Modification of the large size of the iPhone on the GPU

Question

I developed a training AI for use in an iPad game, but its prototype is written in Matlab. I need to perform several actions, such as: (capital denotes a 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). Performing these operations on the GPU can significantly improve the performance of this operation. Is there a linear algebra library more suitable than BLAS acceleration? BLAS only supports up to 4x4 matrices.

Answer 1

We had good results in a previous project using the Eigen library.

• This is a C ++ template library containing only header files, but this is not a problem, since you can write your interfaces in either simple C or Objective-C ++ (.mm).
• Its code is highly optimized for armv7.
• The LGPL3 licensing model should be a pretty good choice for releasing apps in the app store without interfering with Apple TOS.