Q: Is there anything special needed to do/know when recompiling kernels from CUDA 10.1 over to 12.1?
Or should it be a simple tool-chain swap out?
we’re sorting through some non-obvious linear algebra and trig math issues after we simply recompiled on 12.1. It’s all written in straight C and the Eigen header-only vector lib - so started to wonder if there is a porting guide that we might have missed. We are sorting through the CPU side as well, so no specifics worth delving into just yet. General question is just - is there anything to know about moving .cu s to 12.
Thanks & Best Regards…
One consideration, if you’ve not already seen it, is this comment regarding thread scheduling, if you’ve previously been developing for SM < 7.