delete hr_gint_full_ to reduce memory

Author: zgn-26714

source/source_lcao/module_gint/temp_gint/gint_common.cpp

@@ -47,95 +47,149 @@ void compose_hr_gint(HContainer<double>& hr_gint)
     ModuleBase::timer::tick("Gint", "compose_hr_gint");
 }
 
-void compose_hr_gint(const std::vector<HContainer<double>>& hr_gint_part,
-                     HContainer<std::complex<double>>& hr_gint_full)
+template <typename T>
+void transfer_hr_gint_to_hR(const HContainer<T>& hr_gint, HContainer<T>& hR)
 {
-    ModuleBase::TITLE("Gint", "compose_hr_gint");
-    ModuleBase::timer::tick("Gint", "compose_hr_gint");
-    for (int iap = 0; iap < hr_gint_full.size_atom_pairs(); iap++)
+    ModuleBase::TITLE("Gint", "transfer_hr_gint_to_hR");
+    ModuleBase::timer::tick("Gint", "transfer_hr_gint_to_hR");
+#ifdef __MPI
+    int size = 0;
+    MPI_Comm_size(MPI_COMM_WORLD, &size);
+    if (size == 1)
+    {
+        hR.add(hr_gint);
+    }
+    else
     {
-        auto* ap = &(hr_gint_full.get_atom_pair(iap));
-        const int iat1 = ap->get_atom_i();
-        const int iat2 = ap->get_atom_j();
-        if (iat1 <= iat2)
+        hamilt::transferSerials2Parallels(hr_gint, &hR);
+    }
+#else
+    hR.add(hr_gint);
+#endif
+    ModuleBase::timer::tick("Gint", "transfer_hr_gint_to_hR");
+}
+
+
+void merge_hr_part_to_hR(const std::vector<hamilt::HContainer<double>>& hRGint_tmp,
+                         hamilt::HContainer<std::complex<double>>* hR,
+                         const GintInfo& gint_info){
+    ModuleBase::TITLE("Gint_k", "transfer_pvpR");
+    ModuleBase::timer::tick("Gint_k", "transfer_pvpR");
+
+    int mg = hR->get_paraV()->get_global_row_size()/2;
+    int ng = hR->get_paraV()->get_global_col_size()/2;
+    int nb = hR->get_paraV()->get_block_size()/2;
+    const UnitCell* ucell_in = gint_info.get_ucell();
+#ifdef __MPI
+    int blacs_ctxt = hR->get_paraV()->blacs_ctxt;
+    std::vector<int> iat2iwt(ucell_in->nat);
+    for (int iat = 0; iat < ucell_in->nat; iat++) {
+        iat2iwt[iat] = ucell_in->get_iat2iwt()[iat]/2;
+    }
+    Parallel_Orbitals *pv = new Parallel_Orbitals();
+    pv->set(mg, ng, nb, blacs_ctxt);
+    pv->set_atomic_trace(iat2iwt.data(), ucell_in->nat, mg);
+    auto ijr_info = hR->get_ijr_info();
+
+    auto* hR_tmp = new hamilt::HContainer<std::complex<double>>(pv, nullptr, &ijr_info);
+
+    //select hRGint_tmp
+    std::vector<int> first = {0, 1, 1, 0};
+    std::vector<int> second= {3, 2, 2, 3};
+    //select position in the big matrix
+    std::vector<int> row_set = {0, 0, 1, 1};
+    std::vector<int> col_set = {0, 1, 0, 1};
+    //construct complex matrix
+    std::vector<int> clx_i = {1, 0, 0, -1};
+    std::vector<int> clx_j = {0, 1, -1, 0};
+    for (int is = 0; is < 4; is++){
+        if(!PARAM.globalv.domag && (is==1 || is==2)) continue;
+        hR_tmp->set_zero();
+        hamilt::HContainer<std::complex<double>>* hRGint_tmpCd = new hamilt::HContainer<std::complex<double>>(ucell_in->nat);
+        hRGint_tmpCd->insert_ijrs(&ijr_info, *ucell_in);
+        hRGint_tmpCd->allocate(nullptr, true);
+        hRGint_tmpCd->set_zero();
+        for (int iap = 0; iap < hRGint_tmpCd->size_atom_pairs(); iap++)
         {
-            hamilt::AtomPair<std::complex<double>>* upper_ap = ap;
-            hamilt::AtomPair<std::complex<double>>* lower_ap = hr_gint_full.find_pair(iat2, iat1);
-            const hamilt::AtomPair<double>* ap_nspin_0 = hr_gint_part[0].find_pair(iat1, iat2);
-            const hamilt::AtomPair<double>* ap_nspin_3 = hr_gint_part[3].find_pair(iat1, iat2);
-            for (int ir = 0; ir < upper_ap->get_R_size(); ir++)
+            auto* ap = &hRGint_tmpCd->get_atom_pair(iap);
+            const int iat1 = ap->get_atom_i();
+            const int iat2 = ap->get_atom_j();
+            if (iat1 <= iat2)
             {
-                const auto R_index = upper_ap->get_R_index(ir);
-                auto upper_mat = upper_ap->find_matrix(R_index);
-                auto mat_nspin_0 = ap_nspin_0->find_matrix(R_index);
-                auto mat_nspin_3 = ap_nspin_3->find_matrix(R_index);
-
-                // The row size and the col size of upper_matrix is double that of matrix_nspin_0
-                for (int irow = 0; irow < mat_nspin_0->get_row_size(); ++irow)
-                {
-                    for (int icol = 0; icol < mat_nspin_0->get_col_size(); ++icol)
+                hamilt::AtomPair<std::complex<double>>* upper_ap = ap;
+                hamilt::AtomPair<std::complex<double>>* lower_ap = hRGint_tmpCd->find_pair(iat2, iat1);
+                const hamilt::AtomPair<double>* ap_nspin1 = hRGint_tmp[first[is]].find_pair(iat1, iat2);
+                const hamilt::AtomPair<double>* ap_nspin2 = hRGint_tmp[second[is]].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);
+                    auto upper_mat = upper_ap->find_matrix(R_index);
+                    auto mat_nspin1 = ap_nspin1->find_matrix(R_index);
+                    auto mat_nspin2 = ap_nspin2->find_matrix(R_index);
+                    // The row size and the col size of upper_matrix is double that of matrix_nspin_0
+                    for (int irow = 0; irow < mat_nspin1->get_row_size(); ++irow)
                     {
-                        upper_mat->get_value(2*irow, 2*icol) = mat_nspin_0->get_value(irow, icol) + mat_nspin_3->get_value(irow, icol);
-                        upper_mat->get_value(2*irow+1, 2*icol+1) = mat_nspin_0->get_value(irow, icol) - mat_nspin_3->get_value(irow, icol);
+                        for (int icol = 0; icol < mat_nspin1->get_col_size(); ++icol)
+                        {
+                            upper_mat->get_value(irow, icol) = mat_nspin1->get_value(irow, icol) 
+                            + std::complex<double>(clx_i[is], clx_j[is]) * mat_nspin2->get_value(irow, icol);
+                        }
                     }
-                }
-
-                if (PARAM.globalv.domag)
-                {
-                    const hamilt::AtomPair<double>* ap_nspin_1 = hr_gint_part[1].find_pair(iat1, iat2);
-                    const hamilt::AtomPair<double>* ap_nspin_2 = hr_gint_part[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)
+                    //fill the lower triangle matrix
+                    //When is=0 or 3, the real part does not need conjugation; 
+                    //when is=1 or 2, the small matrix is not Hermitian, so conjugation is not needed
+                    if (iat1 < iat2)
                     {
-                        for (int icol = 0; icol < mat_nspin_1->get_col_size(); ++icol)
+                        auto lower_mat = lower_ap->find_matrix(-R_index);
+                        for (int irow = 0; irow < upper_mat->get_row_size(); ++irow)
                         {
-                            upper_mat->get_value(2*irow, 2*icol+1) = mat_nspin_1->get_value(irow, icol) +  std::complex<double>(0.0, 1.0) * mat_nspin_2->get_value(irow, icol);
-                            upper_mat->get_value(2*irow+1, 2*icol) = mat_nspin_1->get_value(irow, icol) -  std::complex<double>(0.0, 1.0) * mat_nspin_2->get_value(irow, icol);
+                            for (int icol = 0; icol < upper_mat->get_col_size(); ++icol)
+                            {
+                                lower_mat->get_value(icol, irow) = upper_mat->get_value(irow, icol);
+                            }
                         }
                     }
+
                 }
+            }
+        }
 
