problems an limitations

- Each tile has a ChildTile cache to perform the optimization process
without changing its own value.

- Forgot to initialize CUDA random generators

optimization computations

- Running out of resources locally

- Added some cache to handle inter-molecule atom-distances pairs

- Optimization still requires some additions
- Check if output is okay-ish

- Since original UEGO algorithm uses dynamic memory allocations during
its execution, the CUDA version must pre-allocate all required device
memory beforehand, so the original algorithm ideas remains the same.
- For that, the allocations have been changed into a 'tile' system.
- Each 'tile' has a buffer to store atoms rotations and such.

- 'Pure' computational functions (VolumeOverlap, Tanimoto, ...) are yet to be implemented.

Jerónimo Sánchez authored 3 years ago and Jerónimo Sánchez committed 3 years ago

- This eases kernel creation and executions