I noticed in several code examples (convolutionFFT2D), lot’s of CUDA functions are wrapped with CUDA_SAFE_CALL, CUT_SAFE_CALL and CUFFT_SAFE_CALL. The code segments in the manual pages make no mention of these wrappers (for example, the CUFFT code examples don’t wrap the malloc or cufftExec calls). I assume they have to do with synchronization, but I might be mistaken. What are these wrappers for and when can I avoid/not use them?
Thanks in advance for any advice.