ngsolve 6.2.2502__cp311-cp311-win_amd64.whl → 6.2.2601__cp311-cp311-win_amd64.whl

netgen/include/mycomplex.hpp

@@ -76,7 +76,7 @@ INLINE const T arg(const MyComplex<T> &A)
 }
 
 template <class T>
-const MyComplex<T> conj(const MyComplex<T> &A)
+INLINE const MyComplex<T> conj(const MyComplex<T> &A)
 {
   return MyComplex<T>(A.a, -A.b);
 }

netgen/include/ngblas.hpp

@@ -412,10 +412,10 @@ namespace ngbla
 
   
   template <bool ADD, bool POS, ORDERING orda, ORDERING ordb>
-  void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double> c);
+  INLINE void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double> c);
 
   template <bool ADD, bool POS, ORDERING orda, ORDERING ordb>
-  void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double,ColMajor> c);
+  INLINE void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double,ColMajor> c);
   
   
 
@@ -428,7 +428,7 @@ namespace ngbla
   // t   t    C += A*B
   
   template <bool ADD, bool POS, ORDERING orda, ORDERING ordb>
-  inline void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double> c)
+  INLINE void NgGEMM (SliceMatrix<double,orda> a, SliceMatrix<double, ordb> b, SliceMatrix<double> c)
   {
     // static Timer t("generic MM, add/pos/ord="+ToString(ADD)+ToString(POS)+ToString(orda)+ToString(ordb));
     // RegionTimer r(t);
@@ -584,12 +584,31 @@ namespace ngbla
       }
   }
 
-  template <typename TM, typename TVX, typename TVY>
-  extern void TestFunc (TM m, TVX x, TVY y);
-
+  // template <typename TM, typename TVX, typename TVY>
+  // extern void TestFunc (TM m, TVX x, TVY y);
+  
+  
+  template <typename TS, typename T> constexpr bool IsVec = false;
+  template <typename TS, int S> constexpr bool IsVec<TS, Vec<S,TS>> = true;  
+  
   template <bool ADD, bool POS, typename TM, ORDERING ORD, typename TX, typename TY>
   INLINE void NgGEMV (BareSliceMatrix<TM,ORD> a, FlatVector<const TX> x, FlatVector<TY> y)
   {
+    if constexpr (std::is_same<TM,double>() && std::is_same<TX,TY>() && IsVec<Complex,TX>)
+      {
+        FlatMatrix<double> mx(x.Size(), sizeof(TX)/sizeof(double), (double*)(void*)x.Addr(0));
+        FlatMatrix<double> my(y.Size(), sizeof(TX)/sizeof(double), (double*)(void*)y.Addr(0));
+        NgGEMM<ADD,POS> (a.AddSize(y.Size(), x.Size()),make_SliceMatrix(mx), make_SliceMatrix(my));
+        return;
+      }
+    if constexpr (std::is_same<TM,Complex>() && std::is_same<TX,TY>() && IsVec<Complex,TX>)
+      {
+        FlatMatrix<Complex> mx(x.Size(), sizeof(TX)/sizeof(Complex), &const_cast<Complex&>(*(x.Data()->Data())));
+        FlatMatrix<Complex> my(y.Size(), sizeof(TX)/sizeof(Complex), y.Data()->Data());
+        NgGEMM<ADD,POS> (a.AddSize(y.Size(), x.Size()),make_SliceMatrix(mx), make_SliceMatrix(my));
+        return;
+      }
+
     if (!ADD)
       {
         if (!POS)
@@ -635,6 +654,16 @@ namespace ngbla
   extern NGS_DLL_HEADER  
   void NgGEMV (double s, BareSliceMatrix<double,ord> a, SliceVector<double> x, SliceVector<double> y) NETGEN_NOEXCEPT;
   */
+
+
+
+
+
+
+
+  /* *********************** GEMV - SliceVector **************************** */
+
+
   
   template <bool ADD, ORDERING ord>
   extern NGS_DLL_HEADER  
@@ -714,7 +743,30 @@ namespace ngbla
   }
 
