We have EigenuBLASAtlas Are there more? Is this all there is?

HTTP/1.1 303 See OtherLocation: http://www.wilmott.com/messageview.cfm? ... adid=87551

BTW, this looks intriguing:http://cppnow.org/session/building-fini ... ract.pdf// upcoming talk (from the C++Now 2013 Schedule), here's more: http://cppnow.org/schedule-table/

QuoteOriginally posted by: PolterBTW, this looks intriguing:http://cppnow.org/session/building-fini ... ract.pdf// upcoming talk (from the C++Now 2013 Schedule), here's more: http://cppnow.org/schedule-table/Looks nice but the percentage hand-crafted/generated code is still high. A FEM engineer might just compute mass and stiffness matrices by hand - a tedious process, admittedly - before spending time learning a DSL? Does it scale? I miss the design 'vision' and how would they do DSL/Proto for one-factor time-dependent problems? Not easy IMO.But the idea is nice. What I like is that they are able to use Eigen from the DSL code and is a very good start.// The real question is how to get a 'handle' on this way of thinking?

Regarding Eigen:It's very easy to use and install (header only) , the syntax is clear (like Fortran) and it is certainly worth looking at IMO.

As an aside for any MATLAB programmers out there, Eigen has syntax that allow performing an operation along all the rows or columns of a matrix just like MATLAB. Many of the vectorization tricks that are part of MATLAB transfer well to Eigen.

Armadillo? http://arma.sourceforge.net/

QuoteOriginally posted by: StaleArmadillo? http://arma.sourceforge.net/Looks like easy-to-use syntax. The basic functionality looks simillar to Boost uBLAS. Row and column iterators are nice looking.Any experience of linking up to LAPACK/ATLAS? (I tried ATLAS in Windows but it was a howler).

On a related topic nice to have1. a minimalist matrix libary with 20% of what is really needed. For MSc students and when you just want to test something AKA "I don't want the full kitchen sink". Later than you can use the joys of ATLAS and LAPACK.2. a related libary to serve as foundation for linear and nonlinear optimisation PLUS a nice way to model vector-* functions (alglib is a wee bit obtuse for me).

QuoteIn practice it's so much easier to use Excel or R when you need to offer a quick solution (coding time) where a factor 10 in performance is completely irrelevant.Yes, this is possible. But it represents one scenario of how developers work and it implies twice the effort (do once on Excel/Malab, then in C++). And it means non-potable software.But why do we need 10 matrix libraries in C++? Fortran never had this problem. C++ 11 could have done a bit it to stabilise things. All that matrix stuff is easy and many of the 10 libraries stop half-way. These days is CS undergrad stuff. I bet there are very few libraries that deal with the issues in my previous post (alglib is the only one I know). Many of these libraries have user-unfriendly install.As I mentioned, students create little matrix libraries them in order to learn even if there are 10 libraries out there.

One issue is if you write software that works with library A and you want it then to with libary B you have to copy the code and modify it, yes?

BTW there is a new Boost ODE package. I see that some solvers (e.g. implicit Euler) use uBlAS LU decomposition which is incorrectly implemented big time and is awful slow. This should be a bug report to Boost. BTW his do the 'masses' (like myself) report on these issues in an effective way to the Boost inner group? Here's problem descriptionA nuance is to logically separate1. Martix data structures2. Algos that use 1.Better than saying that the "combined" (1+2) library is slow.