feat: integrate bytecode and program image reductions by RadNi · Pull Request #19 · LayerZero-Research/jolt

RadNi · 2026-05-21T01:28:43Z

Generated stack PR from amir/bytecode-commitment-merged.

Depends on spec PR: a16z#1565

Stack position: 04
Base branch: amir/bytecode-stack/03-program-foundation

Owned paths:

jolt-core/src/zkvm/claim_reductions/precommitted.rs
jolt-core/src/zkvm/claim_reductions/mod.rs
jolt-core/src/zkvm/bytecode/read_raf_checking.rs
jolt-core/src/zkvm/bytecode/mod.rs
jolt-core/src/zkvm/ram/mod.rs
jolt-core/src/zkvm/ram/val_check.rs
jolt-core/src/zkvm/claim_reductions/hamming_weight.rs
jolt-core/src/poly/opening_proof.rs
jolt-core/src/poly/commitment/dory/wrappers.rs
jolt-core/src/poly/rlc_polynomial.rs
jolt-core/src/utils/errors.rs
jolt-core/src/zkvm/config.rs
jolt-core/src/zkvm/mod.rs
jolt-core/src/zkvm/program.rs
jolt-core/src/zkvm/prover.rs
jolt-core/src/zkvm/transpilable_verifier.rs
jolt-core/src/zkvm/verifier.rs
jolt-core/src/zkvm/witness.rs
jolt-core/src/guest/prover.rs
jolt-core/src/guest/verifier.rs
jolt-core/benches/e2e_profiling.rs
jolt-sdk/macros/src/lib.rs
jolt-sdk/src/host_utils.rs
src/build_wasm.rs
transpiler/src/main.rs

This PR is expected to be updated manually when amir/bytecode-commitment-merged is resliced.

github-actions · 2026-05-21T01:28:55Z

Warning

This PR has more than 500 changed lines and does not include a spec.

Large features and architectural changes benefit from a spec-driven workflow.
See CONTRIBUTING.md for details on how to create a spec.

If this PR is a bug fix, refactor, or doesn't warrant a spec, feel free to ignore this message.

cursor

Cursor Bugbot has reviewed your changes and found 2 potential issues.

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 6150eee. Configure here.}

cursor · 2026-05-21T18:12:27Z

+                    ),
+                    padded_len,
+                },
+            );


Program image words not padded for precommitted polynomial

Medium Severity

When constructing PrecommittedPolynomial::ProgramImage for Stage 8, the words field is set to the raw bytecode_words from RAM preprocessing (unpadded), but padded_len is set to the padded power-of-two length. The vmp_precommitted_contribution function iterates over words and uses padded_len for matrix dimensions (log_2, balanced sigma/nu), creating a mismatch — the words vector is shorter than padded_len, so indexing row_idx = coeff_idx / precommitted_cols may skip contributions that would exist in a properly padded polynomial, and the matrix shape doesn't match the committed polynomial.

^{Reviewed by Cursor Bugbot for commit 6150eee. Configure here.}

cursor · 2026-05-21T18:12:28Z

-
-        let booleanity_cycle_params =
-            BooleanityCyclePhaseParams::new(booleanity.into_params(), &self.opening_accumulator);
+        drop(instances);


ZK sumcheck proof removed from Stage 6a prover

High Severity

The prove_stage6a method previously had a #[cfg(feature = "zk")] block that used BatchedSumcheck::prove_zk with a local BlindFoldAccumulator, because Stage 6a input claims depend on hidden prior-stage outputs in ZK mode. The new code unconditionally uses prove_batched_sumcheck (the non-ZK path), which means Stage 6a no longer produces a ZK sumcheck proof when the zk feature is enabled. This breaks zero-knowledge soundness. The blindfold stage count assertion also changed from 7 to 8, which seems inconsistent with removing a ZK stage.

