Greatly increased performance by: selectively recreating buffers only when needed setting apodisation data to constant graphic card memory (on nvidia GPUs this is faster, but rather small) building the openCL code on startup of the module

some new instability has to be fixed, though