Hi,

I'm currently developing some linear algebra routine that can run on hybrid hardware (ie CPU and GPU) using cublas, blas and lapack ; the goal is to provide some CUDA feature to Scilab.

I didnt know about the existence of eigen until recently, but I'm thinking of switching to eigen instead of blas, because of performance and of code clarity (I'm using c++ features such as templating, functions encapsulation...a lot).

However, I think it would be nice if I could "insert" my work in a more complete framework, in order to benefit from code conduct (I don't have true stable code writing convention for now...), experience from other devs, and so on.

Is there any possibility to contribute a little to the coding of eigen, to provide some CUDA-accelerated routines ?

(and is there a compatibility layer "à la cvs-git" between git and mercurial ?)

Thanks, Vincent.

vlj wrote:Hi,

I'm currently developing some linear algebra routine that can run on hybrid hardware (ie CPU and GPU) using cublas, blas and lapack ; the goal is to provide some CUDA feature to Scilab.

Interesting. Also note that a GSOC project just started to port a part of Scilab to Eigen...

I didnt know about the existence of eigen until recently, but I'm thinking of switching to eigen instead of blas, because of performance and of code clarity (I'm using c++ features such as templating, functions encapsulation...a lot).

I think that's a good idea

However, I think it would be nice if I could "insert" my work in a more complete framework, in order to benefit from code conduct (I don't have true stable code writing convention for now...), experience from other devs, and so on.

Is there any possibility to contribute a little to the coding of eigen, to provide some CUDA-accelerated routines ?

You are very welcome to do so! We need contributors! However, supporting GPU involves non trivial changes and many questions regarding the API. So the best is to start by creating a fork on bitbucket where you will commit your changes. I also strongly recommend you to join the mailing list to discuss your issues, to keep us uptodate on your recent changes, etc. Once the fork will be mature enough we could merge it in the official devel branch.

(and is there a compatibility layer "à la cvs-git" between git and mercurial ?)

I don't know.

If you are ok, I recommend you to continue this discussion on the ML which is a more appropriate place for this kind of discussion, and you will also reach a larger set of people for feedbacks.