Re: [eigen] Optimization advice for a specific expression

[Alberto Luaces <alberto.luaces@xxxxxxxxx>]

Gael Guennebaud writes:

> Hi,
>
> you can avoid one temp with:
>
> I4x4.noalias() += ...;
>

Ok.  Nevertheless, there is no big difference.

>
> Then slowndown might come from the compiler doing a bad job at
> inlining. What is your compiler?

gcc version 5.3.1 20160121 (Debian 5.3.1-7) 

> version?

libeigen3-dev  3.3~beta1-2

> What are the exact compilation flags?
>

-msse4 -O3 -DNDEBUG   -std=gnu++11

....aside from -I preprocessor flags.  "-msse4" seems to make no
difference either.

>
> Also, please paste the complete declaration of
> call_dense_assignment_loop.
>

Eigen::internal::call_dense_assignment_loop<Eigen::Matrix<double, 4, 4, 0, 4, 4>, Eigen::Product<Eigen::Product<Eigen::Matrix<double, 4, 4, 0, 4, 4>, Eigen::Matrix<double, 4, 4, 0, 4, 4>, 0>, Eigen::Matrix<double, 4, 4, 0, 4, 4>, 1>, Eigen::internal::assign_op<double> >

Thanks!