 
-
+  
+  
+  template <bool ADD, bool POS, typename TM, ORDERING ORD, typename TX, typename TY>
+  INLINE void NgGEMV (BareSliceMatrix<TM,ORD> a, SliceVector<TX> x, SliceVector<TY> y)
+  {
+    if constexpr (std::is_same<TM,double>() && std::is_same<TX,TY>() && IsVec<Complex,TX>)
+      {
+        constexpr int VS = sizeof(TX)/sizeof(double);
+        SliceMatrix<double> mx(x.Size(), VS, x.Dist()*VS, (double*)(void*)x.Addr(0));
+        SliceMatrix<double> my(y.Size(), VS, y.Dist()*VS, (double*)(void*)y.Addr(0));
+        NgGEMM<ADD,POS> (a.AddSize(y.Size(), x.Size()),make_SliceMatrix(mx), make_SliceMatrix(my));
+        return;
+      }
+    else if constexpr (std::is_same<TM,Complex>() && std::is_same<TX,TY>() && IsVec<Complex,TX>)
+      {
+        constexpr int VS = sizeof(TX)/sizeof(Complex);
+        SliceMatrix<Complex> mx(x.Size(), VS, x.Dist()*VS, &const_cast<Complex&>(*(x.Data()->Data())));
+        SliceMatrix<Complex> my(y.Size(), VS, y.Dist()*VS, y.Data()->Data());
+        NgGEMM<ADD,POS> (a.AddSize(y.Size(), x.Size()),make_SliceMatrix(mx), make_SliceMatrix(my));
+        return;
+      }
+    else
+      NgGEMV<ADD> (POS ? 1.0 : -1.0, a, x, y);
+  }
 
 
 
@@ -880,9 +932,14 @@ namespace ngbla
                          FlatVector<const TB>(prod.View().B().Range(0,w)),
                          FlatVector<T>(self.Spec().Range(0,h)));
       else
+        NgGEMV<ADD,POS> (make_BareSliceMatrix(prod.View().A()),
+                         SliceVector<TB>(prod.View().B().Range(0,w)),
+                         SliceVector<T>(self.Spec().Range(0,h)));
+        /*
         NgGEMV<ADD> (POS ? 1.0 : -1.0, make_BareSliceMatrix(prod.View().A()),
                      SliceVector<TB>(prod.View().B().Range(0,w)),
                      SliceVector<T>(self.Spec().Range(0,h)));
+        */
       return self.Spec();
     }
   };
@@ -1099,8 +1156,60 @@ namespace ngbla
     }
   };
   
+  // typedef void (*pmatmatcRR)(size_t, size_t, BareSliceMatrix<Complex,RowMajor>, BareSliceMatrix<Complex,RowMajor>,BareSliceMatrix<Complex,RowMajor>);
+
+  template <ORDERING OA, ORDERING OB>
+  using pmatmatc =  void (*)(size_t, size_t, BareSliceMatrix<Complex, OA>, BareSliceMatrix<Complex,OB>,BareSliceMatrix<Complex,RowMajor>);
+  template <bool ADD, bool POS, ORDERING OA, ORDERING OB>
+  extern NGS_DLL_HEADER pmatmatc<OA,OB> dispatch_matmatc[9];
+
+  template <bool ADD, bool POS, ORDERING OA, ORDERING OB>
+  extern NGS_DLL_HEADER void NgGEMMBare (size_t ah, size_t aw, size_t bw, BareSliceMatrix<Complex,OA> a, BareSliceMatrix<Complex,OB> b, BareSliceMatrix<Complex,RowMajor> c);
 
