For example the following solidity snippet

uint160 a, b;
unchecked { a + b }

would be converted into (by the optimizer)

let _1 := sub(shl(160, 1), 1)
let value := and(mul(a, b), _2)

Here _1 is a shorter representation of the constant.

Instead the more optimized version would be

let _1 := shl(160, 1)
let value := mulmod(a, b, _1)

It is shorter and avoids a sub. and + mul has the same gas as mulmod.

Similarly, for addmod, but add + and is 6, and addmod is 8. We save a sub. So total saving of 2 gas and shorter deploy code.

It maybe easier to implement this as a rule and(mul(X, Y), A) -> mulmod(X, Y, 2**N) where A = 2**N - 1. Similarly, for the other?