-
* wrapped all CUDA memory functions in safety calls * fixed CUDALBP invalid memory access which caused performance and stability issues * carryover support for all plugins, only copy over memory once throughout the whole computation
-
was a index error (didn't run enough threads)
-
saves a about 10% on performance
-
produced more of better results
-
doesn't make things too much faster...
-
currently uses CPU code from plugins/core but will start to change things one by one
-
* copyto and copyfrom need work
-
EvalDetection change.
