> For the complete documentation index, see [llms.txt](https://gitbook.hashpower.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://gitbook.hashpower.io/perpetuals/gas_optimizations.md). # HashPowerPerpsDEX — Gas Optimisation Summary Measured on `createOrder` via `tests/gas-createOrder.test.ts`.\ All scenarios use a single seller (maker) placing N resting orders, then a single buyer (taker) sweeping them in one `createOrder` call. *** ## Results | Scenario | Baseline | Round 1–7 | + DLL (Round 8) | + Micro-opts (Round 9, no DLL) | Total saved (baseline → R9) | % | | ------------------------ | ------------: | ------------: | --------------: | -----------------------------: | --------------------------: | ---------: | | Resting only (no match) | 393,610 | 376,018 | **369,001** | **371,260** | -22,350 | -5.7% | | 1 match | 338,467 | 337,190 | **329,391** | **336,196** | -2,271 | -0.7% | | 3 matches (1 level) | 410,999 | 392,865 | **376,061** | **390,916** | -20,083 | -4.9% | | 10 matches (1 level) | 795,267 | 730,676 | **686,135** | **724,689** | -70,578 | -8.9% | | 10 matches (5 levels) | 869,548 | 809,504 | **748,720** | **803,518** | -66,030 | -7.6% | | 20 matches (1 level) | 1,343,788 | 1,212,812 | **1,128,647** | **1,201,057** | -142,731 | -10.6% | | 20 matches (10 levels) | 1,510,922 | 1,390,176 | **1,269,464** | **1,378,422** | -132,500 | -8.8% | | **32 matches (1 level)** | **2,002,040** | **1,791,375** | **1,659,661** | **1,772,699** | **-229,341** | **-11.5%** | ### Average gas per match (32-match scenario) | | Gas | | --------------------------------- | ------------------: | | Baseline | 62,564 | | After rounds 1–7 | 55,980 | | After DLL (round 8) | **51,864** | | Round 9 (micro-opts only, no DLL) | **55,397** | | Total reduction vs baseline (R9) | **-7,167 (-11.5%)** | *** ## Changes Made ### 1. Remove `createdAt` from `Order` struct `Order` shrank from 4 storage slots to 3 by dropping the `createdAt` field.\ The indexer reads timestamps from block metadata instead. **Impact:** saves 1 cold SLOAD per maker order read (\~2,100 gas × N matches). ```solidity // Before (4 slots) struct Order { address participant; uint256 price; int256 quantity; uint256 createdAt; // removed } // After (3 slots) struct Order { address participant; uint256 price; int256 quantity; } ``` *** ### 2. Settle taker funding once before the matching loop `_settleFunding(taker)` was previously called inside `_updateUserPosition` on every match.\ Since `cumulativeFundingPerUnit` is constant within a transaction, calls 2–N were no-op reads. **Fix:** call `_settleFunding(_msgSender())` once in `createOrder` before the loop; `_createPosition` skips settlement for the taker on each iteration. **Impact:** saves \~3 SLOADs × (N−1) redundant calls. *** ### 3. Defer `_removePriceLevelIfEmpty` to after each price level Previously called inside `_removeOrder` on every fully-filled maker order.\ For N orders at one price level, N−1 of those calls were no-ops (queue still had orders). **Fix:** removed the call from `_removeOrder`; `_matchOrdersAtPrice` calls it once after its inner loop, and `cancelOrder` calls it directly. **Impact:** saves \~(N−1) × \~200 gas per price level swept. *** ### 4. Replace `sizeOf()` loop guard with head-pointer check `_matchOrdersAtPrice` checked `makerOrderQueue.sizeOf() > 0` (1 SLOAD) on every iteration before getting the head node (2 SLOADs). **Fix:** read the head pointer once before the loop and re-read it after each removal. **Impact:** saves 1 SLOAD per match. *** ### 5. Cache `makerParticipant` and `makerPrice` in `_executeMatch` `_makerOrder.participant` was read from storage three times (struct copy, `_chargeMatchFee`, `userTotalOrderValue`). `_makerOrder.price` was read twice. **Fix:** cache both fields at the top of `_executeMatch`; pass `makerParticipant` directly to `_createPosition` (eliminating `Order memory orderCopy`) and to `_removeOrder`. **Impact:** saves 2 warm SLOADs per match for participant; eliminates the full struct memory copy. *** ### 6. Move `userTotalOrderValue` decrement out of `_removeOrder` `_removeOrder` always subtracted the order's notional value from `userTotalOrderValue`.