chore: honk verifier opt + fix review by Maddiaa0 · Pull Request #20547 · AztecProtocol/aztec-packages

Maddiaa0 · 2026-02-16T14:08:49Z

Overview

This has been submitted for audit already, so I've attempted to make the changes / optimizations as small as possible. I came across these while reading through it again for the zk verifier.

Commits are squashed to please CI - however ive pushed them to a mirror so they still exist within the repository

removing a scalar mul which was being multiplied by 1
- 699c7e2
LAGRANGE_FIRST is always (1,2), constant_term_acc is always multiplied by (1,2) so add the scalars together
- 6f2c350
the remaining are batching all inversions into the same modexp
- 05217b8
- f1a5830

chore: remove redundant scalar mul chore: combine constant_term_acc and LAGRANGE_FIRST scalar mul chore: batch public inputs inversion chore: deduplicate gemini denominator and document memory layout chore: merge all inversions into single modexp

Maddiaa0 · 2026-02-17T18:22:05Z

@copilot can you remove the commit of barretenberg/sol/src/honk/instance/other.sol

Copilot · 2026-02-17T18:22:14Z

@Maddiaa0 I've opened a new pull request, #20610, to work on those changes. Once the pull request is ready, I'll request review from you.

### 🧾 Audit Context Minor optimizations identified during zk verifier review of already-audited ultra honk verifier code. ### 🛠️ Changes Made - Removed scalar multiplication by 1 (no-op elimination) - Consolidated LAGRANGE_FIRST (1,2) constant term accumulation by adding scalars directly - Batched all inversions into single modexp operation - Removed unintended `barretenberg/sol/src/honk/instance/other.sol` file No functional changes. Purely optimization-focused refactoring. No circuit changes. ### ✅ Checklist - [x] Audited all methods of the relevant module/class - [x] Audited the interface of the module/class with other (relevant) components - [ ] Documented existing functionality and any changes made (as per Doxygen requirements) - [x] Resolved and/or closed all issues/TODOs pertaining to the audited files - [x] Confirmed and documented any security or other issues found (if applicable) - [x] Verified that tests cover all critical paths (and added tests if necessary) - [ ] Updated audit tracking for the files audited (check the start of each file you audited) ### 📌 Notes for Reviewers Changes kept minimal as code already submitted for audit. See individual commits for isolated changes: - 699c7e2: scalar mul removal - 6f2c350: LAGRANGE_FIRST consolidation - 05217b8, f1a5830: inversion batching  --- 💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more [Copilot coding agent tips](https://gh.io/copilot-coding-agent-tips) in the docs. --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Maddiaa0 <47148561+Maddiaa0@users.noreply.github.com>

Maddiaa0 · 2026-02-18T10:12:35Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp

    for (int i = 0; i < BARYCENTRIC_DOMAIN_SIZE; ++i) {
-        print_fr(bary_pointer, "BARYCENTRIC_LAGRANGE_DENOMINATOR_" + std::to_string(i) + "_LOC");
-        bary_pointer += 32;
+        print_fr(pointer, "BARYCENTRIC_LAGRANGE_DENOMINATOR_" + std::to_string(i) + "_LOC");


These were starting from "scratch space", but I've leaned towards not re-using allocations due to memory having a trivial gas overhead

Maddiaa0 · 2026-02-18T10:12:57Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp

    // INVERSIONS
    print_header_centered("SHPLEMINI - RUNTIME MEMORY - INVERSIONS");

+    print_fr(pointer, "GEMINI_R_INV_LOC");


stored in memory now as the inversion is performed before usage

Maddiaa0 · 2026-02-18T10:13:12Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp

    out << "// LOG_N challenge pow minus u\n";
    for (int i = 0; i < log_n; ++i) {
-        print_fr(pointer, "INVERTED_CHALLENEGE_POW_MINUS_U_" + std::to_string(i) + "_LOC");
+        print_fr(pointer, "INVERTED_CHALLENGE_POW_MINUS_U_" + std::to_string(i) + "_LOC");


Maddiaa0 · 2026-02-18T10:14:35Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp


    out << "\n";
-    print_fr(pointer, "LATER_SCRATCH_SPACE");
+    for (int i = 0; i < 8 * log_n; ++i) {


Previously these were located in LATER_SCRATCH_SPACE, i thought it would be better to explicitly provide their memory locations in the layout at the top of the file.

