fix(ssa): do not fold ArraySet to MakeArray in simplifier by TomAFrench · Pull Request #11659 · noir-lang/noir

TomAFrench · 2026-02-23T16:25:27Z

Summary

The DFG simplifier was rewriting predicate-dependent ArraySet instructions into predicate-independent MakeArray when the base array and index were constants. This erased the EnableSideEffectsIf guard that ACIR relies on, making conditional writes unconditional.
Remove the ArraySet→MakeArray constant fold since the simplifier has no access to the current side-effects predicate and cannot determine whether the fold is safe.
Add defensive test confirming array_get through a predicated array_set is also not incorrectly simplified (already handled correctly).

Test plan

Regression test array_set_constant_folding_must_respect_side_effects_predicate verifies simplification preserves semantics under both enabled and disabled side effects
Defensive test array_get_not_simplified_through_predicated_array_set confirms existing conservative behavior
All 1167 noirc_evaluator tests pass
All 868 array-related integration tests pass
Updated snapshots in simplifies_array_get_from_previous_array_set_with_make_array and revisit_block_which_dominates_cache

The DFG simplifier was rewriting predicate-dependent ArraySet instructions into predicate-independent MakeArray when the base array and index were constants. This erased the EnableSideEffectsIf guard that ACIR relies on, making conditional writes unconditional and allowing incorrect witness values to satisfy constraints. Remove the ArraySet→MakeArray constant fold since the simplifier has no access to the current side-effects predicate and cannot determine whether the fold is safe.

…_set NOIR-17 was a false positive — the simplifier already correctly refuses to optimize array_get through a same-index predicated array_set when it lacks predicate context (side_effects = None). Add a test to ensure this behavior is preserved.

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Execution Time'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`rollup-checkpoint-merge`	`0.003` s	`0.002` s	`1.50`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Artifact Size'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `396508f`	Previous: `28c3b5f`	Ratio
`rollup-tx-base-public`	`5696.8` KB	`4693.9` KB	`1.21`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Compilation Time'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `29af7df`	Previous: `128c6eb`	Ratio
`rollup-block-root-single-tx`	`2.24` s	`1.74` s	`1.29`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Opcode count'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.10.

Benchmark suite	Current: `396508f`	Previous: `28c3b5f`	Ratio
`rollup-checkpoint-root-single-block`	`1560093` opcodes	`1387956` opcodes	`1.12`
`rollup-checkpoint-root`	`1561286` opcodes	`1389150` opcodes	`1.12`
`rollup-tx-base-private`	`358504` opcodes	`299181` opcodes	`1.20`
`rollup-tx-base-public`	`314113` opcodes	`254789` opcodes	`1.23`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Compilation Memory'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `9081cf0`	Previous: `824f568`	Ratio
`rollup-tx-base-private`	`1700` MB	`1070` MB	`1.59`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Execution Memory'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `9081cf0`	Previous: `824f568`	Ratio
`rollup-tx-base-private`	`688.35` MB	`527.34` MB	`1.31`
`rollup-tx-base-public`	`556.46` MB	`462.99` MB	`1.20`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Test Suite Duration'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`test_report_noir-lang_sha512_`	`14` s	`11` s	`1.27`
`test_report_zkpassport_noir_rsa_`	`2` s	`1` s	`2`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

Brillig functions do not use side-effects predicates, so the ArraySet-to-MakeArray simplification is safe there. Guard the optimization with a runtime check instead of disabling it entirely.

asterite

Looks good!
Should we capture a follow-up issue to perform this optimization as a pass, if we find that the predicate of the MakeArray instruction is the same as the one for the ArraySet? I guess it could go with the existing SSA pass that does this for ArrayGet, and it could maybe revert these regressions.
When I implemented this for ArrayGet I didn't realize this was only an issue in ACIR, so maybe we could also change that code to take that into account (though I don't think it'll further optimize things).

