt::HContainer<std::complex<double>>* hRGintCd = nullptr; 
 
     //! stores DMR in sparse format
-    std::vector<hamilt::HContainer<double>*> DMRGint; 
+    std::vector<hamilt::HContainer<double>*> dmr_gint; 
 
     //! tmp tools used in transfer_DM2DtoGrid 
     hamilt::HContainer<double>* dm2d_tmp = nullptr;

source/source_lcao/module_gint/gint_force_cpu_interface.cpp

@@ -82,7 +82,7 @@ void Gint::gint_kernel_force(Gint_inout* inout) {
                 cal_flag.get_ptr_2D(),
                 psir_vlbr3.get_ptr_2D(), 
                 psir_vlbr3_DM.get_ptr_2D(), 
-                this->DMRGint[inout->ispin], 
+                this->dmr_gint[inout->ispin], 
                 false);
 
         if(inout->isforce)
@@ -230,19 +230,19 @@ void Gint::gint_kernel_force_meta(Gint_inout* inout) {
 	//calculating g_mu(r) = sum_nu rho_mu,nu f_nu(r)
         Gint_Tools::mult_psi_DMR(*this->gridt, this->bxyz, LD_pool, grid_index, 
             na_grid, block_index.data(), block_size.data(), cal_flag.get_ptr_2D(),
-            psir_vlbr3.get_ptr_2D(), psir_vlbr3_DM.get_ptr_2D(), this->DMRGint[inout->ispin], false);
+            psir_vlbr3.get_ptr_2D(), psir_vlbr3_DM.get_ptr_2D(), this->dmr_gint[inout->ispin], false);
 
         Gint_Tools::mult_psi_DMR(*this->gridt, this->bxyz, LD_pool, grid_index, 
             na_grid, block_index.data(), block_size.data(), cal_flag.get_ptr_2D(),
-            dpsir_x_vlbr3.get_ptr_2D(), dpsirx_v_DM.get_ptr_2D(), this->DMRGint[inout->ispin], false);
+            dpsir_x_vlbr3.get_ptr_2D(), dpsirx_v_DM.get_ptr_2D(), this->dmr_gint[inout->ispin], false);
 
         Gint_Tools::mult_psi_DMR(*this->gridt, this->bxyz, LD_pool, grid_index,
             na_grid, block_index.data(), block_size.data(), cal_flag.get_ptr_2D(),
-            dpsir_y_vlbr3.get_ptr_2D(), dpsiry_v_DM.get_ptr_2D(), this->DMRGint[inout->ispin], false);
+            dpsir_y_vlbr3.get_ptr_2D(), dpsiry_v_DM.get_ptr_2D(), this->dmr_gint[inout->ispin], false);
 
         Gint_Tools::mult_psi_DMR(*this->gridt, this->bxyz, LD_pool, grid_index, 
             na_grid, block_index.data(), block_size.data(), cal_flag.get_ptr_2D(),
-            dpsir_z_vlbr3.get_ptr_2D(), dpsirz_v_DM.get_ptr_2D(), this->DMRGint[inout->ispin], false);
+            dpsir_z_vlbr3.get_ptr_2D(), dpsirz_v_DM.get_ptr_2D(), this->dmr_gint[inout->ispin], false);
 
         if(inout->isforce)
         {

source/source_lcao/module_gint/gint_gpu_interface.cpp

@@ -18,7 +18,7 @@ void Gint::gpu_vlocal_interface(Gint_inout* inout) {
         ylmcoef[i] = ModuleBase::Ylm::ylmcoef[i];
     }
 
-    hamilt::HContainer<double>* hRGint_kernel = PARAM.inp.nspin != 4 ? this->hRGint : this->hRGint_tmp[inout->ispin];
+    hamilt::HContainer<double>* hRGint_kernel = PARAM.inp.nspin != 4 ? this->hRGint : this->hr_gint_tmp[inout->ispin];
     GintKernel::gint_vl_gpu(hRGint_kernel,
                             inout->vl,
                             ylmcoef,
@@ -45,7 +45,7 @@ void Gint::gpu_rho_interface(Gint_inout* inout) {
     int nrxx = this->gridt->ncx * this->gridt->ncy * this->nplane;
     for (int is = 0; is < PARAM.inp.nspin; ++is) {
         ModuleBase::GlobalFunc::ZEROS(inout->rho[is], nrxx);
-        GintKernel::gint_rho_gpu(this->DMRGint[is],
+        GintKernel::gint_rho_gpu(this->dmr_gint[is],
                                        ylmcoef,
                                        dr,
                                        this->gridt->rcuts.data(),
@@ -76,7 +76,7 @@ void Gint::gpu_force_interface(Gint_inout* inout) {
     if (isforce || isstress) {
         std::vector<double> force(nat * 3, 0.0);
         std::vector<double> stress(6, 0.0);
-        GintKernel::gint_fvl_gpu(this->DMRGint[inout->ispin],
+        GintKernel::gint_fvl_gpu(this->dmr_gint[inout->ispin],
                                        inout->vl,
                                        force.data(),
                                        stress.data(),