Hidden allocations could be a source of footguns when refactoring

…ract.hpp previously we were writing to an offset of 0x20 which was not expected - how this lines up correctly

LHerskind

Using my 🥜 brain, but think there are a few unused bits than can be cut off. And then I am just curious about the first_lagrange really.

LHerskind · 2026-02-20T12:48:50Z

barretenberg/sol/src/honk/optimised/honk-optimized.sol.template

+                                mulmod(mload(PUBLIC_INPUTS_DELTA_NUMERATOR_CHALLENGE), pi_delta_inv, p)
+                            )
+                        }
+
                        // Normalise as last loop will have incremented the offset
                        bary_centric_inverses_off := sub(bary_centric_inverses_off, 0x20)
                        for {} gt(bary_centric_inverses_off, SUM_U_CHALLENGE_14) {


Don't this run for too long?

SUM_U_CHALLENGE_14 = 0x3cc0 and bary_centric_inversses_off = 0x4ce0, but the BARYCENTRIC_DENOMINATOR_INVERSES_0_0_LOC = 0x3de0 so it unds up looping more than planned and overwriting BARYCENTRIC_LAGRANGE_DENOMINATOR_7..0 (8 excess runs of the body). The accumulator and other values don't seem to be used afterwards, so it does not break currently, but just writing more than needed?

unbelievable catch, before this pr ( or maybe the last one ) barycentric_lagrange_denominator was in scratch space ( starting at 0x100 ). so this was correct, this pr updates such that the lagrange_denominators are stored in the slab with everything else. But does not update this pointer

fix is replacing SUM_U_CHALLENGE_14 with BARYCENTRIC_LAGRANGE_DENOMINATOR_7

see: https://github.com/AztecProtocol/aztec-packages/pull/20547/changes#r2821481981

LHerskind · 2026-02-20T12:52:31Z

barretenberg/sol/src/honk/optimised/honk-optimized.sol.template

@@ -4203,8 +4553,11 @@ contract BlakeOptHonkVerifier is IVerifier {
                )

                // Accumulator = accumulator + scalar[27] * vk[26]
+                // optimization - Lagrange first is always G - (1,2)


Are we absolutely sure that LAGRANGE_FIRST is always (1, 2)? Because then we should do that in both the optimized and base, to ensure that it is not different between the two variants and that will break together if it changes.

yeah it is - its enforced in the vk

LHerskind · 2026-02-20T12:55:44Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp

                let neg_inverted_denominator := mload(NEG_INVERTED_DENOM_0_LOC)
                let shplonk_nu := mload(SHPLONK_NU_CHALLENGE)

                unshifted_scalar := addmod(pos_inverted_denominator, mulmod(shplonk_nu, neg_inverted_denominator, p), p)


If we are onlysetting it here, we mightn ot need the separate let unshifted_scaler := 0 and then almost instantly setting it?

ah this was getting around a stack too deep

LHerskind · 2026-02-20T13:08:21Z

barretenberg/sol/src/honk/optimised/honk-optimized.sol.template

+                    }
+                    /// {{ UNROLL_SECTION_END ACCUMULATE_INVERSES }}
+
+                    // Invert all elements (barycentric + PI delta + shplemini) as a single batch


Should we apply similar checks as in Fr.sol to avoid trying to invert a 0 value? If there is one 0 we would end up fully making all the accumulator 0. It should be caught by other parts of the verification, but 🤷

we could, but if we invert a zero everything will blow up after

LHerskind · 2026-02-20T13:14:40Z

barretenberg/cpp/src/barretenberg/dsl/acir_proofs/honk_optimized_contract.hpp

I'm not sure you read this value, seems to start at BATCH_SCALAR_1_LOC onwards.

yep its now unused

LHerskind · 2026-02-20T13:16:23Z

barretenberg/sol/src/honk/optimised/honk-optimized.sol.template

+    uint256 internal constant BATCH_EVALUATION_ACCUMULATOR_INVERSION_13_LOC = 0x5f80;
+    uint256 internal constant BATCH_EVALUATION_ACCUMULATOR_INVERSION_14_LOC = 0x5fa0;
+
+    uint256 internal constant BATCHED_EVALUATION_LOC = 0x5fc0;


I don't think you actually use this one. It looks like you store it in memory, but then you use the stack item batched_evaluation right after and not the memory variant.

