Raising Matrices to larger powers using CUDA and exponentiating by squaring

Added a simple project to GitHub which uses Dgemm to perform the exponentiating by squaring algorithm on square Matrices.


I did this mainly to test the double precision of the K20c.

The project compute capability is still set to 3.0 rather than 3.5, will this make a noticeable difference in performance? It is working well as it is, but just wondering if cuBLAS will automatically determine this on its own.