From fe48c25682fd8dcbcdf8fd19db64e69d6154b510 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Tue, 6 Dec 2016 15:04:50 +0100
Subject: [PATCH] Revert vec/y to vec*(1/y) in row-major TRSM: - div is
 extremely costly - this is consistent with the column-major case - this is
 consistent with all other BLAS implementations (grafted from
 eb621413c154b79e0acab3fa0c617964aef11d34 )

---
 Eigen/src/Core/products/TriangularSolverMatrix.h | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/Eigen/src/Core/products/TriangularSolverMatrix.h b/Eigen/src/Core/products/TriangularSolverMatrix.h
index 1bed66ed8..223c38b86 100644
--- a/Eigen/src/Core/products/TriangularSolverMatrix.h
+++ b/Eigen/src/Core/products/TriangularSolverMatrix.h
@@ -183,7 +183,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheLeft,Mode,Conju
     }
   }
 
-/* Optimized triangular solver with multiple left hand sides and the trinagular matrix on the right
+/* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
  */
 template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
 struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
@@ -202,6 +202,7 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index rows = otherSize;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
 
     typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
     typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
@@ -306,9 +307,9 @@ EIGEN_DONT_INLINE void triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conj
               }
               if((Mode & UnitDiag)==0)
               {
-                Scalar b = conj(rhs(j,j));
+                Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
                 for (Index i=0; i<actual_mc; ++i)
-                  r[i] /= b;
+                  r[i] *= inv_rjj;
               }
             }