source/source_lcao/module_gint/gint_k_pvpr.cpp

@@ -89,8 +89,8 @@ void Gint_k::transfer_pvpR(hamilt::HContainer<std::complex<double>>* hR,
         {
             hamilt::AtomPair<std::complex<double>>* upper_ap = ap;
             hamilt::AtomPair<std::complex<double>>* lower_ap = this->hRGintCd->find_pair(iat2, iat1);
-            const hamilt::AtomPair<double>* ap_nspin_0 = this->hRGint_tmp[0]->find_pair(iat1, iat2);
-            const hamilt::AtomPair<double>* ap_nspin_3 = this->hRGint_tmp[3]->find_pair(iat1, iat2);
+            const hamilt::AtomPair<double>* ap_nspin_0 = this->hr_gint_tmp[0]->find_pair(iat1, iat2);
+            const hamilt::AtomPair<double>* ap_nspin_3 = this->hr_gint_tmp[3]->find_pair(iat1, iat2);
             for (int ir = 0; ir < upper_ap->get_R_size(); ir++)
             {   
                 const auto R_index = upper_ap->get_R_index(ir);
@@ -110,8 +110,8 @@ void Gint_k::transfer_pvpR(hamilt::HContainer<std::complex<double>>* hR,
 
                 if (PARAM.globalv.domag)
                 {
-                    const hamilt::AtomPair<double>* ap_nspin_1 = this->hRGint_tmp[1]->find_pair(iat1, iat2);
-                    const hamilt::AtomPair<double>* ap_nspin_2 = this->hRGint_tmp[2]->find_pair(iat1, iat2);
+                    const hamilt::AtomPair<double>* ap_nspin_1 = this->hr_gint_tmp[1]->find_pair(iat1, iat2);
+                    const hamilt::AtomPair<double>* ap_nspin_2 = this->hr_gint_tmp[2]->find_pair(iat1, iat2);
                     const auto mat_nspin_1 = ap_nspin_1->find_matrix(R_index);
                     const auto mat_nspin_2 = ap_nspin_2->find_matrix(R_index);
                     for (int irow = 0; irow < mat_nspin_1->get_row_size(); ++irow)