 
+  template <bool ADD, bool POS, ORDERING OA, ORDERING OB>
+  void NgGEMM (SliceMatrix<Complex,OA> a, SliceMatrix<Complex,OB> b, SliceMatrix<Complex,RowMajor> c)
+  {
+    size_t ah = a.Height();
+    size_t aw = a.Width();
+    size_t bw = b.Width();
+    if (aw < std::size(dispatch_matmatc<ADD,POS,OA,OB>))
+      {
+        (*dispatch_matmatc<ADD,POS,OA,OB>[aw])(ah, bw, make_BareSliceMatrix(a), make_BareSliceMatrix(b), make_BareSliceMatrix(c));
+        return;
+      }
+
+    NgGEMMBare<ADD,POS>(ah, aw, bw, make_BareSliceMatrix(a), make_BareSliceMatrix(b), make_BareSliceMatrix(c));
+  }
+  
+  template <bool ADD, bool POS, ORDERING OA, ORDERING OB>
+  void NgGEMM (SliceMatrix<Complex,OA> a, SliceMatrix<Complex,OB> b, SliceMatrix<Complex,ColMajor> c)
+  {
+    NgGEMM<ADD,POS> (Trans(b), Trans(a), Trans(c));
+  }
+  
+  template <typename OP, typename T, typename TA, typename TB>
+  class assign_trait<OP, T, MultExpr<TA, TB>,
+                     enable_if_t<IsConvertibleToSliceMatrix<TA,Complex>() &&
+                                 IsConvertibleToSliceMatrix<TB,Complex>() &&
+                                 IsConvertibleToSliceMatrix<T,Complex>(), int>>
+  {
+  public:
+    static inline T & Assign (MatExpr<T> & self, const Expr<MultExpr<TA, TB>> & prod) 
+    {
+      constexpr bool ADD = std::is_same<OP,typename MatExpr<T>::AsAdd>::value || std::is_same<OP,typename MatExpr<T>::AsSub>::value;
+      constexpr bool POS = std::is_same<OP,typename MatExpr<T>::As>::value || std::is_same<OP,typename MatExpr<T>::AsAdd>::value;
+
+      size_t n = CombinedSize(prod.View().A().Height(), self.Spec().Height());
+      size_t m = CombinedSize(prod.View().B().Width(), self.Spec().Width());
+      size_t k = CombinedSize(prod.View().A().Width(), prod.View().B().Height());
+      
+      NgGEMM<ADD,POS> (make_BareSliceMatrix(prod.View().A()).AddSize(n,k).RemoveConst(),
+                       make_BareSliceMatrix(prod.View().B()).AddSize(k,m).RemoveConst(),
+                       make_BareSliceMatrix(self.Spec()).AddSize(n,m));
+      return self.Spec();
+    }
+  };
 
   
 

netgen/include/potentialtools.hpp

@@ -0,0 +1,22 @@
+#ifndef FILE_POTENTIALS
+#define FILE_POTENTIALS
+
+/*
+  tools for computing with potentials using multipoles
+ */
+
+
+#include <mptools.hpp>
+#include <meshaccess.hpp>
+
+
+namespace ngsbem
+{
+
+  extern void AddChargeDensity (SingularMLExpansion<Complex> & mp, shared_ptr<CoefficientFunction> current, ngcomp::Region reg);
+  
+  extern void AddCurrentDensity (SingularMLExpansion<Vec<3,Complex>> & mp, shared_ptr<CoefficientFunction> current, ngcomp::Region reg);
+ 
+}
+
+#endif

netgen/include/preconditioner.hpp

@@ -99,12 +99,12 @@ namespace ngcomp
     virtual void InitLevel (shared_ptr<BitArray> freedofs = NULL) { ; }
     virtual void FinalizeLevel (const ngla::BaseMatrix * mat = NULL) { ; }
     virtual void AddElementMatrix (FlatArray<int> dnums,
-				   const FlatMatrix<double> & elmat,
+				   FlatMatrix<double> elmat,
 				   ElementId ei, 
 				   LocalHeap & lh) { ; }
 
     virtual void AddElementMatrix (FlatArray<int> dnums,
-				   const FlatMatrix<Complex> & elmat,
+				   FlatMatrix<Complex> elmat,
 				   ElementId ei, 
 				   LocalHeap & lh) { ; }
 
@@ -186,12 +186,12 @@ namespace ngcomp
     static DocInfo GetDocu ();
     
     ///
-    virtual bool IsComplex() const { return jacobi->IsComplex(); }
+    virtual bool IsComplex() const override { return jacobi->IsComplex(); }
     
     ///
-    virtual void FinalizeLevel (const BaseMatrix * mat);
+    virtual void FinalizeLevel (const BaseMatrix * mat) override;
 
-    virtual void Update ()
+    virtual void Update () override
     {
       if (GetTimeStamp() < bfa->GetTimeStamp())
         FinalizeLevel (&bfa->GetMatrix());
@@ -201,14 +201,14 @@ namespace ngcomp
 
