feat: Delayed merge implementation by federicobarbacovi · Pull Request #22775 · AztecProtocol/aztec-packages

federicobarbacovi · 2026-04-24T16:51:36Z

Implementation of the delayed merge: we replace the single merges in the kernels with a final batch merge in the hiding kernel. Description of the batch merge protocol is the relevant md file

Co-authored-by: Copilot <copilot@github.com>

…y merge in favor of translator-merge

ledwards2225

Looks good! Basically reviewed everything but want to spend a bit more time tomorrow. Submitting this now so you can see some of my comments. Mostly minor suggestions

ledwards2225 · 2026-04-28T18:16:53Z

    update_native_verifier_accumulator(queue_entry, verifier_transcript);
 #endif
-    goblin.prove_merge(prover_accumulation_transcript);
+    // Delayed merge: keep one subtable per folded circuit and prove the batched merge after the tail kernel.


Can't comment on unchanged lines so this is intended for the code above: it seems like HN_TAIL is no longer needed since AFAIK the only reason for it to exist was to add the blinding that's now handled by the batched merge

This is correct, agreed to tackle this in a follow up as it touches Noir as well

ledwards2225 · 2026-04-28T19:31:01Z

+ * @brief Create a mock batch merge proof which has the correct structure but is not necessarily valid
+ *
+ */
+bb::HonkProof create_mock_batch_merge_proof(bool is_zk = true);


It seems this is_zk param is always true. Are you just maintaining it based on the idea of doing goblin resets or something where we'd prefer not to use ZK?

…e in batch verifier Co-authored-by: Copilot <copilot@github.com>

…ification in is_hiding_kernel branch

Co-authored-by: Copilot <copilot@github.com>

…b/delayed_merge_impl

Co-authored-by: Copilot <copilot@github.com>

ledwards2225

LGTM. Did another pass though everything, added some tests and tweaked some out of date comments. Fixed what I think was a bug where we were adding the same random ops 3 times instead of generating them uniquely. Overall quite happy with how simple the changes are here. I'll leave it to you to decide if there's more testing/tweaking/assessing you want to do.

ledwards2225 · 2026-04-29T18:44:52Z

        BB_ASSERT_EQ(current_stdlib_verifier_accumulator.has_value(), false);

+        // Perform batch merge verification
+        auto [batch_pairing_points, batch_merged_table_commitments] =


nice! the changes in this file are so clean now

Co-authored-by: Copilot <copilot@github.com>

…b/delayed_merge_impl

…22906) Fixes the merge-train/barretenberg merge-queue failure: http://ci.aztec-labs.com/1617122ce095114b ## Summary `compute_running_hash` in `batch_merge.test.cpp` (added by #22775) declared `bb::fr hash_challenge(0)` but never used it. Under `-Werror -Wall`, the ARM build fails with `[-Wunused-variable]`. The fix removes the dead declaration. ## Why ci-full on #22775 didn't catch it Clang's `-Wunused-variable` is suppressed when a variable's constructor has visible side effects. On x86 (`DISABLE_ASM=OFF`), `field<T>` constructors call `self_to_montgomery_form()` → `operator*=` → `asm_self_mul_with_coarse_reduction`, which contains an `__asm__()` block with a `"memory"` clobber. Clang treats inline asm as a side effect and silences the warning. On ARM, `CMakeLists.txt` auto-sets `DISABLE_ASM=ON` (line 50–57), so `bb::fr(0)` resolves to a pure `constexpr` arithmetic path with no observable side effect — clang then issues the warning. `ci-full` (label) runs only on amd64 (see `ci.sh` `full` target — single `bootstrap_ec2 ci-full` on x86), so the asm side-effect masked the warning. The merge queue (`ci.sh merge-queue`) is what adds the arm64 leg (`a1-fast arm64 ci-fast`), which is why the failure surfaced only at merge time. ClaudeBox log: https://claudebox.work/s/e0102d5ad296fd89?run=1

BEGIN_COMMIT_OVERRIDE fix(ci): default S3_BUILD_CACHE_AWS_PARAMS in cache_s3_transfer{,_to} (AztecProtocol#22898) chore: low-hanging chonk prover fixes from profiling (AztecProtocol#22855) chore: fuse N `add_scaled` into one `parallel_for` (AztecProtocol#22893) feat: Delayed merge implementation (AztecProtocol#22775) chore: numeric audit response (AztecProtocol#22856) fix: harden BN254 G2 SRS ingress (AztecProtocol#22858) fix: remove unused hash_challenge variable in batch_merge.test.cpp (AztecProtocol#22906) fix(bbup): remove jq dependency (AztecProtocol#22912) chore: fix g2 test failing on merge-train (AztecProtocol#22920) fix(ci): error on disabled-cache in CI hash calculation (AztecProtocol#22904) END_COMMIT_OVERRIDE

federicobarbacovi and others added 7 commits April 27, 2026 09:10

Implement batch merge

64fdd0c

Prover + verifier + tests

5436fcc

Co-authored-by: Copilot <copilot@github.com>

zk + test suite updates

c13aa7f

Co-authored-by: Copilot <copilot@github.com>

1 <= N <= Max merge size + more tests + refactor

5736945

Co-authored-by: Copilot <copilot@github.com>

Use correct exponents

f76f268

Chonk integration

238c930

Full chonk integration + ecc ops table/queue restructuring

de5b622

federicobarbacovi force-pushed the fb/delayed_merge_impl branch from ce5fe78 to de5b622 Compare April 27, 2026 17:18

iakovenkos reviewed Apr 28, 2026

View reviewed changes

Comment thread barretenberg/cpp/src/barretenberg/chonk/chonk.cpp Outdated

iakovenkos reviewed Apr 28, 2026

View reviewed changes

Comment thread barretenberg/cpp/src/barretenberg/chonk/chonk.cpp Outdated

federicobarbacovi and others added 4 commits April 28, 2026 10:00

Clean up (remove unused code + merge is now fixed append only)

