CUDA: better error for FA kernel with 0 occupancy#16643

JohannesGaessler wants to merge 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

JohannesGaessler:cuda-fa-better-errorJohannesGaessler/llama.cpp:cuda-fa-better-errorCopy head branch name to clipboard