Re: [eigen] eigen sqrt compilation error

[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]

May I suggest Eigen/Eigen include ALL functionality that is considered stable and part of the "official" Eigen. I think speeding up compile times is less important than making an easy to use interface. If a developer is concerned with their compile times, they can notice an entry in the FAQ that tells them how to speed things up (selecting a subset of headers, precompiled header options?...) .
Caveat: I'm an Eigen newcomer and so I am unencumbered by knowledge of the specific issues :) 

-- Mark



Benoit Jacob wrote:

Finally, I added a Eigen/Dense header, including all the dense
functionality, but not Sparse, and not the workaround files
QtAlignedMalloc, StdVector.

Rationale:
- this is only 15% slower than Core on my test, while it becomes 35%
slower than Core if I also include Sparse.
- we can still offer Eigen/Eigen alongside if you still think it's
needed. It would include Dense and Sparse, but still not the
workaround files. For that reason it's a bit misleading : the name
Eigen/Eigen suggests that all is included. Tell us if you still want
Eigen/Eigen.

Benoit

2009/6/19 Rohit Garg <rpg.314@xxxxxxxxx>

PLEASE include the nuclear option. I like that.

On Fri, Jun 19, 2009 at 6:13 AM, Mark Borgerding<mark@xxxxxxxxxxxxxx> wrote:

I like this.
* easy to get everything from Eigen you could possibly want with one include
* possible to go through the trouble of minimizing compile time or source
file dependencies.
It seems like a good philosophy.

Guido Von Rossum is fond of saying "Make the easy things easy. Make the hard
things possible."
-- Mark


Benoit Jacob wrote:

oh and there is another "nuclear option":

add a convenience #include<Eigen/Eigen> that includes them all. And
then keep the separate headers for the users who want every % of
compilation speed.

The idea is that Core and Array are the 2 biggest modules, so
Eigen/Eigen won't be much slower than including them two.

Benoit

2009/6/19 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>


2009/6/19 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>


On Fri, Jun 19, 2009 at 12:29 PM, Patrick Mihelich
<patrick.mihelich@xxxxxxxxx> wrote:


On Fri, Jun 19, 2009 at 12:37 AM, Moritz Lenz
<mlenz@xxxxxxxxxxxxxxxxxxxxxxx> wrote:


Considering the number of times people make this error (forgetting to
include it), what about including it by default? Is there a very good reason
not to do it?


I wonder this too. I ran into this error when I started using Eigen,
and it seems to pop up regularly on the list. What exactly is the cost of
including <Eigen/Array> in the core? How much does it increase compile
times?


on my high end computer with gcc 4.3.2 -O2 -DNDEBUG, much less time
than to compile: Vector3f a, b; cout << a+b;, so something around 0.02
sec.


On my computer I tried this program:

#include <Eigen/Core>
//#include <Eigen/Array>

using namespace Eigen;

int main() {
 Matrix3d m = Matrix3d::Identity();
 Matrix3d n = m+m;
 Vector3d v; v << 1, 2, 3;
 Vector3d w = v+v;
 w = m * w;
 std::cout << w << std::endl;
}

with g++ 4.3.3. I repeated the compilation 10 times.

The #include<Eigen/Array> increases compilation times by 2.5% and the
memory usage is not affected.

I'm OK to consider this reasonable, especially as other modules like
Geometry already include Array.
What's your opinion?

The next question would be: should we merge the Array/ files into
Core/ or just #include<Array> in Core.

Benoit








--
Rohit Garg

http://rpg-314.blogspot.com/

Senior Undergraduate
Department of Physics
Indian Institute of Technology
Bombay

Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/