Update namespace ModuleGint

source/module_base/blas_connector.h

@@ -65,7 +65,7 @@ extern "C"
 	void cgemv_(const char *trans, const int *m, const int *n, const std::complex<float> *alpha,
 			const std::complex<float> *a, const int *lda, const std::complex<float> *x, const int *incx,
 			const std::complex<float> *beta, std::complex<float> *y, const int *incy);
-		
+
 	void zgemv_(const char *trans, const int *m, const int *n, const std::complex<double> *alpha,
 			const std::complex<double> *a, const int *lda, const std::complex<double> *x, const int *incx,
 			const std::complex<double> *beta, std::complex<double> *y, const int *incy);
@@ -180,11 +180,36 @@ class BlasConnector
 	// Peize Lin add 2017-10-27
 	// d=x*y
 	static
-	float dot( const int n, const float *X, const int incX, const float *Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+	float dot( const int n, const float*const X, const int incX, const float*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	double dot( const int n, const double*const X, const int incX, const double*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	// d=x*y
+	static
+	float dotu( const int n, const float*const X, const int incX, const float*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	double dotu( const int n, const double*const X, const int incX, const double*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	std::complex<float> dotu( const int n, const std::complex<float>*const X, const int incX, const std::complex<float>*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	std::complex<double> dotu( const int n, const std::complex<double>*const X, const int incX, const std::complex<double>*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	// d=x.conj()*y
+	static
+	float dotc( const int n, const float*const X, const int incX, const float*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	double dotc( const int n, const double*const X, const int incX, const double*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
 	static
-	double dot( const int n, const double *X, const int incX, const double *Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+	std::complex<float> dotc( const int n, const std::complex<float>*const X, const int incX, const std::complex<float>*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
+	static
+	std::complex<double> dotc( const int n, const std::complex<double>*const X, const int incX, const std::complex<double>*const Y, const int incY, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
 	// Peize Lin add 2017-10-27, fix bug trans 2019-01-17
 	// C = a * A.? * B.? + b * C
@@ -231,6 +256,9 @@ class BlasConnector
 		const std::complex<double> alpha, const std::complex<double> *a, const int lda, const std::complex<double> *b, const int ldb,
 		const std::complex<double> beta, std::complex<double> *c, const int ldc, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
+	// side=='L': C = a * A * B + b * C.
+	// side=='R': C = a * B * A + b * C.
+	// 		A == A^T
 	// Because you cannot pack symm or hemm into a row-major kernel by exchanging parameters, so only col-major functions are provided.
 	static
 	void symm_cm(const char side, const char uplo, const int m, const int n,
@@ -252,6 +280,19 @@ class BlasConnector
 		const std::complex<double> alpha, const std::complex<double> *a, const int lda, const std::complex<double> *b, const int ldb,
 		const std::complex<double> beta, std::complex<double> *c, const int ldc, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
+	// side=='L': C = a * A * B + b * C.
+	// side=='R': C = a * B * A + b * C.
+	// 		A == A^H
+	static
+	void hemm_cm(const char side, const char uplo, const int m, const int n,
+		const float alpha, const float *a, const int lda, const float *b, const int ldb,
+		const float beta, float *c, const int ldc, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
+	static
+	void hemm_cm(const char side, const char uplo, const int m, const int n,
+		const double alpha, const double *a, const int lda, const double *b, const int ldb,
+		const double beta, double *c, const int ldc, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
+
 	static
     void hemm_cm(char side, char uplo, int m, int n,
             std::complex<float> alpha, std::complex<float> *a, int lda, std::complex<float> *b, int ldb,
@@ -263,7 +304,6 @@ class BlasConnector
 		std::complex<double> beta, std::complex<double> *c, int ldc, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
 
 	// y = A*x + beta*y
-
 	static
 	void gemv(const char trans, const int m, const int n,
         const float alpha, const float* A, const int lda, const float* X, const int incx,
@@ -283,7 +323,6 @@ class BlasConnector
     void gemv(const char trans, const int m, const int n,
               const std::complex<double> alpha, const std::complex<double> *A, const int lda, const std::complex<double> *X, const int incx,
               const std::complex<double> beta, std::complex<double> *Y, const int incy, base_device::AbacusDevice_t device_type = base_device::AbacusDevice_t::CpuDevice);
-
 
