There are several ways to implement a delay.

1. Using the hardware Timers
2. A software delay loop

Both methods depend on the system clock. The first method requires resources that may be already in use.

The software method is fairly portable with the important proviso that you need to calibrate it for your particular compiler, version memory model ....

Having done the single calibration, all of the project code is going to run at the intended speed. e.g. if you want to delay for one millisecond you call delay_ms(1) rather than some arbitrarily named function with some magic constant.

With many microcontroller compilers, you implement the delay with inline ASM code with the parameters calculated by a C macro solely dependent on F_CPU and the #microseconds required.

If the compiler does not permit inline ASM, you just have to calibrate a C loop that is written to give "predictable" code. ( probably by forcing specific memory model volatile variables). By no means perfect but a practical solution.

And if you change compiler, cpu, memory model... you only have one place to alter or re-calibrate your macro or function.

David.