[preciz]

Assisted-by: Antigravity:Gemini 3.8 Flash

If after div/2 we call rem/2 then the compiler fuses it to one operation.

https://github.com/erlang/otp/blob/master/erts/emulator/beam/jit/x86/ops.tab#L1190-L1213

These changes have been evaluated locally to produce fewer assembly lines and that they run faster.

Bench:

Mix.install([:benchee])

defmodule Bench do
  def old(n, m) do
    div = div(n, m)
    n - div * m
  end

  def new(n, m) do
    _ = div(n, m)
    rem(n, m)
  end
end

Benchee.run(
  %{
    "old" => fn -> Bench.old(10, 3) end,
    "new" => fn -> Bench.new(10, 3) end
  },
  warmup: 1,
  time: 3
)

[preciz]

Assembly lines:

Module & Function	Before (HEAD~1)	After (HEAD)	Net Change
Calendar.ISO.div_rem/2	166 lines div=1, rem=0 mul=1, sub=2	110 lines div_rem=1 mul=0, sub=1	-56 lines
Calendar.ISO.offset_to_iodata/4	139 lines div=2, rem=1	125 lines div_rem=1, div=1	-14 lines
NaiveDateTime.div_rem/2	166 lines div=1, rem=0 mul=1, sub=2	110 lines div_rem=1 mul=0, sub=1	-56 lines
Enumerable.File.Stream.count/1	552 lines div=1, rem=1	542 lines div_rem=1, div=0	-10 lines
Float.bignum_to_float/4	360 lines div=1, rem=1	351 lines div_rem=1, div=0	-9 lines
Integer.floor_div/2	131 lines div=2, rem=1	112 lines div_rem=1, div=0	-19 lines
Integer.ceil_div/2	133 lines div=2, rem=1	114 lines div_rem=1, div=0	-19 lines
Integer.extended_gcd/6	299 lines div=1, rem=0 mul=3, sub=3	261 lines div_rem=1 mul=2, sub=2	-38 lines
String.build_filler/5	355 lines div=1, rem=1	340 lines div_rem=1, div=0	-15 lines

[josevalim]

Thanks but chasing this level of optimization is unlikely to be worth our time investment in most cases!