Maddiaa0 added 7 commits September 4, 2025 23:02

fix: templating issue

34b197c

Merge branch 'next' into md/fix-verifier

e40fc96

chore: fix optimized verifier

ef166f2

fox

03f0543

chore: fix lookup tables

18e1603

chore: use opt verifier

8ed3b40

chore: various optimisations

965d231

chore: remove redundant scalar mul chore: combine constant_term_acc and LAGRANGE_FIRST scalar mul chore: batch public inputs inversion chore: deduplicate gemini denominator and document memory layout chore: merge all inversions into single modexp

Maddiaa0 force-pushed the md/honk-golf branch from f1a5830 to 965d231 Compare February 16, 2026 15:20

Maddiaa0 added 6 commits February 17, 2026 11:55

chore: fix no memory overlaps

9c158c3

chore: code gen

aea2d8c

chore: more template fixes

ac39a5c

chore: update template

8ed0ac9

Merge branch 'next' into md/fix-verifier

77c1e87

revert using optimized verifier in prod

7987361

Copilot AI mentioned this pull request Feb 17, 2026

chore: lil bit of golfing on the ultra honk verifier #20610

Merged

7 tasks

Maddiaa0 and others added 3 commits February 18, 2026 09:56

fix

fb7e597

Merge branch 'next' into md/fix-verifier

19451c7

Maddiaa0 changed the base branch from next to md/fix-verifier February 18, 2026 10:01

Merge branch 'md/fix-verifier' into md/honk-golf

da4e454

Maddiaa0 marked this pull request as ready for review February 18, 2026 10:03

Maddiaa0 requested a review from LHerskind February 18, 2026 10:03

Maddiaa0 commented Feb 18, 2026

View reviewed changes

AztecBot force-pushed the md/fix-verifier branch from 19451c7 to ed81ff5 Compare February 18, 2026 10:55

chore: fix honk optimized allocations

885ed4f

Maddiaa0 added 2 commits February 18, 2026 15:37

chore: make sure allocations line up - applied to honk_optimized_cont…

7de5747

…ract.hpp previously we were writing to an offset of 0x20 which was not expected - how this lines up correctly

opt: recomputing in looks is cheaper than using constants

8e26466

Base automatically changed from md/fix-verifier to next February 19, 2026 15:22

Maddiaa0 added 2 commits February 19, 2026 15:44

Merge branch 'next' into md/honk-golf

409ccf7

apply opts to cpp

fa0c321

LHerskind reviewed Feb 20, 2026

View reviewed changes

Maddiaa0 added 4 commits February 20, 2026 14:34

chore: review comments

e10287b

chore: remove inverted gemini

93353ca

Merge branch 'next' into md/honk-golf

60a785a

chore: add large circuit test

b86aedf

Maddiaa0 changed the title ~~chore: lil bit of golfing on the ultra honk verifier~~ chore: honk verifier opt + fix review Mar 10, 2026

Maddiaa0 added 2 commits March 10, 2026 16:56

chore: remove unused vk_fr

6311540

fix: sub(gate_challenge, 1)

58f7ada

Conversation

Maddiaa0 commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Overview

Uh oh!

Maddiaa0 commented Feb 17, 2026

Uh oh!

Copilot AI commented Feb 17, 2026

Uh oh!

Maddiaa0 Feb 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LHerskind left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Maddiaa0 Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Maddiaa0 commented Feb 16, 2026 •

edited

Loading

Maddiaa0 Feb 18, 2026 •

edited

Loading

Maddiaa0 Feb 20, 2026 •

edited

Loading