fix(ssa): Constant fold Brillig calls using the SSA interpreter by vezenovm · Pull Request #9365 · noir-lang/noir

vezenovm · 2025-07-30T23:31:12Z

Description

Problem*

Resolves #9015
Resolves #9097

This also resolves a panic from a Noir sync when attempting to debug log an array value (the same error as #9097)

Let's take the snippet from #9097:

fn main() {
    // Safety:
    let result = unsafe { foo([0]) };
    println(result);
}

unconstrained fn foo(bytes: [u8; 1]) -> Field {
    for _ in 0..bytes.len() {
        let _ = bytes[0];
    }
    0
}

We can see the following SSA when we fold constants with Brillig:

Details

After Inlining Brillig Calls (1) (step 42):
acir(inline) impure fn main f0 {
  b0():
    v1 = make_array [u8 0] : [u8; 1]                    // src/main.nr:3:32
    call f2(u1 1, Field 0)                              // std/lib.nr:40:9
    return
}
brillig(inline) predicate_pure fn foo f1 {
  b0(v0: [u8; 1]):
    inc_rc v0                                           // src/main.nr:8:17
    return Field 0
}
brillig(inline) impure fn print_unconstrained f2 {
  b0(v0: u1, v1: Field):
    v13 = make_array b"{\"kind\":\"field\"}"            // std/lib.nr:40:9
    call print(v0, v1, v13, u1 0)                       // std/lib.nr:34:5
    return
}

Remove Unreachable Instructions (2) (step 43): 0 ms
After Remove Unreachable Instructions (2) (step 43):
acir(inline) impure fn main f0 {
  b0():
    v1 = make_array [u8 0] : [u8; 1]                    // src/main.nr:3:32
    call f1(u1 1, Field 0)                              // std/lib.nr:40:9
    return
}
brillig(inline) predicate_pure fn print_unconstrained f1 {
  b0(v0: u1, v1: Field):
    v13 = make_array b"{\"kind\":\"field\"}"            // std/lib.nr:40:9
    call print(v0, v1, v13, u1 0)                       // std/lib.nr:34:5
    return
}

We can see that one Brillig function was folding into f0. However, the print remains. However, the Brillig context structure contains all the artifacts already has a set mapping for function ids to Brillig artifacts. Thus, when we remove functions after Brillig gen and change their IDs (e.g., like when normalizing) we end up fetching an artifact for the wrong function when it comes times to generate an entry point. So we were fetching

Summary*

I realized this could have been fixed for #9097 by simply updating the IDs point to the Brillig artifacts as part of normalization. However, this felt brittle and I felt the problem laid out above still feels like it could come up elsewhere.

Aside the benefit of avoiding extra Brillig compilation, switching to using the SSA interpreter for constant folding Brillig entry points also avoids these kinds of situations where Brillig context needs to be updated due to secondary SSA passes post Brillig gen.

Additional Context

Documentation*

Check one:

No documentation needed.
Documentation included in this PR.
[For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

I have tested the changes locally.
I have formatted the changes with Prettier and/or cargo fmt on default settings.

vezenovm · 2025-07-30T23:43:17Z

regression_7451 +2 ❌ +200.00% +2,848 ❌ +16752.94%
inline_decompose_hint_brillig_call +15 ❌ +1500.00% +74 ❌ +435.29%
strings +1 ❌ +3.45% +3 ❌ +9.09%

Hmm looks like we are getting some regressions. I was not expecting any regression so I will have to investigate before we can merge.

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: `c987274`	Previous: `ee2ac1a`	Ratio
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_rollup-lib`	`909` s	`682` s	`1.33`
`test_report_noir-lang_sha512_`	`216` s	`13` s	`16.62`
`test_report_zkpassport_noir_rsa_`	`1` s	`0` s	`+∞`

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

CC: @TomAFrench

github-actions · 2025-07-31T14:55:44Z

Changes to Brillig bytecode sizes

Generated at commit: de310253ba179dc73ad18b4b234ed8381668b804, compared to commit: ee2ac1a04e85134079fbe557d4029d3f64aff7b6

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
global_var_regression_simple_inliner_max	-1 ✅	-2.56%
global_var_regression_simple_inliner_zero	-1 ✅	-2.56%
binary_operator_overloading_inliner_max	-26 ✅	-6.79%

Full diff report 👇

Program	Brillig opcodes (+/-)	%
nested_array_in_slice_inliner_max	953 (-2)	-0.21%
nested_array_in_slice_inliner_min	809 (-2)	-0.25%
nested_array_in_slice_inliner_zero	809 (-2)	-0.25%
array_oob_regression_7965_inliner_max	259 (-2)	-0.77%
array_oob_regression_7965_inliner_min	259 (-2)	-0.77%
array_oob_regression_7965_inliner_zero	259 (-2)	-0.77%
higher_order_functions_inliner_zero	625 (-5)	-0.79%
nested_array_dynamic_inliner_max	1,723 (-15)	-0.86%
nested_array_dynamic_inliner_min	1,431 (-15)	-1.04%
nested_array_dynamic_inliner_zero	1,431 (-15)	-1.04%
global_consts_inliner_max	197 (-4)	-1.99%
global_consts_inliner_zero	197 (-4)	-1.99%
global_var_regression_simple_inliner_max	38 (-1)	-2.56%
global_var_regression_simple_inliner_zero	38 (-1)	-2.56%
binary_operator_overloading_inliner_max	357 (-26)	-6.79%

github-actions · 2025-07-31T14:55:47Z

Changes to number of Brillig opcodes executed

Generated at commit: de310253ba179dc73ad18b4b234ed8381668b804, compared to commit: ee2ac1a04e85134079fbe557d4029d3f64aff7b6

