The CUDA equivalents for coulomb_tab_climg2D, nbfp_climg2D, nbfp_comb_climg2D are textures.
The OpenCL implementation uses normal memory buffers for these data arrays instead of textures because of some issues on AMD cards (has to be retested)
What needs to be done:
- recheck if using textures produces incorrect results on AMD cards
- evaluate the performance difference between using textures and normal buffers
- if necessary, switch back to using textures for coulomb_tab_climg2D, nbfp_climg2D, nbfp_comb_climg2D