Re: [eigen] great performance with eigen 3.3

[Gael Guennebaud <gael.guennebaud@xxxxxxxxx>]

Hi,

that's good to know! thanks for sharing. 

I guess that your doing a lot of very small dense products using "Dynamic" size matrices: this use case has been indeed optimized. If that's the case, then you might consider using fixed-size matrices (if possible), and do not forget the .noalias() if there is indeed no aliasing:

A.noalias() += B*C.

Another explanation would be if you are using SparseMatrix::coeffRef/insert without a proper preallocation. This use case has been optimized too.

gael

On Tue, Feb 9, 2016 at 6:02 PM, Lorenzo Alessio Botti <bottilorenzo@xxxxxxxxx> wrote:

Just to let you know that I've recently given a shot to eigen 3.3 and it works great (as compared to 3.2).
I have an old 2010 macbook pro (so no avx nor fma) but my finite element assembly and solution routines are almost twice as fast.
I use a lot of matrix products, outer products (often with constant expressions and outer of vectors) and += assignments.
Great job!
Thanks to the community.
Bests
Lorenzo