finally here is a simple solution making (a*b).diagonal() even faster than a.lazyProduct(b).diagonal() !!

1 files changed, 16 insertions, 0 deletions

diff --git a/Eigen/src/Core/products/CoeffBasedProduct.h b/Eigen/src/Core/products/CoeffBasedProduct.h
index c9d333072..f030d59b5 100644
--- a/Eigen/src/Core/products/CoeffBasedProduct.h
+++ b/Eigen/src/Core/products/CoeffBasedProduct.h
@@ -127,8 +127,14 @@ class CoeffBasedProduct
                                   Unroll ? InnerSize-1 : Dynamic,
                                   _LhsNested, _RhsNested, Scalar> ScalarCoeffImpl;
 
+    typedef CoeffBasedProduct<LhsNested,RhsNested,NestByRefBit> LazyCoeffBasedProductType;
+
   public:
 
+    inline CoeffBasedProduct(const CoeffBasedProduct& other)
+      : Base(), m_lhs(other.m_lhs), m_rhs(other.m_rhs)
+    {}
+
     template<typename Lhs, typename Rhs>
     inline CoeffBasedProduct(const Lhs& lhs, const Rhs& rhs)
       : m_lhs(lhs), m_rhs(rhs)
@@ -185,6 +191,16 @@ class CoeffBasedProduct
     const _LhsNested& lhs() const { return m_lhs; }
     const _RhsNested& rhs() const { return m_rhs; }
 
+    const Diagonal<LazyCoeffBasedProductType,0> diagonal() const
+    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
+
+    template<int Index>
+    const Diagonal<LazyCoeffBasedProductType,Index> diagonal() const
+    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
+
+    const Diagonal<LazyCoeffBasedProductType,Dynamic> diagonal(int index) const
+    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this).diagonal(index); }
+
   protected:
     const LhsNested m_lhs;
     const RhsNested m_rhs;