 	// Peize Lin add 2018-06-12
 	// out = ||x||_2

source/module_hamilt_lcao/module_gint/temp_gint/gint_atom.cpp

@@ -203,5 +203,6 @@ void GintAtom::set_phi_dphi(
 
 // explicit instantiation
 template void GintAtom::set_phi(const std::vector<Vec3d>& coords, const int stride, double* phi) const;
+template void GintAtom::set_phi(const std::vector<Vec3d>& coords, const int stride, std::complex<double>* phi) const;
 template void GintAtom::set_phi_dphi(const std::vector<Vec3d>& coords, const int stride, double* phi, double* dphi_x, double* dphi_y, double* dphi_z) const;
 }

source/module_hamilt_lcao/module_gint/temp_gint/gint_common.cpp

@@ -23,7 +23,7 @@ void compose_hr_gint(std::shared_ptr<HContainer<double>> hr_gint)
             assert(upper_ap != nullptr);
 #endif
             for (int ir = 0; ir < ap.get_R_size(); ir++)
-            {   
+            {
                 auto R_index = ap.get_R_index(ir);
                 auto upper_mat = upper_ap->find_matrix(-R_index);
                 auto lower_mat = lower_ap->find_matrix(R_index);
@@ -36,7 +36,7 @@ void compose_hr_gint(std::shared_ptr<HContainer<double>> hr_gint)
                 }
             }
         }
-    } 
+    }
 }
 
 void compose_hr_gint(std::vector<std::shared_ptr<HContainer<double>>> hr_gint_part,
@@ -54,7 +54,7 @@ void compose_hr_gint(std::vector<std::shared_ptr<HContainer<double>>> hr_gint_pa
             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++)
-            {   
+            {
                 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);
@@ -124,10 +124,11 @@ void transfer_hr_gint_to_hR(std::shared_ptr<const HContainer<T>> hr_gint, HConta
 
 // 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>
 void transfer_dm_2d_to_gint(
     std::shared_ptr<const GintInfo> gint_info,
-    std::vector<HContainer<double>*> dm,
-    std::vector<std::shared_ptr<HContainer<double>>> dm_gint)
+    std::vector<HContainer<T>*> dm,
+    std::vector<std::shared_ptr<HContainer<T>>> dm_gint)
 {
     // To check whether input parameter dm_2d has been initialized
 #ifdef __DEBUG
@@ -150,12 +151,12 @@ void transfer_dm_2d_to_gint(
     {
 #ifdef __MPI
         const int npol = 2;
-        std::shared_ptr<HContainer<double>> dm_full = gint_info->get_hr<double>(npol);
+        std::shared_ptr<HContainer<T>> dm_full = gint_info->get_hr<T>(npol);
         hamilt::transferParallels2Serials(*dm[0], dm_full.get());
 #else
-        HContainer<double>* dm_full = dm[0];
+        HContainer<T>* dm_full = dm[0];
 #endif
-        std::vector<double*> tmp_pointer(4, nullptr);
+        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);
@@ -166,10 +167,10 @@ void transfer_dm_2d_to_gint(
                 const ModuleBase::Vector3<int> r_index = ap.get_R_index(ir);
                 for (int is = 0; is < 4; is++)
                 {
-                    tmp_pointer[is] = 
+                    tmp_pointer[is] =
                         dm_gint[is]->find_matrix(iat1, iat2, r_index)->get_pointer();
                 }
-                double* data_full = ap.get_pointer(ir);
+                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)
@@ -191,6 +192,18 @@ void transfer_dm_2d_to_gint(
 }
 
 
-template void transfer_hr_gint_to_hR(std::shared_ptr<const HContainer<double>> hr_gint, HContainer<double>* hR);
-template void transfer_hr_gint_to_hR(std::shared_ptr<const HContainer<std::complex<double>>> hr_gint, HContainer<std::complex<double>>* hR);
+template void transfer_hr_gint_to_hR(
+    std::shared_ptr<const HContainer<double>> hr_gint,
+    HContainer<double>* hR);
+template void transfer_hr_gint_to_hR(
+    std::shared_ptr<const HContainer<std::complex<double>>> hr_gint,
+    HContainer<std::complex<double>>* hR);
+template void transfer_dm_2d_to_gint(
+    std::shared_ptr<const GintInfo> gint_info,
+    std::vector<HContainer<double>*> dm,
+    std::vector<std::shared_ptr<HContainer<double>>> dm_gint);
+template void transfer_dm_2d_to_gint(
+    std::shared_ptr<const GintInfo> gint_info,
+    std::vector<HContainer<std::complex<double>>*> dm,
+    std::vector<std::shared_ptr<HContainer<std::complex<double>>>> dm_gint);
 }

