Make products always eval into expressions. Improves performance

in benchmark. Still not as fasts as explicit eval(), strangely.

3 files changed, 3 insertions, 2 deletions

diff --git a/Eigen/src/Core/Assign.h b/Eigen/src/Core/Assign.h
index 362e3e9e7..3862ae831 100644
--- a/Eigen/src/Core/Assign.h
+++ b/Eigen/src/Core/Assign.h
@@ -114,6 +114,7 @@ Derived& MatrixBase<Derived>
 {
 //   std::cout << typeid(OtherDerived).name() << "\n";
 //   std::cout << "lazyAssign = " << (Derived::Flags&VectorizableBit) << " " << (OtherDerived::Flags&VectorizableBit) << "\n";
+//  std::cout << __PRETTY_FUNCTION__ << std::endl;
   ei_assignment_impl<Derived, OtherDerived>::execute(derived(),other.derived());
   return derived();
 }
diff --git a/Eigen/src/Core/Product.h b/Eigen/src/Core/Product.h
index 86f37ec43..cdb4e88b4 100644
--- a/Eigen/src/Core/Product.h
+++ b/Eigen/src/Core/Product.h
@@ -141,7 +141,7 @@ struct ei_traits<Product<Lhs, Rhs, EvalMode> >
               | ((RowsAtCompileTime == Dynamic || ColsAtCompileTime == Dynamic) ? 0 : LargeBit)),
     Flags = ((unsigned int)(LhsFlags | RhsFlags) & _LostBits)
           | EvalBeforeAssigningBit
-          | ((int)EvalMode == (int)CacheOptimalProduct ? EvalBeforeNestingBit : 0)
+          | EvalBeforeNestingBit
           | (_Vectorizable ? VectorizableBit : 0),
     CoeffReadCost
       = Lhs::ColsAtCompileTime == Dynamic
diff --git a/bench/benchmark.cpp b/bench/benchmark.cpp
index 53693cbf6..4061b0211 100644
--- a/bench/benchmark.cpp
+++ b/bench/benchmark.cpp
@@ -39,7 +39,7 @@ int main(int argc, char *argv[])
     asm("#begin");
     for(int a = 0; a < REPEAT; a++)
     {
-        m = I + 0.00005 * (m + m*m);
+        m = Matrix<SCALAR,MATSIZE,MATSIZE>::ones() + 0.00005 * (m + m*m);
     }
     asm("#end");
     cout << m << endl;