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
higher_order_functions_inliner_zero	-12 ✅	-1.03%
global_var_regression_simple_inliner_max	-1 ✅	-3.23%
global_var_regression_simple_inliner_zero	-1 ✅	-3.23%

Full diff report 👇

Program	Brillig opcodes (+/-)	%
poseidon_bn254_hash_width_3_inliner_max	137,347 (+234)	+0.17%
regression_5252_inliner_max	760,074 (+1,200)	+0.16%
poseidonsponge_x5_254_inliner_max	152,905 (+240)	+0.16%
poseidon_bn254_hash_width_3_inliner_zero	167,759 (+234)	+0.14%
global_consts_inliner_max	1,371 (-4)	-0.29%
global_consts_inliner_zero	1,371 (-4)	-0.29%
array_oob_regression_7965_inliner_max	313 (-1)	-0.32%
array_oob_regression_7965_inliner_min	313 (-1)	-0.32%
array_oob_regression_7965_inliner_zero	313 (-1)	-0.32%
nested_array_dynamic_inliner_min	2,828 (-10)	-0.35%
nested_array_dynamic_inliner_zero	2,828 (-10)	-0.35%
nested_array_dynamic_inliner_max	2,644 (-10)	-0.38%
higher_order_functions_inliner_zero	1,153 (-12)	-1.03%
global_var_regression_simple_inliner_max	30 (-1)	-3.23%
global_var_regression_simple_inliner_zero	30 (-1)	-3.23%

compiler/noirc_evaluator/src/ssa/interpreter/mod.rs

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

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

compiler/noirc_evaluator/src/ssa/mod.rs

github-actions

ACVM Benchmarks

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`purely_sequential_opcodes`	`246544` ns/iter (`± 770`)	`246963` ns/iter (`± 660`)	`1.00`
`perfectly_parallel_opcodes`	`218489` ns/iter (`± 3724`)	`219542` ns/iter (`± 2180`)	`1.00`
`perfectly_parallel_batch_inversion_opcodes`	`2779355` ns/iter (`± 2092`)	`2776812` ns/iter (`± 1780`)	`1.00`

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

github-actions

Compilation Time

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`1.794` s	`1.686` s	`1.06`
`private-kernel-reset`	`7.714` s	`7.822` s	`0.99`
`private-kernel-tail`	`1.378` s	`1.296` s	`1.06`
`rollup-base-private`	`15.94` s	`15.34` s	`1.04`
`rollup-base-public`	`13.64` s	`13.32` s	`1.02`
`rollup-block-root-empty`	`21.5` s	`20.48` s	`1.05`
`rollup-block-root-single-tx`	`231` s	`189` s	`1.22`
`rollup-block-root`	`229` s	`211` s	`1.09`
`rollup-merge`	`1.314` s	`1.378` s	`0.95`
`rollup-root`	`1.438` s	`1.456` s	`0.99`
`semaphore-depth-10`	`0.806` s	`0.759` s	`1.06`
`sha512-100-bytes`	`1.788` s	`1.747` s	`1.02`

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

github-actions

Artifact Size

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`709.1` KB	`709.1` KB	`1`
`private-kernel-reset`	`2033.1` KB	`2033.1` KB	`1`
`private-kernel-tail`	`535.9` KB	`535.9` KB	`1`
`rollup-base-private`	`4313.9` KB	`4314.6` KB	`1.00`
`rollup-base-public`	`3327.1` KB	`3327.1` KB	`1`
`rollup-block-root-empty`	`3841.7` KB	`3842.8` KB	`1.00`
`rollup-block-root-single-tx`	`30722.2` KB	`30722.5` KB	`1.00`
`rollup-block-root`	`30771.3` KB	`30771.4` KB	`1.00`
`rollup-merge`	`183` KB	`183` KB	`1`
`rollup-root`	`388.9` KB	`388.9` KB	`1`
`semaphore-depth-10`	`632.1` KB	`632.1` KB	`1`
`sha512-100-bytes`	`525.2` KB	`525.2` KB	`1`

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

github-actions

Execution Time

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`0.014` s	`0.014` s	`1`
`private-kernel-reset`	`0.155` s	`0.154` s	`1.01`
`private-kernel-tail`	`0.01` s	`0.01` s	`1`
`rollup-base-private`	`0.264` s	`0.264` s	`1`
`rollup-base-public`	`0.164` s	`0.162` s	`1.01`
`rollup-block-root`	`13.2` s	`12.9` s	`1.02`
`rollup-merge`	`0.002` s	`0.002` s	`1`
`rollup-root`	`0.004` s	`0.004` s	`1`
`semaphore-depth-10`	`0.019` s	`0.019` s	`1`
`sha512-100-bytes`	`0.101` s	`0.096` s	`1.05`

