The biggest wins are structural: fewer storage writes, packed structs, `calldata` instead of `memory`, short-circuiting checks, and `unchecked` blocks for counters that provably cannot overflow.
for (uint256 i; i < n;) {
// ...
unchecked { ++i; }
}Micro-tricks matter far less than avoiding an extra SSTORE. Measure with `forge snapshot` rather than guessing.