 
     ///
-    virtual const BaseMatrix & GetMatrix() const
+    virtual const BaseMatrix & GetMatrix() const override
     {
       if (!jacobi)
         ThrowPreconditionerNotReady();
       return *jacobi;
     }
     
-    virtual shared_ptr<BaseMatrix> GetMatrixPtr()
+    virtual shared_ptr<BaseMatrix> GetMatrixPtr() override
     {
       if (!jacobi)
         ThrowPreconditionerNotReady();
@@ -216,12 +216,12 @@ namespace ngcomp
     }
 
     ///
-    virtual const BaseMatrix & GetAMatrix() const
+    virtual const BaseMatrix & GetAMatrix() const override
     {
       return bfa->GetMatrix(); 
     }
     ///
-    virtual const char * ClassName() const
+    virtual const char * ClassName() const override
     { return "Local Preconditioner"; }
     void LocPrecTest () const;
   };
@@ -258,29 +258,29 @@ namespace ngcomp
       ; // delete pre;
     }
     
-    virtual void InitLevel (shared_ptr<BitArray> _freedofs);
+    virtual void InitLevel (shared_ptr<BitArray> _freedofs) override;
 
-    virtual void FinalizeLevel (const BaseMatrix *);
+    virtual void FinalizeLevel (const BaseMatrix *) override;
     virtual void AddElementMatrix (FlatArray<int> dnums,
-				   const FlatMatrix<SCAL> & elmat,
+				   FlatMatrix<SCAL> elmat,
 				   ElementId id, 
-				   LocalHeap & lh);
+				   LocalHeap & lh) override;
 
-    virtual void Update ()
+    virtual void Update () override
     {
       if (timestamp < bfa->GetTimeStamp())
         throw Exception("A BDDC preconditioner must be defined before assembling");
     }  
 
-    virtual const BaseMatrix & GetAMatrix() const
+    virtual const BaseMatrix & GetAMatrix() const override
     {
       return bfa->GetMatrix();
     }
 
-    virtual const BaseMatrix & GetMatrix() const;
-    virtual shared_ptr<BaseMatrix> GetMatrixPtr();
+    virtual const BaseMatrix & GetMatrix() const override;
+    virtual shared_ptr<BaseMatrix> GetMatrixPtr() override;
 
-    virtual void CleanUpLevel ()
+    virtual void CleanUpLevel () override
     {
       /*
       delete pre;
@@ -290,10 +290,10 @@ namespace ngcomp
     }
 
 
-    virtual void Mult (const BaseVector & x, BaseVector & y) const;
-    virtual void MultAdd (double s, const BaseVector & x, BaseVector & y) const;
+    virtual void Mult (const BaseVector & x, BaseVector & y) const override;
+    virtual void MultAdd (double s, const BaseVector & x, BaseVector & y) const override;
 
-    virtual const char * ClassName() const
+    virtual const char * ClassName() const override
     { return "BDDC Preconditioner"; }
   };
 
@@ -414,7 +414,7 @@ namespace ngcomp
     ///
     // MGPreconditioner * low_order_preconditioner;
     ///
-    shared_ptr<Preconditioner> coarse_pre;
+    shared_ptr<BaseMatrix> coarse_pre;
     ///
     int finesmoothingsteps;
     ///
@@ -462,7 +462,7 @@ namespace ngcomp
     void MgTest () const;
 
     void SetDirectSolverCluster(shared_ptr<Array<int>> cluster);
-    void SetCoarsePreconditioner(shared_ptr<Preconditioner> prec);
+    void SetCoarsePreconditioner(shared_ptr<BaseMatrix> prec);
   };
 
   class CommutingAMGPreconditioner : public Preconditioner

netgen/include/prolongation.hpp

@@ -67,6 +67,11 @@ namespace ngmg
       y = tmp.Range(0, VWidth());
     }
 
+    shared_ptr<BaseSparseMatrix> CreateSparseMatrix() const override
+    {
+      return prol->CreateProlongationMatrix(level);
+    }
+
     AutoVector CreateRowVector() const override { return make_unique<VVector<double>> (VWidth()); }
     AutoVector CreateColVector() const override { return make_unique<VVector<double>> (VHeight()); }
   };
@@ -219,9 +224,7 @@ namespace ngmg
     }
 
