You can always make sure they are not optimized to zero, by giving them a side effect that the compiler is forced to honor.

But C delays not circling around a hardware-based timing construct should be avoided at any time where there is a hard requirement that a delay must have a minimum (or maximum) time. Withour making use of some known hardware for the timing, the C code will just result in a delay from an unknown sequence of instructions run a hard-coded number of times. And a developer can't know the time it takes for an unknown sequence of instructions to run.

A processor can have a large number of different ways to present the C code with timing feedback. A traditional timer is just one of them. It might be possible to read back information from a baudrate generator. Or abuse a unused SPI device for performing a dummy transfer. Or maybe there is a RTC.