Additional Locations (1)

jolt-core/src/zkvm/prover.rs#L2611-L2613

^{Reviewed by Cursor Bugbot for commit 6150eee. Configure here.}

moodlezoup · 2026-05-22T12:48:58Z

+        let entry_bytecode_index = self
+            .preprocessing
+            .shared
+            .program_meta
+            .entry_address
+            .saturating_sub(self.preprocessing.shared.program_meta.min_bytecode_address)
+            as usize
+            / BYTES_PER_INSTRUCTION
+            + 1;


This duplicates the logic from BytecodePreprocessing::entry_bytecode_index(). We can move entry_bytecode_index onto ProgramMetadata instead so it can be used by both prover and verifier

moodlezoup · 2026-05-22T12:51:56Z

        bytecode_read_raf_params: BytecodeReadRafSumcheckParams<F>,
        booleanity_params: BooleanitySumcheckParams<F>,
    ) -> Result<StageVerifyResult<F>, ProofVerifyError> {
+        let bytecode_reduction_seed_params = bytecode_read_raf_params.clone();


this is a potentially heavy clone, when BytecodeClaimReductionParams really only needs the gamma values

moodlezoup · 2026-05-22T12:56:18Z

@@ -1293,46 +1343,25 @@
        write_instance_flamegraph_svg(&instances, "stage6a_start_flamechart.svg");
        tracing::info!("Stage 6a proving");

-        #[cfg(feature = "zk")]
-        let (sumcheck_proof, _r_stage6a, _initial_claim) = {
-            // Stage 6a input claims depend on hidden prior-stage outputs in ZK mode,
-            // so we prove it with a ZK sumcheck proof. We keep a local blindfold
-            // accumulator so this split-internal phase does not add a new global
-            // BlindFold stage.
-            let mut rng = rand::thread_rng();
-            let mut local_blindfold =
-                crate::subprotocols::blindfold::BlindFoldAccumulator::<F, C>::new();
-            BatchedSumcheck::prove_zk::<F, C, _, _>(
-                instances.iter_mut().map(|v| &mut **v as _).collect(),
-                &mut self.opening_accumulator,
-                &mut local_blindfold,
-                &mut self.transcript,
-                &self.pedersen_generators,
-                &mut rng,
-            )
-        };
-        #[cfg(not(feature = "zk"))]
        let (sumcheck_proof, _r_stage6a, _initial_claim) =
            self.prove_batched_sumcheck(instances.iter_mut().map(|v| &mut **v as _).collect());

        #[cfg(feature = "allocative")]
        write_instance_flamegraph_svg(&instances, "stage6a_end_flamechart.svg");
-
-        let booleanity_cycle_params =


this sort of Arc::new(bytecode.clone()) appears a couple of times in this PR; we should avoid cloning the bytecode as much as possible. Can we return an Arc from a preprocessing method?

Co-authored-by: Amirhossein Khajehpour <khajepour.amirhossein@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

Update bytecode and program-image verifier integration to use the shared precommitted phase transition helper after cycle-phase openings are cached. Made-with: Cursor Co-authored-by: Cursor <cursoragent@cursor.com>

Thread committed bytecode metadata consistently through prover and verifier paths so bytecode reduction state stays synchronized. Made-with: Cursor Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

github-actions Bot added the no-spec PR has no spec file label May 21, 2026

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from e4cdbb5 to 744ada7 Compare May 21, 2026 01:44

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 01ce43e to 8d14d29 Compare May 21, 2026 01:44

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 744ada7 to d8d5384 Compare May 21, 2026 01:51

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 8d14d29 to b2cab0e Compare May 21, 2026 01:51

RadNi changed the title ~~Integrate bytecode program reductions~~ [04/05] Integrate bytecode program reductions May 21, 2026

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from d8d5384 to 0b06125 Compare May 21, 2026 01:56

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from b2cab0e to 5d0aa55 Compare May 21, 2026 01:56

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 0b06125 to f33facb Compare May 21, 2026 15:37

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 5d0aa55 to 4741cdb Compare May 21, 2026 15:37

RadNi changed the title ~~[04/05] Integrate bytecode program reductions~~ [04/05] Integrate bytecode and program image polynomials reductions May 21, 2026

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from f33facb to 77801f2 Compare May 21, 2026 18:02

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 4741cdb to 6150eee Compare May 21, 2026 18:02

