merge with default branch

3 files changed, 28 insertions, 22 deletions

diff --git a/Eigen/src/Core/products/CoeffBasedProduct.h b/Eigen/src/Core/products/CoeffBasedProduct.h
index 3343b1875..e8016e915 100644
--- a/Eigen/src/Core/products/CoeffBasedProduct.h
+++ b/Eigen/src/Core/products/CoeffBasedProduct.h
@@ -305,10 +305,7 @@ struct ei_product_coeff_vectorized_dyn_selector
 {
   EIGEN_STRONG_INLINE static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
   {
-    res = ei_dot_impl<
-      Block<Lhs, 1, ei_traits<Lhs>::ColsAtCompileTime>,
-      Block<Rhs, ei_traits<Rhs>::RowsAtCompileTime, 1>,
-      LinearVectorizedTraversal, NoUnrolling>::run(lhs.row(row), rhs.col(col));
+    res = lhs.row(row).cwiseProduct(rhs.col(col)).sum();
   }
 };
 
@@ -319,10 +316,7 @@ struct ei_product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,RhsCols>
 {
   EIGEN_STRONG_INLINE static void run(int /*row*/, int col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
   {
-    res = ei_dot_impl<
-      Lhs,
-      Block<Rhs, ei_traits<Rhs>::RowsAtCompileTime, 1>,
-      LinearVectorizedTraversal, NoUnrolling>::run(lhs, rhs.col(col));
+    res = lhs.cwiseProduct(rhs.col(col)).sum();
   }
 };
 
@@ -331,10 +325,7 @@ struct ei_product_coeff_vectorized_dyn_selector<Lhs,Rhs,LhsRows,1>
 {
   EIGEN_STRONG_INLINE static void run(int row, int /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
   {
-    res = ei_dot_impl<
-      Block<Lhs, 1, ei_traits<Lhs>::ColsAtCompileTime>,
-      Rhs,
-      LinearVectorizedTraversal, NoUnrolling>::run(lhs.row(row), rhs);
+    res = lhs.row(row).cwiseProduct(rhs).sum();
   }
 };
 
@@ -343,10 +334,7 @@ struct ei_product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,1>
 {
   EIGEN_STRONG_INLINE static void run(int /*row*/, int /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
   {
-    res = ei_dot_impl<
-      Lhs,
-      Rhs,
-      LinearVectorizedTraversal, NoUnrolling>::run(lhs, rhs);
+    res = lhs.cwiseProduct(rhs).sum();
   }
 };
 
diff --git a/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
index 6836a10de..8ac5afb05 100644
--- a/Eigen/src/Core/products/GeneralBlockPanelKernel.h
+++ b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -31,6 +31,7 @@
 #define CJMADD(A,B,C,T)  C = cj.pmadd(A,B,C);
 #else
 #define CJMADD(A,B,C,T)  T = B; T = cj.pmul(A,T); C = ei_padd(C,T);
+// #define CJMADD(A,B,C,T)  T = A; T = cj.pmul(T,B); C = ei_padd(C,T);
 #endif
 
 // optimized GEneral packed Block * packed Panel product kernel
@@ -146,7 +147,7 @@ struct ei_gebp_kernel
         #endif
 
         // performs "inner" product
-        // TODO let's check wether the flowing peeled loop could not be
+        // TODO let's check wether the folowing peeled loop could not be
         //      optimized via optimal prefetching from one loop to the other
         const Scalar* blB = unpackedB;
         for(int k=0; k<peeled_kc; k+=4)
@@ -409,6 +410,7 @@ struct ei_gebp_kernel
             CJMADD(A0,B2,C2,B2);
             B2 = ei_pload(&blB[14*PacketSize]);
             CJMADD(A0,B3,C3,B3);
+
             A0 = ei_pload(&blA[3*PacketSize]);
             B3 = ei_pload(&blB[15*PacketSize]);
             CJMADD(A0,B0,C0,B0);
diff --git a/Eigen/src/Core/products/SelfadjointMatrixMatrix.h b/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
index 785045db4..2e71b5fd4 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
@@ -43,7 +43,10 @@ struct ei_symm_pack_lhs
     {
       for(int w=0; w<h; w++)
         blockA[count++] = ei_conj(lhs(k, i+w)); // transposed
-      for(int w=h; w<BlockRows; w++)
+
+      blockA[count++] = ei_real(lhs(k,k));   // real (diagonal)
+
+      for(int w=h+1; w<BlockRows; w++)
         blockA[count++] = lhs(i+w, k);          // normal
       ++h;
     }
@@ -71,8 +74,11 @@ struct ei_symm_pack_lhs
     // do the same with mr==1
     for(int i=peeled_mc; i<rows; i++)
     {
-      for(int k=0; k<=i; k++)
+      for(int k=0; k<i; k++)
         blockA[count++] = lhs(i, k);              // normal
+
+      blockA[count++] = ei_real(lhs(i, i));       // real (diagonal)
+
       for(int k=i+1; k<cols; k++)
         blockA[count++] = ei_conj(lhs(k, i));     // transposed
     }
@@ -129,8 +135,11 @@ struct ei_symm_pack_rhs
         // normal
         for (int w=0 ; w<h; ++w)
           blockB[count+w] = alpha*rhs(k,j2+w);
+
+        blockB[count+h] = alpha*rhs(k,k);
+
         // transpose
-        for (int w=h ; w<nr; ++w)
+        for (int w=h+1 ; w<nr; ++w)
           blockB[count+w] = alpha*ei_conj(rhs(j2+w,k));
         count += nr;
         ++h;
@@ -175,8 +184,15 @@ struct ei_symm_pack_rhs
         blockB[count] = alpha*ei_conj(rhs(j2,k));
         count += 1;
       }
+
+      if(half==j2)
+      {
+        blockB[count] = alpha*ei_real(rhs(j2,j2));
+        count += 1;
+      }
+
       // normal
-      for(int k=half; k<k2+rows; k++)
+      for(int k=half+1; k<k2+rows; k++)
       {
         blockB[count] = alpha*rhs(k,j2);
         count += 1;
@@ -389,7 +405,7 @@ struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>
                                * RhsBlasTraits::extractScalarFactor(m_rhs);
 
     ei_product_selfadjoint_matrix<Scalar,
-      EIGEN_LOGICAL_XOR(LhsIsUpper, 
+      EIGEN_LOGICAL_XOR(LhsIsUpper,
                         ei_traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
       NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsIsUpper,bool(LhsBlasTraits::NeedToConjugate)),
       EIGEN_LOGICAL_XOR(RhsIsUpper,