     ///
-    virtual shared_ptr<SparseMatrix< double >> CreateProlongationMatrix( int finelevel ) const override
-    { return NULL; }
-
+    shared_ptr<SparseMatrix< double >> CreateProlongationMatrix( int finelevel ) const override;
 
     ///
     virtual void ProlongateInline (int finelevel, BaseVector & v) const override;
@@ -233,12 +236,14 @@ namespace ngmg
   class NGS_DLL_HEADER HarmonicProlongation : public Prolongation
   {
     shared_ptr<Prolongation> baseprol;
-    shared_ptr<BilinearForm> bfa;
+    weak_ptr<BilinearForm> wbfa;
     Array<shared_ptr<BaseMatrix>> innerinverses;
-    Array<size_t> edges_on_level;
+    Array<size_t> facets_on_level;
+    Array<shared_ptr<BitArray>> freedofs;
+    string inverse;
   public:
     HarmonicProlongation (shared_ptr<Prolongation> abaseprol,
-                          shared_ptr<BilinearForm> abfa);
+                          shared_ptr<BilinearForm> abfa, string ainverse);
 
     virtual void Update (const FESpace & fes) override;
     
@@ -248,6 +253,127 @@ namespace ngmg
     virtual void ProlongateInline (int finelevel, BaseVector & v) const override;
     virtual void RestrictInline (int finelevel, BaseVector & v) const override;
   };
+
+
+
+
+  
+  template <int DIM>
+  class FacetProlongation : public Prolongation
+  {
+  protected:
+    shared_ptr<MeshAccess> ma;
+    
+    int dofs_per_facet = -1;
+    Array<Array<int>> first_dofs;
+    Array<int> prol_class;  // which prol to use ?
+    Array<size_t> facets_on_level;
+
+    static constexpr int NUMPROL = (DIM==2) ? 2 : 32;
+    array<Matrix<double>,NUMPROL> facetprol;
+    
+    bool haveprols = false;
+    
+  public:
+    FacetProlongation (shared_ptr<MeshAccess> ama)
+      : ma(ama) { }
+
+
+    virtual shared_ptr<SparseMatrix< double >> CreateProlongationMatrix( int finelevel ) const override
+    {
+      return nullptr;  // or NULL ? 
+    }
+
+    
+    virtual void ProlongateInline (int finelevel, BaseVector & v) const override
+    {
+      auto fv = v.FV<double>();
+      Matrix<double> tmp(facets_on_level[finelevel], dofs_per_facet);
+      tmp = 0.0;
+      
+      for (size_t i = 0; i < first_dofs[finelevel-1].Size()-1; i++)
+        {
+          IntRange r(first_dofs[finelevel-1][i], first_dofs[finelevel-1][i+1]);
+          if (r.Size() > 0)
+            tmp.Row(i) = fv.Range(r);
+        }
+
+      if constexpr (DIM==2)
+        {
+          for (int i = facets_on_level[finelevel-1]; i < facets_on_level[finelevel]; i++)
+            if (auto parents = get<1>(ma->GetParentEdges(i)); parents[1] == -1)
+              {
+                int pe = parents[0];
+                tmp.Row(i) = facetprol[prol_class[i]] * tmp.Row(pe);
+              }
+        }
+      else
+        {
+          for (int i = facets_on_level[finelevel-1]; i < facets_on_level[finelevel]; i++)
+          if (auto parents = get<1>(ma->GetParentFaces(i)); parents[1] == -1)
+            {
+              int pe = parents[0];
+              tmp.Row(i) = facetprol[prol_class[i]] * tmp.Row(pe);
+            }
+        }
+      
+      for (size_t i = 0; i < first_dofs[finelevel].Size()-1; i++)
+        {
+          IntRange r(first_dofs[finelevel][i], first_dofs[finelevel][i+1]);
+          if (r.Size() > 0)
+            fv.Range(r) = tmp.Row(i);
+        }
+    }
+    
+    virtual void RestrictInline (int finelevel, BaseVector & v) const override
+    {
+      auto fv = v.FV<double>();
+      Matrix<double> tmp(facets_on_level[finelevel], dofs_per_facet);
+      tmp = 0.0;
+      
+      for (size_t i = 0; i < first_dofs[finelevel].Size()-1; i++)
+        {
+          IntRange r(first_dofs[finelevel][i], first_dofs[finelevel][i+1]);
+          if (r.Size() > 0)
+            tmp.Row(i) = fv.Range(r);
+        }
+
+
+      if constexpr (DIM==2)
+        {
+          for (int i = facets_on_level[finelevel-1]; i < facets_on_level[finelevel]; i++)
+            if (auto parents = get<1>(ma->GetParentEdges(i)); parents[1] == -1)
+              {
+                int pe = parents[0];
+                tmp.Row(pe) += Trans(facetprol[prol_class[i]]) * tmp.Row(i);
+              }
+        }
+      else
+        {
+          // for (int i = facets_on_level[finelevel-1]; i < facets_on_level[finelevel]; i++)
+          for (int i = facets_on_level[finelevel]-1; i >= facets_on_level[finelevel-1]; i--)      
+            if (auto parents = get<1>(ma->GetParentFaces(i)); parents[1] == -1)
+              {
+                int pe = parents[0];
+                tmp.Row(pe) += Trans(facetprol[prol_class[i]]) * tmp.Row(i);
+                tmp.Row(i) = 0.0;  // optional, for testing
+              }
+        }
+      
+      fv = 0.0;
+      for (size_t i = 0; i < first_dofs[finelevel-1].Size()-1; i++)
+        {
+          IntRange r(first_dofs[finelevel-1][i], first_dofs[finelevel-1][i+1]);
+          if (r.Size() > 0)
+            fv.Range(r) = tmp.Row(i);
+        }
+
+    }
+    
+  };
+  
+  
+  
 }
 