source/source_lcao/module_gint/gint_old.cpp

@@ -24,13 +24,13 @@ Gint::~Gint() {
 
     delete this->hRGint;
     delete this->hRGintCd;
-    // in gamma_only case, DMRGint.size()=0, 
-    // in multi-k case, DMRGint.size()=nspin
-    for (int is = 0; is < this->DMRGint.size(); is++) {
-        delete this->DMRGint[is];
+    // in gamma_only case, dmr_gint.size()=0, 
+    // in multi-k case, dmr_gint.size()=nspin
+    for (int is = 0; is < this->dmr_gint.size(); is++) {
+        delete this->dmr_gint[is];
     }
-    for(int is = 0; is < this->hRGint_tmp.size(); is++) {
-        delete this->hRGint_tmp[is];
+    for(int is = 0; is < this->hr_gint_tmp .size(); is++) {
+        delete this->hr_gint_tmp [is];
     }
 #ifdef __MPI
     delete this->dm2d_tmp;
@@ -141,13 +141,12 @@ void Gint::prep_grid(const Grid_Technique& gt,
 void Gint::initialize_pvpR(const UnitCell& ucell_in, const Grid_Driver* gd, const int& nspin)
 {
     ModuleBase::TITLE("Gint", "initialize_pvpR");
-
     int npol = 1;
-    // there is the only resize code of DMRGint
-    if (this->DMRGint.size() == 0) {
-        this->DMRGint.resize(nspin);
+    // there is the only resize code of dmr_gint
+    if (this->dmr_gint.size() == 0) {
+        this->dmr_gint.resize(nspin);
     }
-    hRGint_tmp.resize(nspin);
+    hr_gint_tmp.resize(nspin);
     if (nspin != 4) {
         if (this->hRGint != nullptr) {
             delete this->hRGint;
@@ -161,22 +160,21 @@ void Gint::initialize_pvpR(const UnitCell& ucell_in, const Grid_Driver* gd, cons
         this->hRGintCd
             = new hamilt::HContainer<std::complex<double>>(ucell_in.nat);
         for (int is = 0; is < nspin; is++) {
-            if (this->DMRGint[is] != nullptr) {
-                delete this->DMRGint[is];
+            if (this->dmr_gint[is] != nullptr) {
+                delete this->dmr_gint[is];
             }
-            if (this->hRGint_tmp[is] != nullptr) {
-                delete this->hRGint_tmp[is];
+            if (this->hr_gint_tmp[is] != nullptr) {
+                delete this->hr_gint_tmp[is];
             }
-            this->DMRGint[is] = new hamilt::HContainer<double>(ucell_in.nat);
-            this->hRGint_tmp[is] = new hamilt::HContainer<double>(ucell_in.nat);
+            this->dmr_gint[is] = new hamilt::HContainer<double>(ucell_in.nat);
+            this->hr_gint_tmp[is] = new hamilt::HContainer<double>(ucell_in.nat);
         }
 #ifdef __MPI
         if (this->dm2d_tmp != nullptr) {
             delete this->dm2d_tmp;
         }
 #endif
     }
-
     if (PARAM.globalv.gamma_only_local && nspin != 4) {
         this->hRGint->fix_gamma();
     }
@@ -185,94 +183,100 @@ void Gint::initialize_pvpR(const UnitCell& ucell_in, const Grid_Driver* gd, cons
         this->hRGint->allocate(nullptr, true);
         ModuleBase::Memory::record("Gint::hRGint",
                             this->hRGint->get_memory_size());
-        // initialize DMRGint with hRGint when NSPIN != 4
-        for (int is = 0; is < this->DMRGint.size(); is++) {
-            if (this->DMRGint[is] != nullptr) {
-                delete this->DMRGint[is];
+        // initialize dmr_gint with hRGint when NSPIN != 4
+        for (int is = 0; is < this->dmr_gint.size(); is++) {
+            if (this->dmr_gint[is] != nullptr) {
+                delete this->dmr_gint[is];
             }
-            this->DMRGint[is] = new hamilt::HContainer<double>(*this->hRGint);
+            this->dmr_gint[is] = new hamilt::HContainer<double>(*this->hRGint);
         }
-        ModuleBase::Memory::record("Gint::DMRGint",
-                                   this->DMRGint[0]->get_memory_size()
-                                       * this->DMRGint.size());
+        ModuleBase::Memory::record("Gint::dmr_gint",
+                                   this->dmr_gint[0]->get_memory_size()
+                                       * this->dmr_gint.size());
     } else {
         this->hRGintCd->insert_ijrs(this->gridt->get_ijr_info(), ucell_in, npol);
         this->hRGintCd->allocate(nullptr, true);
         for(int is = 0; is < nspin; is++) {
-            this->hRGint_tmp[is]->insert_ijrs(this->gridt->get_ijr_info(), ucell_in);
-            this->DMRGint[is]->insert_ijrs(this->gridt->get_ijr_info(), ucell_in);
-            this->hRGint_tmp[is]->allocate(nullptr, true);
-            this->DMRGint[is]->allocate(nullptr, true);
+            this->hr_gint_tmp[is]->insert_ijrs(this->gridt->get_ijr_info(), ucell_in);
+            this->dmr_gint[is]->insert_ijrs(this->gridt->get_ijr_info(), ucell_in);
+            this->hr_gint_tmp[is]->allocate(nullptr, true);
+            this->dmr_gint[is]->allocate(nullptr, true);
         }
-        ModuleBase::Memory::record("Gint::hRGint_tmp",
-                                       this->hRGint_tmp[0]->get_memory_size()*nspin);
-        ModuleBase::Memory::record("Gint::DMRGint",
-                                       this->DMRGint[0]->get_memory_size()
-                                           * this->DMRGint.size()*nspin);
+        ModuleBase::Memory::record("Gint::hr_gint_tmp",
+                                       this->hr_gint_tmp[0]->get_memory_size()*nspin);
+        ModuleBase::Memory::record("Gint::dmr_gint",
+                                       this->dmr_gint[0]->get_memory_size()
+                                           * this->dmr_gint.size()*nspin);
     }
 }
 