TomAFrench · 2026-02-23T17:15:51Z

Ah, lemme take another look at #11586 to see if I can tie this into your changes a bit more.

…dedicated pass Move the constant-array ArraySet folding out of the simplifier (which lacks predicate context) into the array_set_optimization pass. Extract mutable_array_set as a separate module and fix test expectations.

compiler/noirc_evaluator/src/ssa/opt/array_set.rs

github-actions

ACVM Benchmarks

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`purely_sequential_opcodes`	`251600` ns/iter (`± 615`)	`251446` ns/iter (`± 484`)	`1.00`
`perfectly_parallel_opcodes`	`220410` ns/iter (`± 4690`)	`221480` ns/iter (`± 6603`)	`1.00`
`perfectly_parallel_batch_inversion_opcodes`	`2741913` ns/iter (`± 2640`)	`2745617` ns/iter (`± 6703`)	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

TomAFrench · 2026-02-24T13:42:14Z

I've pulled the renaming of the old array_set_optimization into #11673 so simplify the diff

github-actions

Elaboration Time

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`0.993` s	`0.932` s	`1.07`
`private-kernel-reset`	`0.92` s	`0.984` s	`0.93`
`private-kernel-tail`	`0.994` s	`0.929` s	`1.07`
`rollup-block-root-first-empty-tx`	`1.464` s	`1.48` s	`0.99`
`rollup-block-root-single-tx`	`1.4` s	`1.36` s	`1.03`
`rollup-block-root`	`1.38` s	`1.42` s	`0.97`
`rollup-checkpoint-merge`	`1.468` s	`1.396` s	`1.05`
`rollup-checkpoint-root-single-block`	`1.43` s	`1.36` s	`1.05`
`rollup-checkpoint-root`	`1.4` s	`1.35` s	`1.04`
`rollup-root`	`1.414` s	`1.392` s	`1.02`
`rollup-tx-base-private`	`1.394` s	`1.388` s	`1.00`
`rollup-tx-base-public`	`1.388` s	`1.396` s	`0.99`
`rollup-tx-merge`	`1.398` s	`1.398` s	`1`
`semaphore-depth-10`	`0.157` s	`0.147` s	`1.07`
`sha512-100-bytes`	`0.133` s	`0.136` s	`0.98`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Opcode count

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`19357` opcodes	`19357` opcodes	`1`
`private-kernel-reset`	`85604` opcodes	`85604` opcodes	`1`
`private-kernel-tail`	`8739` opcodes	`8739` opcodes	`1`
`rollup-block-root-first-empty-tx`	`1088` opcodes	`1082` opcodes	`1.01`
`rollup-block-root-single-tx`	`975` opcodes	`969` opcodes	`1.01`
`rollup-block-root`	`2177` opcodes	`2171` opcodes	`1.00`
`rollup-checkpoint-merge`	`1271` opcodes	`1271` opcodes	`1`
`rollup-checkpoint-root-single-block`	`1412521` opcodes	`1387956` opcodes	`1.02`
`rollup-checkpoint-root`	`1413715` opcodes	`1389150` opcodes	`1.02`
`rollup-root`	`1525` opcodes	`1525` opcodes	`1`
`rollup-tx-base-private`	`307656` opcodes	`299181` opcodes	`1.03`
`rollup-tx-base-public`	`263264` opcodes	`254789` opcodes	`1.03`
`rollup-tx-merge`	`1302` opcodes	`1302` opcodes	`1`
`semaphore-depth-10`	`5699` opcodes	`5699` opcodes	`1`
`sha512-100-bytes`	`13173` opcodes	`13173` opcodes	`1`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Artifact Size

