Question

I’ve designed a learning AI for use in a iPad game but its prototype is written in matlab. I need to perform several actions such as: (capital denotes matrix)

A = B > c;
A = B * C;
A = B' * C;
A = B .* C;
A = B - C;

And some of my matrices are large (2601 x 100). Running these operations on a GPU could significantly improve the performance of this operation. Is there a Linear Algebra library more appropriate than Accelerate’s BLAS? BLAS only supports up to 4×4 matrices.

Answer 1

We had pretty good results in a former project using the (open source) Eigen library.

• It’s a C++ template library containing header files only, but this is no problem as you can write your interfaces either in plain C or Objective-C++ (.mm).
• Its code is heavily optimised for armv7.
• The LGPL3 licensing model should be a pretty good choice for releasing apps on the app-store without interfering with Apple’s TOS.