 

netgen/include/recursive_pol.hpp

@@ -1507,9 +1507,24 @@ namespace ngfem
   class NormalizedLegendreFunctions : public RecursivePolynomialNonStatic<NormalizedLegendreFunctions>
   {
     int  m;
+    static Matrix<double,ColMajor> matA, matB;
   public:
     NormalizedLegendreFunctions (size_t _m) : m(_m) { }
     
+
+    static void Calc(int mmax)
+    {
+      if (matA.Height() >= mmax+1) return;
+      matA.SetSize(mmax+1, mmax+1);
+      matB.SetSize(mmax+1, mmax+1);
+      for (int m = 0; m <= mmax; m++)
+        for (int n = m+2; n <= mmax; n++)
+          {
+            matA(n,m) = (2*n-1) / sqrt((n-m+0.0)*(n+m));
+            matB(n,m) = sqrt((n+m-1.0)*(n-m-1.0)) / sqrt((n-m+0.0)*(n+m));
+          }
+    }
+    
     template <typename T, typename S>
     NormalizedLegendreFunctions (size_t mmax, size_t nmax, T x, S && values)
     {
@@ -1524,23 +1539,60 @@ namespace ngfem
       auto y = Trans(values);
       
       y = 0.0;
-      double u = -sqrt((1-x)*(1+x));
+      T u = -sqrt((1-x)*(1+x));
       y(0,0)=1;
-      
-      for (double m = 0; m <= mmax; m++)
+
+      if (mmax+1 > matA.Height() || nmax > matA.Width())
+        {
+          for (int m = 0; m <= mmax; m++)
+            {
+              if (m > 0)
+                y(m,m)=y(m-1,m-1)*u*sqrt((2*m-1.0)/(2*m));
+              if (m < nmax)
+                y(m+1,m)=x*y(m,m)*sqrt(2*m+1.0);
+              for (int n = m+2; n <= nmax; n++)
+                y(n,m)=((2*n-1)*x*y(n-1,m) - 
+                        sqrt((n+m-1.0)*(n-m-1.0))*y(n-2,m))
+                  /sqrt((n-m+0.0)*(n+m));
+            }
+        }
+      else
         {
-          if (m > 0)
-            y(m,m)=y(m-1,m-1)*u*sqrt((2*m-1.0)/(2*m));
-          if (m < nmax)
-            y(m+1,m)=x*y(m,m)*sqrt(2*m+1.0);
-          for (int n = m+2; n <= nmax; n++)
-            y(n,m)=((2*n-1)*x*y(n-1,m) - 
-                    sqrt((n+m-1.0)*(n-m-1.0))*y(n-2,m))
-              /sqrt((n-m+0.0)*(n+m));
+          for (int m = 0; m <= mmax; m++)
+            {
+              if (m > 0)
+                y(m,m)=y(m-1,m-1)*u*sqrt((2*m-1.0)/(2*m));
+              if (m < nmax)
+                {
+                  T valold = y(m,m);
+                  
+                  T val = x*valold*sqrt(2*m+1.0);
+                  y(m+1,m)=val;
+                  
+                  auto coefsA = matA.Col(m);
+                  auto coefsB = matB.Col(m);
+                  
+                  for (int n = m+2; n <= nmax; n++)
+                    {
+                      T valnew = coefsA(n) * x * val - coefsB(n) * valold;
+                      y(n,m) = valnew;
+                      valold = val;