RadNi marked this pull request as ready for review May 21, 2026 18:10

cursor Bot reviewed May 21, 2026

View reviewed changes

RadNi changed the title ~~[04/05] Integrate bytecode and program image polynomials reductions~~ feat: integrate bytecode and program image reductions May 21, 2026

moodlezoup mentioned this pull request May 22, 2026

refactor: split stage 6 sumchecks #16

Open

moodlezoup reviewed May 22, 2026

View reviewed changes

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 77801f2 to d1eb7e0 Compare May 22, 2026 15:57

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 6150eee to 42d671b Compare May 22, 2026 15:57

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from d1eb7e0 to 78c82ba Compare May 22, 2026 17:11

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 42d671b to 5c76a6b Compare May 22, 2026 17:11

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 78c82ba to 9d5b82e Compare May 22, 2026 17:30

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 5c76a6b to 546aa02 Compare May 22, 2026 17:30

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 9d5b82e to e587e65 Compare May 22, 2026 17:48

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 546aa02 to 30d2a9b Compare May 22, 2026 17:48

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from e587e65 to 36378dd Compare May 22, 2026 19:53

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch 2 times, most recently from 48e80c2 to f307a64 Compare May 23, 2026 06:57

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch 2 times, most recently from 850790c to 268b434 Compare May 25, 2026 18:46

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 3e65aa6 to 4637de5 Compare May 25, 2026 18:46

quangvdao and others added 8 commits May 25, 2026 20:37

feat(zkvm): add bytecode program reductions

5ea73f2

Co-authored-by: Amirhossein Khajehpour <khajepour.amirhossein@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): reconcile committed reduction integration with source branch

05ef2ab

Co-authored-by: Amirhossein Khajehpour <khajepour.amirhossein@gmail.com> Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): clean up reduction rebase artifacts

37f8287

Co-authored-by: Cursor <cursoragent@cursor.com>

chore(dory): remove setup cache debug bypass

5b43cee

Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): transition committed reductions from accumulator state

282c829

Update bytecode and program-image verifier integration to use the shared precommitted phase transition helper after cycle-phase openings are cached. Made-with: Cursor Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): align committed bytecode metadata flow

6cabff2

Thread committed bytecode metadata consistently through prover and verifier paths so bytecode reduction state stays synchronized. Made-with: Cursor Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): gate prover-only reduction code

1236121

Co-authored-by: Cursor <cursoragent@cursor.com>

fix(zkvm): align shared opening helper inputs

881b8fc

Co-authored-by: Cursor <cursoragent@cursor.com>

RadNi force-pushed the amir/bytecode-stack/03-program-foundation branch from 268b434 to bf067dd Compare May 26, 2026 04:17

RadNi force-pushed the amir/bytecode-stack/04-bytecode-program-reduction branch from 4637de5 to 881b8fc Compare May 26, 2026 04:38

chore(poly): remove debug opening accessors

e295e75

Co-authored-by: Cursor <cursoragent@cursor.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: integrate bytecode and program image reductions#19

feat: integrate bytecode and program image reductions#19
RadNi wants to merge 9 commits into
amir/bytecode-stack/03-program-foundationfrom
amir/bytecode-stack/04-bytecode-program-reduction

RadNi commented May 21, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented May 21, 2026

Uh oh!

cursor Bot left a comment

Uh oh!

cursor Bot May 21, 2026

Uh oh!

cursor Bot May 21, 2026

Uh oh!

moodlezoup May 22, 2026

Uh oh!

moodlezoup May 22, 2026

Uh oh!

moodlezoup May 22, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

RadNi commented May 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented May 21, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot May 21, 2026

Choose a reason for hiding this comment

Program image words not padded for precommitted polynomial

Uh oh!

cursor Bot May 21, 2026

Choose a reason for hiding this comment

ZK sumcheck proof removed from Stage 6a prover

Uh oh!

moodlezoup May 22, 2026

Choose a reason for hiding this comment

Uh oh!

moodlezoup May 22, 2026

Choose a reason for hiding this comment

Uh oh!

moodlezoup May 22, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

RadNi commented May 21, 2026 •

edited

Loading