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`993.8` KB	`993.8` KB	`1`
`private-kernel-reset`	`2200.5` KB	`2200.5` KB	`1`
`private-kernel-tail`	`489.9` KB	`490` KB	`1.00`
`rollup-block-root-first-empty-tx`	`230.9` KB	`230.7` KB	`1.00`
`rollup-block-root-single-tx`	`234.3` KB	`234.2` KB	`1.00`
`rollup-block-root`	`306.4` KB	`306.3` KB	`1.00`
`rollup-checkpoint-merge`	`365.2` KB	`365.2` KB	`1`
`rollup-checkpoint-root-single-block`	`31154.9` KB	`30782.4` KB	`1.01`
`rollup-checkpoint-root`	`31207.3` KB	`30834.5` KB	`1.01`
`rollup-root`	`389.8` KB	`389.8` KB	`1`
`rollup-tx-base-private`	`5515.8` KB	`5384.3` KB	`1.02`
`rollup-tx-base-public`	`4813.9` KB	`4689.9` KB	`1.03`
`rollup-tx-merge`	`178.5` KB	`178.5` KB	`1`
`semaphore-depth-10`	`488.6` KB	`488.6` KB	`1`
`sha512-100-bytes`	`473.7` KB	`473.7` KB	`1`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Brillig Compilation Time

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`1.54` s	`1.5` s	`1.03`
`private-kernel-reset`	`1.522` s	`1.578` s	`0.96`
`private-kernel-tail`	`1.378` s	`1.294` s	`1.06`
`rollup-block-root-first-empty-tx`	`1.782` s	`1.892` s	`0.94`
`rollup-block-root-single-tx`	`1.7` s	`1.69` s	`1.01`
`rollup-block-root`	`1.74` s	`1.77` s	`0.98`
`rollup-checkpoint-merge`	`1.796` s	`1.7` s	`1.06`
`rollup-checkpoint-root-single-block`	`2.29` s	`2.24` s	`1.02`
`rollup-checkpoint-root`	`2.33` s	`2.3` s	`1.01`
`rollup-root`	`1.85` s	`1.756` s	`1.05`
`rollup-tx-base-private`	`2.078` s	`1.872` s	`1.11`
`rollup-tx-base-public`	`1.962` s	`2.04` s	`0.96`
`rollup-tx-merge`	`1.702` s	`1.702` s	`1`
`semaphore-depth-10`	`0.297` s	`0.282` s	`1.05`
`sha512-100-bytes`	`0.244` s	`0.243` s	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Compilation Time

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`2.97` s	`2.638` s	`1.13`
`private-kernel-reset`	`7.948` s	`8.434` s	`0.94`
`private-kernel-tail`	`2.57` s	`2.622` s	`0.98`
`rollup-block-root-first-empty-tx`	`1.854` s	`1.86` s	`1.00`
`rollup-block-root-single-tx`	`1.75` s	`1.7` s	`1.03`
`rollup-block-root`	`1.8` s	`1.83` s	`0.98`
`rollup-checkpoint-merge`	`1.856` s	`1.778` s	`1.04`
`rollup-checkpoint-root-single-block`	`332` s	`299` s	`1.11`
`rollup-checkpoint-root`	`316` s	`307` s	`1.03`
`rollup-root`	`1.86` s	`1.806` s	`1.03`
`rollup-tx-base-private`	`20.64` s	`19.06` s	`1.08`
`rollup-tx-base-public`	`110` s	`102.8` s	`1.07`
`rollup-tx-merge`	`1.756` s	`1.75` s	`1.00`
`semaphore-depth-10`	`1.193` s	`1.118` s	`1.07`
`sha512-100-bytes`	`1.719` s	`2.051` s	`0.84`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Execution Time

