Fix cost evaluation of partial reduxions -> improve performance of vectorwise/replicate expressions involving partial reduxions

1 files changed, 10 insertions, 0 deletions

diff --git a/test/vectorwiseop.cpp b/test/vectorwiseop.cpp
index 9d60b0388..6cd1acdda 100644
--- a/test/vectorwiseop.cpp
+++ b/test/vectorwiseop.cpp
@@ -101,6 +101,16 @@ template<typename ArrayType> void vectorwiseop_array(const ArrayType& m)
 
   VERIFY_RAISES_ASSERT(m2.rowwise() /= rowvec.transpose());
   VERIFY_RAISES_ASSERT(m1.rowwise() / rowvec.transpose());
+  
+  m2 = m1;
+  // yes, there might be an aliasing issue there but ".rowwise() /="
+  // is suppposed to evaluate " m2.colwise().sum()" into to temporary to avoid
+  // evaluating the reducions multiple times
+  if(ArrayType::RowsAtCompileTime>2 || ArrayType::RowsAtCompileTime==Dynamic)
+  {
+    m2.rowwise() /= m2.colwise().sum();
+    VERIFY_IS_APPROX(m2, m1.rowwise() / m1.colwise().sum());
+  }
 }
 
 template<typename MatrixType> void vectorwiseop_matrix(const MatrixType& m)