 void Gint::reset_DMRGint(const int& nspin)
 {
     if (this->hRGint)
     {
-        for (auto& d : this->DMRGint) { delete d; }
-        this->DMRGint.resize(nspin);
-        this->DMRGint.shrink_to_fit();
-        for (auto& d : this->DMRGint) { d = new hamilt::HContainer<double>(*this->hRGint); }
+        for (auto& d : this->dmr_gint) { delete d; }
+        this->dmr_gint.resize(nspin);
+        this->dmr_gint.shrink_to_fit();
+        for (auto& d : this->dmr_gint) { d = new hamilt::HContainer<double>(*this->hRGint); }
         if (nspin == 4)
         {
-            for (auto& d : this->DMRGint) { d->allocate(nullptr, false); }
+            for (auto& d : this->dmr_gint) { d->allocate(nullptr, false); }
 #ifdef __MPI
             delete this->dm2d_tmp;
 #endif
         }
     }
 }
 
-void Gint::transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> DM2D) {
+void Gint::transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> dm2d) {
     ModuleBase::TITLE("Gint", "transfer_DMR");
-
-    // To check whether input parameter DM2D has been initialized
+    // To check whether input parameter dm2d has been initialized
 #ifdef __DEBUG
-    assert(!DM2D.empty()
-           && "Input parameter DM2D has not been initialized while calling "
+    assert(!dm2d.empty()
+           && "Input parameter dm2d has not been initialized while calling "
               "function transfer_DM2DtoGrid!");
 #endif
-
     ModuleBase::timer::tick("Gint", "transfer_DMR");
     if (PARAM.inp.nspin != 4) {
-        for (int is = 0; is < this->DMRGint.size(); is++) {
+        for (int is = 0; is < this->dmr_gint.size(); is++) {
 #ifdef __MPI
-            hamilt::transferParallels2Serials(*DM2D[is], DMRGint[is]);
+            hamilt::transferParallels2Serials(*dm2d[is], dmr_gint[is]);
 #else
-            this->DMRGint[is]->set_zero();
-            this->DMRGint[is]->add(*DM2D[is]);
+            this->dmr_gint[is]->set_zero();
+            this->dmr_gint[is]->add(*dm2d[is]);
 #endif
         }
     } else // NSPIN=4 case
     {
-#ifdef __MPI
         // is=0:↑↑, 1:↑↓, 2:↓↑, 3:↓↓
         const int row_set[4] = {0, 0, 1, 1};
         const int col_set[4] = {0, 1, 0, 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 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;
+        auto ijr_info = dm2d[0]->get_ijr_info();
+#ifdef __MPI
+        int blacs_ctxt = dm2d[0]->get_paraV()->blacs_ctxt;
         std::vector<int> iat2iwt(ucell->nat);
         for (int iat = 0; iat < ucell->nat; iat++) {
             iat2iwt[iat] = ucell->get_iat2iwt()[iat]/2;
         }
         Parallel_Orbitals pv{};
         pv.set(mg, ng, nb, blacs_ctxt);
         pv.set_atomic_trace(iat2iwt.data(), ucell->nat, mg);
-        auto ijr_info = DM2D[0]->get_ijr_info();
         this-> dm2d_tmp = new hamilt::HContainer<double>(&pv, nullptr, &ijr_info);
+#else
+        if (this->dm2d_tmp != nullptr) {
+            delete this->dm2d_tmp;
+        }
+        this-> dm2d_tmp = new hamilt::HContainer<double>(*this->hRGint);
+        this-> dm2d_tmp -> insert_ijrs(this->gridt->get_ijr_info(), *(this->ucell));
+        this-> dm2d_tmp -> allocate(nullptr, true);
+#endif
         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);
+            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) {
@@ -288,13 +292,15 @@ void Gint::transfer_DM2DtoGrid(std::vector<hamilt::HContainer<double>*> DM2D) {
                     }
                 }
             }
-            hamilt::transferParallels2Serials( *(this->dm2d_tmp), this->DMRGint[is]);
-        }
-        delete this->dm2d_tmp;
-        this->dm2d_tmp = nullptr;
+#ifdef __MPI
+            hamilt::transferParallels2Serials( *(this->dm2d_tmp), this->dmr_gint[is]);
 #else
-        //this->DMRGint_full = DM2D[0];
+            this->dmr_gint[is]->set_zero();
+            this->dmr_gint[is]->add(*(this->dm2d_tmp));
 #endif
+        }//is=4
+        delete this->dm2d_tmp;
+        this->dm2d_tmp = nullptr;
     }
     ModuleBase::timer::tick("Gint", "transfer_DMR");
 }