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`0.04` s	`0.041` s	`0.98`
`private-kernel-reset`	`0.174` s	`0.176` s	`0.99`
`private-kernel-tail`	`0.009` s	`0.008` s	`1.13`
`rollup-block-root-first-empty-tx`	`0.003` s	`0.003` s	`1`
`rollup-block-root-single-tx`	`0.003` s	`0.003` s	`1`
`rollup-block-root`	`0.003` s	`0.004` s	`0.75`
`rollup-checkpoint-merge`	`0.003` s	`0.002` s	`1.50`
`rollup-checkpoint-root-single-block`	`10.8` s	`10.7` s	`1.01`
`rollup-checkpoint-root`	`9.85` s	`10.4` s	`0.95`
`rollup-root`	`0.003` s	`0.003` s	`1`
`rollup-tx-base-private`	`0.336` s	`0.332` s	`1.01`
`rollup-tx-base-public`	`0.251` s	`0.252` s	`1.00`
`rollup-tx-merge`	`0.002` s	`0.002` s	`1`
`semaphore-depth-10`	`0.009` s	`0.009` s	`1`
`sha512-100-bytes`	`0.058` s	`0.083` s	`0.70`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Brillig Artifact Size

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`766` KB	`766` KB	`1`
`private-kernel-reset`	`706.3` KB	`706.3` KB	`1`
`private-kernel-tail`	`338.6` KB	`338.6` KB	`1`
`rollup-block-root-first-empty-tx`	`272` KB	`272` KB	`1`
`rollup-block-root-single-tx`	`276.3` KB	`276.3` KB	`1`
`rollup-block-root`	`337.3` KB	`337.3` KB	`1`
`rollup-checkpoint-merge`	`268.4` KB	`268.4` KB	`1`
`rollup-checkpoint-root-single-block`	`539.1` KB	`539.1` KB	`1`
`rollup-checkpoint-root`	`581.6` KB	`581.6` KB	`1`
`rollup-root`	`405.8` KB	`405.8` KB	`1`
`rollup-tx-base-private`	`636.4` KB	`636.4` KB	`1`
`rollup-tx-base-public`	`802.4` KB	`802.4` KB	`1`
`rollup-tx-merge`	`202.8` KB	`202.8` KB	`1`
`semaphore-depth-10`	`2067.4` KB	`2067.4` KB	`1`
`sha512-100-bytes`	`163.5` KB	`163.5` KB	`1`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Brillig Execution Time

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`private-kernel-inner`	`0.031` s	`0.032` s	`0.97`
`private-kernel-reset`	`0.05` s	`0.052` s	`0.96`
`private-kernel-tail`	`0.004` s	`0.005` s	`0.80`
`rollup-block-root-first-empty-tx`	`0.003` s	`0.003` s	`1`
`rollup-block-root-single-tx`	`0.002` s	`0.002` s	`1`
`rollup-block-root`	`0.002` s	`0.003` s	`0.67`
`rollup-checkpoint-merge`	`0.001` s	`0.001` s	`1`
`rollup-root`	`0.002` s	`0.002` s	`1`
`rollup-tx-base-private`	`0.028` s	`0.028` s	`1`
`rollup-tx-base-public`	`0.031` s	`0.032` s	`0.97`
`rollup-tx-merge`	`0.001` s	`0.001` s	`1`
`semaphore-depth-10`	`0.023` s	`0.023` s	`1`
`sha512-100-bytes`	`0.016` s	`0.014` s	`1.14`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Test Suite Duration

Details

Benchmark suite	Current: `75f893b`	Previous: `001f104`	Ratio
`test_report_AztecProtocol_aztec-packages_noir-projects_aztec-nr`	`246` s	`233` s	`1.06`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-contracts`	`183` s	`178` s	`1.03`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_blob`	`239` s	`269` s	`0.89`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_private-kernel-lib`	`492` s	`464` s	`1.06`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_types`	`162` s	`156` s	`1.04`
`test_report_noir-lang_noir-bignum_`	`207` s	`186` s	`1.11`
`test_report_noir-lang_noir_bigcurve_`	`374` s	`335` s	`1.12`
`test_report_noir-lang_sha256_`	`18` s	`18` s	`1`
`test_report_noir-lang_sha512_`	`14` s	`11` s	`1.27`
`test_report_zkpassport_noir-ecdsa_`	`1` s	`1` s	`1`
`test_report_zkpassport_noir_rsa_`	`2` s	`1` s	`2`

This comment was automatically generated by workflow using github-action-benchmark.

aakoshh · 2026-03-04T14:30:48Z

My experiment with merging is in #11760

At the same time we confirmed with @guipublic that assigning incorrect values to the witnesses representing the inputs of poseidon2_permutation, when we will zero out the outputs of the call, should not matter from the circuit correctness perspective.

