Is it expected for production grade CUDA code to have error checking around all of the GPU operations like cudaMalloc
? Or is it more idiomatic to remove the checks once you have checked that the operations are correct.
Also when error checking in CUDA 9+, should one do something along the lines of this stackoverflow answer - What is the canonical way to check for errors using the CUDA runtime API? - Stack Overflow ?(asking since it is a rather old answer)