No, these macros expand to calls to __builtin_avr_delay_cycles(), which are compiled into delay loops. These loops cycle-accurate, but they are affected by interrupts. The delay you get may then be larger than what you asked for.

No, these macros expand to calls to __builtin_avr_delay_cycles(), which are compiled into delay loops. It should be noted that:

• the arguments to these macros should be compile-time constants, they can be floating point
• the macros are cycle-accurate, e.g. _delay_us(0.125) will take exactly two CPU cycles on an Uno
• whereas the CPU time used by the call is exactly what you ask for, the physical time will be larger if the call gets interrupted.