We were wondering with @asterite whether it is even possible to construct a program that would make this unwanted side effect of unconditionally replacing the result of a disabled array_set even possible, or will the value merger always mask the results.

…pattern

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'ACVM Benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `a012f4a`	Previous: `28c3b5f`	Ratio
`perfectly_parallel_batch_inversion_opcodes`	`2746089` ns/iter (`± 2902`)	`2209309` ns/iter (`± 4639`)	`1.24`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

github-actions

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'ACVM Benchmarks'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: `396508f`	Previous: `28c3b5f`	Ratio
`perfectly_parallel_batch_inversion_opcodes`	`2743531` ns/iter (`± 2148`)	`2209309` ns/iter (`± 4639`)	`1.24`

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

…raySet` (#11760)

asterite · 2026-03-05T15:07:51Z

One thing I'm noticing is that when we compile hashmap in master, we just get a warning:

warning: unused type alias MyMap
    ┌─ src/main.nr:227:6
    │
227 │ type MyMap = HashMap<u8, u32, 10, BuildHasherDefault<Poseidon2Hasher>>;
    │      ----- unused type alias

But when we compile it in this branch we get bugs (in addition to the warning above):

bug: Input to Brillig function is in a separate subgraph to output
    ┌─ std/array/mod.nr:284:31
    │
284 │         let sorted = unsafe { quicksort::quicksort(self, ordering) };
    │                               ------------------------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
    │
    = Call stack:
      1. src/main.nr:38:5
      2. src/main.nr:198:32
      3. std/array/mod.nr:284:31

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:106:17
     8. std/field/bn254.nr:82:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:106:17
     8. std/field/bn254.nr:83:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:109:17
     8. std/field/bn254.nr:82:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:109:17
     8. std/field/bn254.nr:83:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:106:17
     8. std/field/bn254.nr:82:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:106:17
     8. std/field/bn254.nr:83:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:109:17
     8. std/field/bn254.nr:82:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/field/bn254.nr:40:22
   │
40 │         let borrow = lte_hint(alo, blo);
   │                      ------------------ There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:289:21
     4. src/main.nr:25:68
     5. std/field/mod.nr:199:13
     6. std/field/bn254.nr:117:5
     7. std/field/bn254.nr:109:17
     8. std/field/bn254.nr:83:23
     9. std/field/bn254.nr:68:13
     10. std/field/bn254.nr:40:22

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/array/check_shuffle.nr:49:23
   │
49 │             let idx = __get_index(shuffle_indices, i);
   │                       ------------------------------- There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:293:13
     4. std/array/check_shuffle.nr:49:23

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/array/check_shuffle.nr:49:23
   │
49 │             let idx = __get_index(shuffle_indices, i);
   │                       ------------------------------- There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:293:13
     4. std/array/check_shuffle.nr:49:23

bug: Input to Brillig function is in a separate subgraph to output
   ┌─ std/array/check_shuffle.nr:49:23
   │
49 │             let idx = __get_index(shuffle_indices, i);
   │                       ------------------------------- There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability
   │
   = Call stack:
     1. src/main.nr:38:5
     2. src/main.nr:198:32
     3. std/array/mod.nr:293:13
     4. std/array/check_shuffle.nr:49:23

I wonder if those bugs are legitimate or this PR introduces a bug...

