Subject: Re: [ublas] eigen Vs. ublas
From: Umut Tabak (u.tabak_at_[hidden])
Date: 2011-04-11 08:17:17
On 04/11/2011 01:42 PM, Nasos Iliopoulos wrote:
> Umut
> a partial answer to the items of your question:
>
> The main performance difference between eigen and ublas is that eigen
> performs explicit vectorization: the operations are put on special
> registers that allow for the execution of multiple instructions per
> cycle (SIMD).
>
> Modern cpus have 128 bits per register, allowing 4 floats or 2 doubles
> to be stored (and operated upon) simultaneously. This means that the
> performance gain you have for double precision (If you are using a
> linear algebra system for anything other than programming graphics,
> probably you want to be using double precision), is at most about 2.
> That is what is expected from eigen over uBlas at most, but this is
> not the full story.
Hi Nasos,
Thanks for the detailed explanation, very nice pointer to gcc pages
>
>
> Please also be aware that there are things that can be done to improve
> the performance of certain uBlas algorithms that are not optimal at
> the moment.
>
What are they, could you share them with me as far as you know. For the
moment, what I would like to do is to test my prototype codes, these are
basically in MATLAB or Octave, in a fast language, where I can check the
real cpu time on large models, I have had experience with uBlas and
digged a lot through the mailing list to find some important
undocumented information, I would not like to do the same with eigen,
that is the reason I am cautious on the topic. As far as I can see
sparse module of eigen is also not that mature and you even have to
define a preprocessor macro to use the sparse module to circumvent this
problem.
I am guessing that ublas is not that drastically slow in comparison to
eigen for, especially large matrix sizes, because the benchmarks on
eigen page are for relatively lower size of matrix-vector products,
however I will give a test myself shortly.
Thanks and best regards,
Umut