-                // fill the lower triangle matrix
-                if (iat1 < iat2)
+        // transfer hRGint_tmpCd to parallel hR_tmp
+        hamilt::transferSerials2Parallels( *hRGint_tmpCd, hR_tmp);
+        // merge hR_tmp to hR
+        for (int iap = 0; iap < hR->size_atom_pairs(); iap++)
+        {
+            auto* ap = &hR->get_atom_pair(iap);
+            const int iat1 = ap->get_atom_i();
+            const int iat2 = ap->get_atom_j();
+            auto* ap_nspin = hR_tmp ->find_pair(iat1, iat2);
+            for (int ir = 0; ir < ap->get_R_size(); ir++)
+            {   
+                const auto R_index = ap->get_R_index(ir);
+                auto upper_mat = ap->find_matrix(R_index);
+                auto mat_nspin = ap_nspin->find_matrix(R_index);
+                // The row size and the col size of upper_matrix is double that of matrix_nspin_0
+                for (int irow = 0; irow < mat_nspin->get_row_size(); ++irow)
                 {
-                    auto lower_mat = lower_ap->find_matrix(-R_index);
-                    for (int irow = 0; irow < upper_mat->get_row_size(); ++irow)
+                    for (int icol = 0; icol < mat_nspin->get_col_size(); ++icol)
                     {
-                        for (int icol = 0; icol < upper_mat->get_col_size(); ++icol)
-                        {
-                            lower_mat->get_value(icol, irow) = conj(upper_mat->get_value(irow, icol));
-                        }
+                        upper_mat->get_value(2*irow+row_set[is], 2*icol+col_set[is]) = 
+                        mat_nspin->get_value(irow, icol);
                     }
                 }
             }
         }
-    }
-    ModuleBase::timer::tick("Gint", "compose_hr_gint");
-}
-
-template <typename T>
-void transfer_hr_gint_to_hR(const HContainer<T>& hr_gint, HContainer<T>& hR)
-{
-    ModuleBase::TITLE("Gint", "transfer_hr_gint_to_hR");
-    ModuleBase::timer::tick("Gint", "transfer_hr_gint_to_hR");
-#ifdef __MPI
-    int size = 0;
-    MPI_Comm_size(MPI_COMM_WORLD, &size);
-    if (size == 1)
-    {
-        hR.add(hr_gint);
-    }
-    else
-    {
-        hamilt::transferSerials2Parallels(hr_gint, &hR);
+        delete hRGint_tmpCd;
     }
 #else
-    hR.add(hr_gint);
+
 #endif
-    ModuleBase::timer::tick("Gint", "transfer_hr_gint_to_hR");
+    ModuleBase::timer::tick("Gint_k", "transfer_pvpR");
+    return;
 }
 