Then it seems #11773 doesn't fix the regression in hashmap, thought it does seem to fix the regression that was happening in Aztec-Packages, so maybe there was a bug in hashmap after all. I'm going to check now why those bugs happen.

aakoshh · 2026-03-05T16:06:52Z

There is no path from the output of this Brillig call to either return values or inputs of the circuit, which creates an independent subgraph. This is quite likely a soundness vulnerability

It is true that the results of the call it points at are not constrained against any input, nor are they returned:

noir/noir_stdlib/src/field/bn254.nr

Lines 33 to 48 in 8ed9293

    
           fn assert_gt_limbs(a: (Field, Field), b: (Field, Field)) { 
        
               let (alo, ahi) = a; 
        
               let (blo, bhi) = b; 
        
               // Safety: borrow is enforced to be boolean due to its type. 
        
               // if borrow is 0, it asserts that (alo > blo && ahi >= bhi) 
        
               // if borrow is 1, it asserts that (alo <= blo && ahi > bhi) 
        
               unsafe { 
        
                   let borrow = lte_hint(alo, blo); 
        
                   let rlo = alo - blo - 1 + (borrow as Field) * TWO_POW_128; 
        
                   let rhi = ahi - bhi - (borrow as Field); 
        
                   rlo.assert_max_bit_size::<128>(); 
        
                   rhi.assert_max_bit_size::<128>(); 
        
               } 
        
           }

TomAFrench · 2026-03-05T16:25:03Z

It is true that the results of the call it points at are not constrained against any input, nor are they returned:

This is only half true as the range constraints are expected to constrain the value of borrow.

aakoshh · 2026-03-05T16:25:19Z

It is also true that hashmap sorts an array that is a constant defined in the test:

noir/test_programs/execution_success/hashmap/src/main.nr

Lines 160 to 166 in 8ed9293

    
           let mut hashmap = ALLOCATE_HASHMAP(); 
        
           hashmap.insert(2, 3); 
        
           hashmap.insert(5, 7); 
        
           hashmap.insert(11, 13); 
        
           let keys: [K; 3] = cut(hashmap.keys()).sort_via(K_CMP);

Maybe this is whey the vectors test has this kind of stuff:

noir/test_programs/execution_success/vectors/src/main.nr

Line 57 in 8ed9293

    
           // The parameters to this function must come from witness values (inputs to main)

aakoshh · 2026-03-05T16:33:47Z

This is only half true as the range constraints are expected to constrain the value of borrow.

It looks like the method that considers things connected only looks for return values and inputs:

noir/compiler/noirc_evaluator/src/ssa/checks/check_for_underconstrained_values.rs

Lines 703 to 715 in 8ed9293

    
               /// Find sets that contain input or output value of the function 
        
               /// 
        
               /// Goes through each set of connected ValueIds and see if function arguments or return values are in the set 
        
               fn find_sets_connected_to_function_inputs_or_outputs( 
        
                   &self, 
        
                   function: &Function, 
        
               ) -> BTreeSet<usize> { 
        
                   let returns = function.returns().unwrap_or_default(); 
        
                   let variable_parameters_and_return_values = function 
        
                       .parameters() 
        
                       .iter() 
        
                       .chain(returns) 
        
                       .filter(|id| !is_numeric_constant(function, **id));

Where it connects range constraints, ~~I only see them as things that need to be connected~~:

noir/compiler/noirc_evaluator/src/ssa/checks/check_for_underconstrained_values.rs

Line 832 in 8ed9293

self.value_sets.push(instruction_arguments_and_results);

Ah no, they are actually not connected 🙈 :

noir/compiler/noirc_evaluator/src/ssa/checks/check_for_underconstrained_values.rs

Lines 808 to 812 in 8ed9293

    
           Value::Intrinsic(intrinsic) => match intrinsic { 
        
               Intrinsic::ApplyRangeConstraint 
        
               | Intrinsic::AssertConstant 
        
               | Intrinsic::AsWitness 
        
               | Intrinsic::IsUnconstrained => {}

asterite · 2026-03-05T19:56:50Z

It seems the hashmap regression is gone! 🎉 And then #11773 optimized the case that Aztec-Packages was bumping into.

aakoshh · 2026-03-06T11:04:49Z

It seems the hashmap regression is gone!

