Bug fix regarding to open shell with cuda and libxc

MALBECC · Sep 6, 2023 · e30dc87 · e30dc87
1 parent 4ee8f59
commit e30dc87
Show file tree

Hide file tree

Showing 4 changed files with 47 additions and 0 deletions.
diff --git a/g2g/cuda/iteration.cu b/g2g/cuda/iteration.cu
@@ -939,6 +939,24 @@ void PointGroupGPU<scalar_type>::solve_opened(
   //cudaPrintMemoryInfo();
 }
 
+template<class scalar_type>
+void PointGroupGPU<scalar_type>::compute_rmm_libxc(const uint& group_m, const scalar_type* fv,
+		const scalar_type* gxv, const scalar_type* gyv, const scalar_type* gzv,
+		const scalar_type& wp, double* coef_a, double* coef_b,
+		const G2G::vec_type<scalar_type, 3>& dxyz_a, const G2G::vec_type<scalar_type, 3>& dxyz_b,
+		double* smallFock_a, double* smallFock_b) {}
+
+template<class scalar_type>
+void PointGroupGPU<scalar_type>::compute_forces_libxc(const uint& group_m, const scalar_type& wp, int& local_atoms,
+		const scalar_type* fv, const scalar_type* gxv, const scalar_type* gyv, const scalar_type* gzv,
+		const scalar_type* hpxv, const scalar_type* hpyv, const scalar_type* hpzv,
+		const scalar_type* hixv, const scalar_type* hiyv, const scalar_type* hizv,
+		HostMatrix<scalar_type>& rmm_input_a, HostMatrix<scalar_type>& rmm_input_b,
+		const G2G::vec_type<scalar_type, 3>& dxyz_a, const G2G::vec_type<scalar_type, 3>& dxyz_b,
+		double* coef_a, double* coef_b, 
+		HostMatrix<scalar_type>& ddx_a, HostMatrix<scalar_type>& ddy_a, HostMatrix<scalar_type>& ddz_a,
+		HostMatrix<scalar_type>& ddx_b, HostMatrix<scalar_type>& ddy_b, HostMatrix<scalar_type>& ddz_b,
+		double* smallFor_a, double* smallFor_b) {}
 
 
 /*******************************

diff --git a/g2g/excited/cuda/fake_cuda.cu b/g2g/excited/cuda/fake_cuda.cu
@@ -22,6 +22,9 @@ void PointGroupGPU<scalar_type>::lr_closed_init() { }
 template<class scalar_type>
 void PointGroupGPU<scalar_type>::solve_closed_lr(double* T,HostMatrix<double>& Fock,int DER) { }
 
+template<class scalar_type>
+void PointGroupGPU<scalar_type>::solve_open_lr(double* Ta, double* Tb, HostMatrix<double>& FockA, HostMatrix<double>& FockB,int DER) {}
+
 template <class scalar_type>
 void PointGroupGPU<scalar_type>::get_tred_input(
      HostMatrix<scalar_type>& tred_input, HostMatrix<double>& source) const { }

diff --git a/g2g/excited/cuda/g2g_calculateXC.cu b/g2g/excited/cuda/g2g_calculateXC.cu
@@ -288,6 +288,13 @@ void PointGroupGPU<scalar_type>::solve_closed_lr(double* T, HostMatrix<double>&
    }
 }
 
+// For open shell does not work
+template<class scalar_type>
+void PointGroupGPU<scalar_type>::solve_open_lr(double* Ta, double* Tb, HostMatrix<double>& FockA, HostMatrix<double>& FockB,int DER) {
+	std::cout << " Linear Response does not work for open shell and cuda, compile lio in cpu mode!" << std::endl;
+	exit(-1);
+}
+
 template <class scalar_type>
 void PointGroupGPU<scalar_type>::get_tred_input(
     HostMatrix<scalar_type>& tred_input, HostMatrix<double>& source) const

diff --git a/g2g/partition.h b/g2g/partition.h
@@ -280,10 +280,29 @@ class PointGroupGPU: public PointGroup<scalar_type> {
     virtual void solve(Timers& timers, bool compute_rmm, bool lda, bool compute_forces,
         bool compute_energy, double& energy, double &, double &, double &, double &,
         HostMatrix<double> &, int, HostMatrix<double> &, bool);
+    //
+    // For open shell with libxc, only works in cpu
+    virtual void compute_rmm_libxc(const uint& group_m, const scalar_type* fv,
+                 const scalar_type* gxv, const scalar_type* gyv, const scalar_type* gzv,
+                 const scalar_type& wp, double* coef_a, double* coef_b,
+                 const G2G::vec_type<scalar_type, 3>& dxyz_a, const G2G::vec_type<scalar_type, 3>& dxyz_b,
+                 double* smallFock_a, double* smallFock_b);
+
+    virtual void compute_forces_libxc(const uint& group_m, const scalar_type& wp, int& local_atoms,
+                 const scalar_type* fv, const scalar_type* gxv, const scalar_type* gyv, const scalar_type* gzv,
+                 const scalar_type* hpxv, const scalar_type* hpyv, const scalar_type* hpzv,
+                 const scalar_type* hixv, const scalar_type* hiyv, const scalar_type* hizv,
+                 HostMatrix<scalar_type>& rmm_input_a, HostMatrix<scalar_type>& rmm_input_b,
+                 const G2G::vec_type<scalar_type, 3>& dxyz_a, const G2G::vec_type<scalar_type, 3>& dxyz_b,
+                 double* coef_a, double* coef_b, 
+                 HostMatrix<scalar_type>& ddx_a, HostMatrix<scalar_type>& ddy_a, HostMatrix<scalar_type>& ddz_a,
+                 HostMatrix<scalar_type>& ddx_b, HostMatrix<scalar_type>& ddy_b, HostMatrix<scalar_type>& ddz_b,
+                 double* smallFor_a, double* smallFor_b);
 
     virtual void get_tred_input(G2G::HostMatrix<scalar_type>& tre_input,G2G::HostMatrix<double>& source) const;
     virtual void lr_closed_init();
     virtual void solve_closed_lr(double* T,HostMatrix<double>& Fock,int DER);
+    virtual void solve_open_lr(double* Ta, double* Tb, HostMatrix<double>& FockA, HostMatrix<double>& FockB,int DER);
     virtual void solve_for_exc(double* P,double* V,HostMatrix<double>& F,int met);
     virtual void calc_W_mat(HostMatrix<double>& , CDFTVars&);