\ When called from `_executeMatch` after a full fill, `order.quantity` was already 0, making it a no-op SSTORE. **Fix:** callers own the decrement. `_executeMatch` decrements by `notionalValue` (already done before the call). `cancelOrder` decrements by the remaining order value before calling `_removeOrder`.\ `_removeOrder` now only handles structural removal: queue, participant index, and storage delete. **Impact:** saves 1 warm SSTORE per full fill. *** ### 7. Cache `sizeOf()` in `_addPriceLevel` `priceList.sizeOf()` was called twice (capacity check + empty check). **Fix:** read once into a local variable. **Impact:** saves 1 SLOAD per resting order placement. *** ### 8. Remove dead code: `_isOppositeSign` The function was defined but never called. Removed for clarity. *** ### 8. Replace `StructuredLinkedList` with custom `DLL` library `StructuredLinkedList` uses nested mappings (`mapping(uint256 => mapping(bool => uint256))`), requiring **two** keccak256 slot derivations per node pointer access, and `remove()` always zeros both pointers and decrements the `size` counter (3 extra SSTOREs). **Fix:** new `DLL` library (`contracts/DLL.sol`) with: * **Flat mappings** — `mapping(uint256 => uint256) next` and `prev` (1 keccak256 per access). * **`removeFast`** — skips zeroing node pointers and the size decrement. Safe for order queues where node IDs are unique keccak256 hashes and are never reinserted. * **`isEmpty()`** — checks `next[0] == 0` instead of `sizeOf() == 0`, correct without a maintained size counter. * **`getNext(node)`** — returns plain `uint256` (no discarded `bool` in every call site). Order queues use `pushBack` / `removeFast` / `isEmpty` / `getNext`.\ Price lists use `pushFront` / `insertBefore` / `insertAfter` / `remove` / `nodeExists` / `sizeOf` / `getNext` (full semantics, same as before). **Impact:** \~4,100 gas saved per match; \~7,000 gas saved on the resting-only path. *** ### 9. Micro-optimisations (Round 9, no DLL — applied 2026-02-20) Applied worktree optimisations without swapping the linked-list implementation. 1. **Skip `_settleFunding` when funding is not configured**\ Early return when `lastFundingUpdateTime == 0` or `fundingPeriod == 0`. Benchmarks use disabled funding, so this saves storage reads on the hot path. 2. **Cache `_msgSender()` in `createOrder`**\ Resolve `address sender = _msgSender()` once; reuse throughout. 3. **Validation inlining**\ Inline `_quantity == 0` and `_price` checks at function start instead of calling `_validateQuantity` / `_validatePrice`. 4. **`_executeMatch` micro-opts**\ Cache `makerQty`, compute `newMakerQty` once, use `unchecked { return _remainingQty - matchQty }`. 5. **`_createPosition` simplification**\ Remove redundant taker parameter (was passed twice); simplify signature to `(makerParticipant, taker, ...)`. 6. **Cheaper order IDs**\ Replace `keccak256(abi.encode(...))` with `bytes32(++nonce)`. Sequential counter is unique within the contract and much cheaper. **Impact (Round 9 vs Round 1–7, no DLL):** \~4.8K gas on resting-only; \~1–19K gas on matching scenarios. DLL migration (Round 8) would yield additional \~4.1K gas per match. *** ## What was considered but not done | Idea | Reason skipped | | ---------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | | Cache `takerFeeBps` / `liquidationFee` before loop | Slots become warm after first match; EVM warm-slot cache already handles subsequent reads at 100 gas | | Skip `_settleFunding` for maker when same address appears in multiple consecutive orders | Maker addresses vary in production; the optimisation would only help the synthetic single-seller benchmark |