cuda-mat

Contains some matrix transformation functions writed on cuda.

Why not opencv::cuda?

opencv::cuda has simple and fast implementation, but there are several problems:

  1. Not thread safe - some transformation functions from opencv::cuda uses gpu constant memory, so you can't use same function in other thread without blocking

  2. There are problem with signed integer overflow, see this issue

TODO

  1. Add algorithm for inversing 3x3 transformation matricies and remove opencv dependency