Before it was gone, the SSA could not eliminate the Brillig calls as constants, the array_sets where not turned into make_array, and then the call to Brillig stayed in. After some further changes, they were all make_array, a constant input to the unconstrained quick sort related calls, which were then constant folded, and with no Brillig calls left, no disconnected subgraphs either.

TomAFrench · 2026-03-06T11:46:38Z

Great work on this @aakoshh and @asterite, thank you!

TomAFrench added 3 commits February 23, 2026 16:19

chore: update test name to match new functionality

0298f55

github-actions bot reviewed Feb 23, 2026

View reviewed changes

fix(ssa): only skip ArraySet-to-MakeArray folding for ACIR functions

9081cf0

Brillig functions do not use side-effects predicates, so the ArraySet-to-MakeArray simplification is safe there. Guard the optimization with a runtime check instead of disabling it entirely.

TomAFrench requested a review from a team February 23, 2026 16:51

TomAFrench added the bench-show Display benchmark results on PR label Feb 23, 2026

asterite approved these changes Feb 23, 2026

View reviewed changes

TomAFrench marked this pull request as draft February 24, 2026 12:32

asterite reviewed Feb 24, 2026

View reviewed changes

compiler/noirc_evaluator/src/ssa/opt/array_set.rs Outdated Show resolved Hide resolved

Merge branch 'master' into tf/fix-array-set-predicate-simplification

e984fa4

github-actions bot reviewed Feb 24, 2026

View reviewed changes

aakoshh mentioned this pull request Mar 4, 2026

fix(ssa): Use the ValueMerger on the element instead of keeping ArraySet #11760

Merged

5 tasks

aakoshh and others added 2 commits March 4, 2026 14:57

Add a regression test to start tracking the opcode count of this use …

770555f

…pattern

Add a test to remove_if_else to show a potential optimization

a012f4a

github-actions bot reviewed Mar 4, 2026

View reviewed changes

Fix test

396508f

github-actions bot reviewed Mar 4, 2026

View reviewed changes

asterite mentioned this pull request Mar 4, 2026

chore: array_set_window_optimization #11773

Merged

5 tasks

fix(ssa): Use the ValueMerger on the element instead of keeping `Ar…

c865890

…raySet` (#11760)

Merge origin/master into tf/fix-array-set-predicate-simplification

2bcf3bb

This was referenced Mar 5, 2026

chore: let simple_optimization optimize values whose dependent values changed #11784

Merged

chore: optimize array_get through unconditional array_set #11785

Merged

asterite added 3 commits March 5, 2026 16:03

Merge branch 'master' into tf/fix-array-set-predicate-simplification

a06c850

Remove unrelated changes

ce6269f

Merge branch 'master' into tf/fix-array-set-predicate-simplification

44e5b1d

Raise timeout

75f893b

aakoshh marked this pull request as ready for review March 6, 2026 11:04

TomAFrench added this pull request to the merge queue Mar 6, 2026

Merged via the queue into master with commit ebb5ff1 Mar 6, 2026
135 checks passed

TomAFrench deleted the tf/fix-array-set-predicate-simplification branch March 6, 2026 12:27

noirwhal mentioned this pull request Mar 6, 2026

chore: Release Noir(1.0.0-beta.20) #11611

Open

Conversation

TomAFrench commented Feb 23, 2026

Summary

Test plan

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

⚠️ Performance Alert ⚠️

Uh oh!

asterite left a comment

Choose a reason for hiding this comment

Uh oh!

TomAFrench commented Feb 23, 2026

Uh oh!

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

ACVM Benchmarks

Uh oh!

TomAFrench commented Feb 24, 2026

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Elaboration Time

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Opcode count

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Artifact Size

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Brillig Compilation Time

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Compilation Time

Uh oh!

github-actions bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

github-actions bot left a comment •

edited

Loading

aakoshh commented Mar 5, 2026 •

edited

Loading