source/module_hamilt_lcao/module_gint/temp_gint/gint_common.h

@@ -13,9 +13,10 @@ namespace ModuleGint
     template <typename T>
     void transfer_hr_gint_to_hR(std::shared_ptr<const HContainer<T>> hr_gint, HContainer<T>* hR);
 
+    template<typename T>
     void transfer_dm_2d_to_gint(
         std::shared_ptr<const GintInfo> gint_info,
-        std::vector<HContainer<double>*> dm,
-        std::vector<std::shared_ptr<HContainer<double>>> dm_gint);
+        std::vector<HContainer<T>*> dm,
+        std::vector<std::shared_ptr<HContainer<T>>> dm_gint);
 
 }

source/module_hamilt_lcao/module_gint/temp_gint/gint_rho.cpp

@@ -44,7 +44,7 @@ void Gint_rho::cal_rho_()
             for (int is = 0; is < nspin_; is++)
             {
                 phi_op.phi_mul_dm(phi.data(), *dm_gint_vec_[is], true, phi_dm.data());
-                phi_op.phi_dot_phi_dm(phi.data(), phi_dm.data(), rho_[is]);
+                phi_op.phi_dot_phi(phi.data(), phi_dm.data(), rho_[is]);
             }
         }
     }

source/module_hamilt_lcao/module_gint/temp_gint/gint_tau.cpp

@@ -54,9 +54,9 @@ void Gint_tau::cal_tau_()
                 phi_op.phi_mul_dm(dphi_x.data(), *dm_gint_vec_[is], true, dphi_x_dm.data());
                 phi_op.phi_mul_dm(dphi_y.data(), *dm_gint_vec_[is], true, dphi_y_dm.data());
                 phi_op.phi_mul_dm(dphi_z.data(), *dm_gint_vec_[is], true, dphi_z_dm.data());
-                phi_op.phi_dot_phi_dm(dphi_x.data(), dphi_x_dm.data(), kin_[is]);
-                phi_op.phi_dot_phi_dm(dphi_y.data(), dphi_y_dm.data(), kin_[is]);
-                phi_op.phi_dot_phi_dm(dphi_z.data(), dphi_z_dm.data(), kin_[is]);
+                phi_op.phi_dot_phi(dphi_x.data(), dphi_x_dm.data(), kin_[is]);
+                phi_op.phi_dot_phi(dphi_y.data(), dphi_y_dm.data(), kin_[is]);
+                phi_op.phi_dot_phi(dphi_z.data(), dphi_z_dm.data(), kin_[is]);
             }
         }
     }

source/module_hamilt_lcao/module_gint/temp_gint/gint_vl.cpp

@@ -36,7 +36,7 @@ void Gint_vl::cal_hr_gint_()
 #pragma omp for schedule(dynamic)
         for(const auto& biggrid: gint_info_->get_biggrids())
         {
-            if(biggrid->get_atoms().size() == 0)
+            if(biggrid->get_atoms().empty())
             {
                 continue;
             }
@@ -46,7 +46,7 @@ void Gint_vl::cal_hr_gint_()
             phi_vldr3.resize(phi_len);
             phi_op.set_phi(phi.data());
             phi_op.phi_mul_vldr3(vr_eff_, dr3_, phi.data(), phi_vldr3.data());
-            phi_op.phi_mul_phi_vldr3(phi.data(), phi_vldr3.data(), *hr_gint_);
+            phi_op.phi_mul_phi(phi.data(), phi_vldr3.data(), *hr_gint_, PhiOperator::Triangular_Matrix::Upper);
         }
     }
 }

