Generally, optimization options would only change when there's a substantial departure from an older microarchitecture, be it a new instruction set or a latency/throughput change. Cannon Lake introduces the AVX512 extension, while Kaby Lake just isn't different enough from, I guess, Skylake, to get a different set of optimization parameters.
You may ask the compiler itself for its opinion on the native architecture with `gcc -Q -march=native --help=target | grep march`.