+
+
 // gint_info should not have been a parameter, but it was added to initialize dm_gint_full
 // In the future, we might try to remove the gint_info parameter
 template<typename T>

source/source_lcao/module_gint/temp_gint/gint_common.h

@@ -6,12 +6,14 @@ namespace ModuleGint
 {
     // fill the lower triangle matrix with the upper triangle matrix
     void compose_hr_gint(HContainer<double>& hr_gint);
-    // for nspin=4 case
-    void compose_hr_gint(const std::vector<HContainer<double>>& hr_gint_part,
-                         HContainer<std::complex<double>>& hr_gint_full);
+    
 
     template <typename T>
     void transfer_hr_gint_to_hR(const HContainer<T>& hr_gint, HContainer<T>& hR);
+    // for nspin=4 case
+    void merge_hr_part_to_hR(const std::vector<hamilt::HContainer<double>>& hRGint_tmp,
+                         hamilt::HContainer<std::complex<double>>* hR,
+                         const GintInfo& gint_info);
 
     template<typename T>
     void transfer_dm_2d_to_gint(

source/source_lcao/module_gint/temp_gint/gint_vl_metagga_nspin4.cpp

@@ -14,8 +14,7 @@ void Gint_vl_metagga_nspin4::cal_gint()
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
     init_hr_gint_();
     cal_hr_gint_();
-    compose_hr_gint(hr_gint_part_, hr_gint_full_);
-    transfer_hr_gint_to_hR(hr_gint_full_, *hR_);
+    merge_hr_part_to_hR(hr_gint_part_, hR_, *gint_info_);
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
 }
 
@@ -26,8 +25,6 @@ void Gint_vl_metagga_nspin4::init_hr_gint_()
     {
         hr_gint_part_[i] = gint_info_->get_hr<double>();
     }
-    const int npol = 2;
-    hr_gint_full_ = gint_info_->get_hr<std::complex<double>>(npol);
 }
 
 void Gint_vl_metagga_nspin4::cal_hr_gint_()

source/source_lcao/module_gint/temp_gint/gint_vl_metagga_nspin4.h

@@ -37,7 +37,6 @@ class Gint_vl_metagga_nspin4 : public Gint
     const int nspin_ = 4;
 
     std::vector<HContainer<double>> hr_gint_part_;
-    HContainer<std::complex<double>> hr_gint_full_;
 };
 
