Yes, there now are several construction sites (or things that should be reviewed, at least) :
- Replace exception tests after Get*Critical calls with NULL checks, to prevent (unnecessary) warnings
- Clean up pointer data handling, particularly PointersArrayPointerData
- Consider “deferring” the Get*Critical calls as long as possible
- Consider a dedicated utility class for the kernel arguments (in any case: In addition to the one that mimics the original API)
- Consider a special treatment of the kernelArgs-pointer-to-pointer
I’m not sure when I can tackle each of these points, and the progress partially depends on schedules that I can not influence (CUDA 6.0 will come soon), but I’ll try to allocate some time for that.