Agner Fog’s manual lists a latency/reciprocal-throughput of 15/10 for 64-bit divide on Ivy Lake & Tiger Lake, vs 3/1 for 64-bit multiply. So division is 5-10x slower than multiplication.

https://www.agner.org/optimize/#manuals

It makes a significant difference in profiling, versus i.e. a bitwise mask to find a modulus for a power-of-two size.