 }

source/source_lcao/module_gint/temp_gint/gint_vl_metagga_nspin4_gpu.cpp

@@ -13,8 +13,7 @@ void Gint_vl_metagga_nspin4_gpu::cal_gint()
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
     init_hr_gint_();
     cal_hr_gint_();
-    compose_hr_gint(hr_gint_part_, hr_gint_full_);
-    transfer_hr_gint_to_hR(hr_gint_full_, *hR_);
+    merge_hr_part_to_hR(hr_gint_part_, hR_, *gint_info_);
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
 }
 
@@ -25,8 +24,6 @@ void Gint_vl_metagga_nspin4_gpu::init_hr_gint_()
     {
         hr_gint_part_[i] = gint_info_->get_hr<double>();
     }
-    const int npol = 2;
-    hr_gint_full_ = gint_info_->get_hr<std::complex<double>>(npol);
 }
 
 void Gint_vl_metagga_nspin4_gpu::transfer_cpu_to_gpu_()

source/source_lcao/module_gint/temp_gint/gint_vl_metagga_nspin4_gpu.h

@@ -42,7 +42,6 @@ class Gint_vl_metagga_nspin4_gpu : public Gint
     const int nspin_ = 4;
 
     std::vector<HContainer<double>> hr_gint_part_;
-    HContainer<std::complex<double>> hr_gint_full_;
 
     std::vector<CudaMemWrapper<double>> vr_eff_d_;
     std::vector<CudaMemWrapper<double>> vofk_d_;

source/source_lcao/module_gint/temp_gint/gint_vl_nspin4.cpp

@@ -13,8 +13,7 @@ void Gint_vl_nspin4::cal_gint()
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
     init_hr_gint_();
     cal_hr_gint_();
-    compose_hr_gint(hr_gint_part_, hr_gint_full_);
-    transfer_hr_gint_to_hR(hr_gint_full_, *hR_);
+    merge_hr_part_to_hR(hr_gint_part_, hR_, *gint_info_);
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
 }
 
@@ -25,8 +24,6 @@ void Gint_vl_nspin4::init_hr_gint_()
     {
         hr_gint_part_[i] = gint_info_->get_hr<double>();
     }
-    const int npol = 2;
-    hr_gint_full_ = gint_info_->get_hr<std::complex<double>>(npol);
 }
 
 void Gint_vl_nspin4::cal_hr_gint_()

source/source_lcao/module_gint/temp_gint/gint_vl_nspin4.h

@@ -39,7 +39,6 @@ class Gint_vl_nspin4 : public Gint
     const int nspin_ = 4;
 
     std::vector<HContainer<double>> hr_gint_part_;
-    HContainer<std::complex<double>> hr_gint_full_;
 };
 
 } // namespace ModuleGint

source/source_lcao/module_gint/temp_gint/gint_vl_nspin4_gpu.cpp

@@ -13,8 +13,7 @@ void Gint_vl_nspin4_gpu::cal_gint()
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
     init_hr_gint_();
     cal_hr_gint_();
-    compose_hr_gint(hr_gint_part_, hr_gint_full_);
-    transfer_hr_gint_to_hR(hr_gint_full_, *hR_);
+    merge_hr_part_to_hR(hr_gint_part_, *hR_, *gint_info_);
     ModuleBase::timer::tick("Gint", "cal_gint_vl");
 }
 
@@ -25,8 +24,6 @@ void Gint_vl_nspin4_gpu::init_hr_gint_()
     {
         hr_gint_part_[i] = gint_info_->get_hr<double>();
     }
-    const int npol = 2;
-    hr_gint_full_ = gint_info_->get_hr<std::complex<double>>(npol);
 }
 
 void Gint_vl_nspin4_gpu::transfer_cpu_to_gpu_()

source/source_lcao/module_gint/temp_gint/gint_vl_nspin4_gpu.h

@@ -44,7 +44,6 @@ class Gint_vl_nspin4_gpu : public Gint
     const int nspin_ = 4;
 
     std::vector<HContainer<double>> hr_gint_part_;
-    HContainer<std::complex<double>> hr_gint_full_;
 
     std::vector<CudaMemWrapper<double>> vr_eff_d_;
     std::vector<CudaMemWrapper<double>> hr_gint_part_d_;