279d66d

DSL integration

67d7d42

Co-authored-by: Copilot <copilot@github.com>

Update vks

ba849fa

Update README

c906659

federicobarbacovi added ci-full Run all master checks. ci-no-fail-fast Sets NO_FAIL_FAST in the CI so the run is not aborted on the first failure labels Apr 28, 2026

federicobarbacovi marked this pull request as ready for review April 28, 2026 13:38

ledwards2225 self-requested a review April 28, 2026 17:01

minor ecc ops table simplicifaction; more possible if we remove legac…

f009ba8

…y merge in favor of translator-merge

ledwards2225 reviewed Apr 28, 2026

View reviewed changes

federicobarbacovi and others added 10 commits April 29, 2026 12:32

Build transcript internally in batch prover/verifier + split challeng…

d3a1d27

…e in batch verifier Co-authored-by: Copilot <copilot@github.com>

Use non-empty polys + docs

00c3988

Remove optional table commitments and place recursive batch merge ver…

0f0c07c

…ification in is_hiding_kernel branch

Bump max Chonk size

05617d4

Remove merge queue

d5481fc

Co-authored-by: Copilot <copilot@github.com>

Add tests for templated pow

67d8249

Co-authored-by: Copilot <copilot@github.com>

Merge remote-tracking branch 'origin/merge-train/barretenberg' into f…

1b82545

…b/delayed_merge_impl

Update max stack size + add assertion

c57e9b4

Co-authored-by: Copilot <copilot@github.com>

Update gate counts

98fcc9b

Update vks

1f6a724

ledwards2225 added 2 commits April 29, 2026 21:06

general comment update; remove refs to prepending etc

60fa60d

comment about quirks of num_ultra_rows()

a4c3da2

ledwards2225 reviewed Apr 29, 2026

View reviewed changes

federicobarbacovi and others added 13 commits April 30, 2026 10:32

Remove is_zk flag, move batch prove call

9b90251

Co-authored-by: Copilot <copilot@github.com>

Fix batched translator tests

f5ea77c

Co-authored-by: Copilot <copilot@github.com>

Updates after removing is_zk flag

32be0a2

Update vks

6cc350d

Fixes

07d602e

Fix + update vks

36c8314

Update batch merge protocol description

2c22d6a

Update batch merge md

524a7af

Clean up + defensive assertions

6883849

Merge remote-tracking branch 'origin/merge-train/barretenberg' into f…

211d25c

…b/delayed_merge_impl

Profiling infra

bb8b04f

Update gate count

62fca92

Update vks

85e4b6d

federicobarbacovi self-assigned this May 1, 2026

federicobarbacovi added 3 commits May 1, 2026 17:41

Merge remote-tracking branch 'origin/merge-train/barretenberg' into f…

93cbb86

…b/delayed_merge_impl

Merge remote-tracking branch 'origin/merge-train/barretenberg' into f…

bb9bfb4

…b/delayed_merge_impl

Use batch add scaled in batch merge prover

94b345a

federicobarbacovi requested a review from ledwards2225 May 1, 2026 19:20

federicobarbacovi enabled auto-merge (squash) May 1, 2026 19:21

ledwards2225 approved these changes May 1, 2026

View reviewed changes

federicobarbacovi merged commit d4cd48e into merge-train/barretenberg May 1, 2026
15 checks passed

federicobarbacovi deleted the fb/delayed_merge_impl branch May 1, 2026 20:03

This was referenced May 1, 2026

feat: merge-train/barretenberg #22901

Merged

fix: remove unused hash_challenge variable in batch_merge.test.cpp #22906

Merged

federicobarbacovi mentioned this pull request May 5, 2026

Perform multiple merge verifications at once to reduce the number of MSMs AztecProtocol/barretenberg#1471

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: Delayed merge implementation#22775

feat: Delayed merge implementation#22775
federicobarbacovi merged 45 commits into
merge-train/barretenbergfrom
fb/delayed_merge_impl

federicobarbacovi commented Apr 24, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

ledwards2225 left a comment

Uh oh!

Uh oh!

ledwards2225 Apr 28, 2026

Uh oh!

federicobarbacovi Apr 29, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ledwards2225 Apr 28, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ledwards2225 left a comment

Uh oh!

ledwards2225 Apr 29, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

federicobarbacovi commented Apr 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ledwards2225 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ledwards2225 Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

federicobarbacovi Apr 29, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ledwards2225 Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ledwards2225 left a comment

Choose a reason for hiding this comment

Uh oh!

ledwards2225 Apr 29, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

federicobarbacovi commented Apr 24, 2026 •

edited

Loading