source/module_hamilt_lcao/module_gint/temp_gint/gint_vl_metagga.cpp

@@ -61,10 +61,10 @@ void Gint_vl_metagga::cal_hr_gint_()
             phi_op.phi_mul_vldr3(vofk_, dr3_, dphi_x.data(), dphi_x_vldr3.data());
             phi_op.phi_mul_vldr3(vofk_, dr3_, dphi_y.data(), dphi_y_vldr3.data());
             phi_op.phi_mul_vldr3(vofk_, dr3_, dphi_z.data(), dphi_z_vldr3.data());
-            phi_op.phi_mul_phi_vldr3(phi.data(), phi_vldr3.data(), *hr_gint_);
-            phi_op.phi_mul_phi_vldr3(dphi_x.data(), dphi_x_vldr3.data(), *hr_gint_);
-            phi_op.phi_mul_phi_vldr3(dphi_y.data(), dphi_y_vldr3.data(), *hr_gint_);
-            phi_op.phi_mul_phi_vldr3(dphi_z.data(), dphi_z_vldr3.data(), *hr_gint_);
+            phi_op.phi_mul_phi(phi.data(), phi_vldr3.data(), *hr_gint_, PhiOperator::Triangular_Matrix::Upper);
+            phi_op.phi_mul_phi(dphi_x.data(), dphi_x_vldr3.data(), *hr_gint_, PhiOperator::Triangular_Matrix::Upper);
+            phi_op.phi_mul_phi(dphi_y.data(), dphi_y_vldr3.data(), *hr_gint_, PhiOperator::Triangular_Matrix::Upper);
+            phi_op.phi_mul_phi(dphi_z.data(), dphi_z_vldr3.data(), *hr_gint_, PhiOperator::Triangular_Matrix::Upper);
         }
     }
 }

source/module_hamilt_lcao/module_gint/temp_gint/gint_vl_metagga_nspin4.cpp

@@ -65,10 +65,10 @@ void Gint_vl_metagga_nspin4::cal_hr_gint_()
                 phi_op.phi_mul_vldr3(vofk_[is], dr3_, dphi_x.data(), dphi_x_vldr3.data());
                 phi_op.phi_mul_vldr3(vofk_[is], dr3_, dphi_y.data(), dphi_y_vldr3.data());
                 phi_op.phi_mul_vldr3(vofk_[is], dr3_, dphi_z.data(), dphi_z_vldr3.data());
-                phi_op.phi_mul_phi_vldr3(phi.data(), phi_vldr3.data(), *hr_gint_part_[is]);
-                phi_op.phi_mul_phi_vldr3(dphi_x.data(), dphi_x_vldr3.data(), *hr_gint_part_[is]);
-                phi_op.phi_mul_phi_vldr3(dphi_y.data(), dphi_y_vldr3.data(), *hr_gint_part_[is]);
-                phi_op.phi_mul_phi_vldr3(dphi_z.data(), dphi_z_vldr3.data(), *hr_gint_part_[is]);
+                phi_op.phi_mul_phi(phi.data(), phi_vldr3.data(), *hr_gint_part_[is], PhiOperator::Triangular_Matrix::Upper);
+                phi_op.phi_mul_phi(dphi_x.data(), dphi_x_vldr3.data(), *hr_gint_part_[is], PhiOperator::Triangular_Matrix::Upper);
+                phi_op.phi_mul_phi(dphi_y.data(), dphi_y_vldr3.data(), *hr_gint_part_[is], PhiOperator::Triangular_Matrix::Upper);
+                phi_op.phi_mul_phi(dphi_z.data(), dphi_z_vldr3.data(), *hr_gint_part_[is], PhiOperator::Triangular_Matrix::Upper);
             }
         }
     }

source/module_hamilt_lcao/module_gint/temp_gint/gint_vl_nspin4.cpp

@@ -49,7 +49,7 @@ void Gint_vl_nspin4::cal_hr_gint_()
             for(int is = 0; is < nspin_; is++)
             {
                 phi_op.phi_mul_vldr3(vr_eff_[is], dr3_, phi.data(), phi_vldr3.data());
-                phi_op.phi_mul_phi_vldr3(phi.data(), phi_vldr3.data(), *hr_gint_part_[is]);
+                phi_op.phi_mul_phi(phi.data(), phi_vldr3.data(), *hr_gint_part_[is], PhiOperator::Triangular_Matrix::Upper);
             }
         }
     }