Division is the slow arithmetic; multiplication is fast. So compute the reciprocal and multiply. Newton’s method on f(y)=1/y − d gives the iteration y ← y·(2 − d·y) — each step doubles the correct digits, using only multiply and subtract, converging quadratically to 1/d. Then a/d = a·(1/d). This is how many CPUs and GPUs actually divide: a seed from a table, then two or three Newton steps — no long-division circuit at all.
The demo finds 1/8 by Newton from a rough seed 0.1 (multiply/subtract only), then computes 40/8 = 5 without a division: live demo
“To divide you need a divider.” — you need a reciprocal, and Newton finds that with only multiply and subtract. cited
The reciprocal grown by multiply-and-subtract, the quotient a single scaling — division without a divider. resource
On the canonical compiler, Newton from seed 0.1 reaches 1/8 = 0.125 in six steps; 40×0.125 = 5, no divide used: