Divide vector instead of matrix saves time

Now for homogeneous

src/local_parametrization_homogeneous.cpp

@@ -60,11 +60,11 @@ bool LocalParametrizationHomogeneous::computeJacobian(const Eigen::Map<const Eig
     assert(_h.size() == global_size_ && "Wrong size of input quaternion.");
     assert(_jacobian.rows() == global_size_ && _jacobian.cols() == local_size_ && "Wrong size of Jacobian matrix.");
 
-    _jacobian <<  _h(3),  _h(2), -_h(1),
-                 -_h(2),  _h(3),  _h(0),
-                  _h(1), -_h(0),  _h(3),
-                 -_h(0), -_h(1), -_h(2) ;
-    _jacobian /= 2;
+    Eigen::Vector4s hh = _h/2;
+    _jacobian <<  hh(3),  hh(2), -hh(1),
+                 -hh(2),  hh(3),  hh(0),
+                  hh(1), -hh(0),  hh(3),
+                 -hh(0), -hh(1), -hh(2) ;
     return true;
 }
 

[Jeremie Deray]

@jsola In the same spirit multiplying by 0.5 in faster than dividing by 2 ;)

[Joan Solà Ortega]

Isn't the compiler taking care of these smallities?

[Jeremie Deray]

Only if optimization is enabled (-Ox) modern compilers will do . Was just a side note.

[Joan Solà Ortega]

OK