+                      val = valnew;
+                    }
+                }
+              /*
+              for (int n = m+2; n <= nmax; n++)
+                y(n,m) = matA(n,m) * x * y(n-1,m) - matB(n,m) * y(n-2,m);
+              */
+            }
         }
+      
       for (int n = 0; n <= nmax; n++)
+        y.Row(n).Range(0,n+1) *= sqrt(2*n+1.0);
+        /*
         for (int m = 0; m <= n; m++)
           y(n,m)=y(n,m)*sqrt(2*n+1.0);
+        */
     }
     
     template <class S>

netgen/include/simd_complex.hpp

@@ -56,6 +56,17 @@ namespace ngcore
     auto & imag() { return im; }
 
 
+    auto Lo() const
+    {
+      if constexpr (N == 2) return Complex(re.Lo(), im.Lo());
+      else return SIMD<Complex,N/2> (re.Lo(), im.Lo());
+    }
+    auto Hi() const
+    {
+      if constexpr (N == 2) return Complex(re.Hi(), im.Hi());
+      else return SIMD<Complex,N/2> (re.Hi(), im.Hi());
+    }
+    
     // Numbers in SIMD structure are not necessarily in same order as in memory
     // for instance:
     // [x0,y0,x1,y1,x2,y2,x3,y3] -> [x0,x2,x1,x3,y0,y2,y1,y3]
@@ -266,6 +277,40 @@ namespace ngcore
     return SIMD<Complex, N> (IfPos (a.real(), b.real(), c.real()),
                              IfPos (a.real(), b.imag(), c.imag()));
   }
+
+
+  template <int N>
+  INLINE SIMD<Complex, N> If (SIMD<mask64, N> a, SIMD<Complex, N> b, SIMD<Complex, N> c)
+  {
+    return SIMD<Complex, N> (If (a, b.real(), c.real()),
+                             If (a, b.imag(), c.imag()));
+  }
+
+
+
+  
+  template <typename T, size_t S> class MakeSimdCl;
+  
+  template <size_t S>
+  class MakeSimdCl<Complex,S>
+  {
+    std::array<Complex,S> a;
+  public:
+    MakeSimdCl (std::array<Complex,S> aa) : a(aa)  { ; }
+    auto Get() const
+    {
+      std::array<double,S> ar, ai;
+      for (int j = 0; j < S; j++)
+        {
+          ar[j] = ngbla::Real(a[j]);
+          ai[j] = ngbla::Imag(a[j]);
+        }
+      
+      return SIMD<Complex,S> (MakeSimd(ar), MakeSimd(ai));
+    }
+  };
+  
+  
 }
 
 

netgen/include/sparsecholesky.hpp

@@ -28,10 +28,14 @@ namespace ngla
 
   public:
     SparseFactorization (shared_ptr<const BaseSparseMatrix> amatrix,
-			 shared_ptr<BitArray> ainner,
-			 shared_ptr<const Array<int>> acluster);
+			 shared_ptr<BitArray> ainner = nullptr,
+			 shared_ptr<const Array<int>> acluster = nullptr);
     SparseFactorization() {}
 
+    virtual void SetSubset
+      (shared_ptr<BitArray> ainner,
+        shared_ptr<const Array<int>> acluster=nullptr);
+
     void DoArchive(Archive& ar) override;
 
     // virtual bool IsComplex() const override { return matrix.lock()->IsComplex(); }