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

github-actions

Opcode count

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`14792` opcodes	`14792` opcodes	`1`
`private-kernel-reset`	`68868` opcodes	`68868` opcodes	`1`
`private-kernel-tail`	`11176` opcodes	`11176` opcodes	`1`
`rollup-base-private`	`221419` opcodes	`221419` opcodes	`1`
`rollup-base-public`	`159972` opcodes	`159972` opcodes	`1`
`rollup-block-root-empty`	`68110` opcodes	`68110` opcodes	`1`
`rollup-block-root-single-tx`	`963865` opcodes	`963865` opcodes	`1`
`rollup-block-root`	`965153` opcodes	`965153` opcodes	`1`
`rollup-merge`	`1411` opcodes	`1411` opcodes	`1`
`rollup-root`	`2631` opcodes	`2631` opcodes	`1`
`semaphore-depth-10`	`5763` opcodes	`5763` opcodes	`1`
`sha512-100-bytes`	`13173` opcodes	`13173` opcodes	`1`

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

github-actions

Execution Memory

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`205.02` MB	`205.02` MB	`1`
`private-kernel-reset`	`241.84` MB	`241.84` MB	`1`
`private-kernel-tail`	`194.26` MB	`194.26` MB	`1`
`rollup-base-private`	`498.48` MB	`498.48` MB	`1`
`rollup-base-public`	`431.12` MB	`431.12` MB	`1`
`rollup-block-root`	`1500` MB	`1500` MB	`1`
`rollup-merge`	`326.44` MB	`326.44` MB	`1`
`rollup-root`	`328.85` MB	`328.85` MB	`1`
`semaphore_depth_10`	`69.51` MB	`69.51` MB	`1`
`sha512_100_bytes`	`55` MB	`55` MB	`1`

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

github-actions

Compilation Memory

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`private-kernel-inner`	`238.23` MB	`238.26` MB	`1.00`
`private-kernel-reset`	`547.42` MB	`547.41` MB	`1.00`
`private-kernel-tail`	`212.14` MB	`212.14` MB	`1`
`rollup-base-private`	`1340` MB	`1340` MB	`1`
`rollup-base-public`	`1420` MB	`1420` MB	`1`
`rollup-block-root-empty`	`1140` MB	`1090` MB	`1.05`
`rollup-block-root-single-tx`	`10670` MB	`9340` MB	`1.14`
`rollup-block-root`	`10670` MB	`9350` MB	`1.14`
`rollup-merge`	`330.52` MB	`330.52` MB	`1`
`rollup-root`	`339.74` MB	`339.69` MB	`1.00`
`semaphore_depth_10`	`104.78` MB	`104.76` MB	`1.00`
`sha512_100_bytes`	`234.74` MB	`234.84` MB	`1.00`

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

github-actions

Test Suite Duration

Details

Benchmark suite	Current: `c987274`	Previous: `ee2ac1a`	Ratio
`test_report_AztecProtocol_aztec-packages_noir-projects_aztec-nr`	`103` s	`103` s	`1`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-contracts`	`114` s	`113` s	`1.01`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_blob`	`194` s	`168` s	`1.15`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_private-kernel-lib`	`209` s	`220` s	`0.95`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_reset-kernel-lib`	`34` s	`32` s	`1.06`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_rollup-lib`	`909` s	`682` s	`1.33`
`test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_types`	`100` s	`101` s	`0.99`
`test_report_noir-lang_noir-bignum_`	`399` s	`413` s	`0.97`
`test_report_noir-lang_noir_bigcurve_`	`353` s	`299` s	`1.18`
`test_report_noir-lang_noir_json_parser_`	`55` s
`test_report_noir-lang_sha512_`	`216` s	`13` s	`16.62`
`test_report_zkpassport_noir-ecdsa_`	`1` s	`2` s	`0.50`
`test_report_zkpassport_noir_rsa_`	`1` s	`0` s	`+∞`

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

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

asterite

Looks good!

vezenovm · 2025-07-31T20:57:53Z

sha512_100_bytes +221 ❌ +1.68% +443 ❌ +1.12%

Looks like we still have a minor regression here. Most likely due to us not folding Brillig calls as we constant fold all the other instructions. I'm going to look into a fixpoint loop quickly to see if it removes the regression. Otherwise, I'm going to merge as this regression is small and we in fact see improvements in other tests. Optimizing this can then come in a separate PR as it would most likely require changing the interpreter interface to not borrow Ssa so that we can evaluate calls during the main constant_folding pass.

…nder constant_folding

vezenovm · 2025-07-31T21:31:01Z

I ended up making a fixed point loop in constant_folding. The loop is contained in a top-level method in constant_folding and fold_constant_brillig_calls is now a submodule of constant_folding.

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: `c987274`	Previous: `ee2ac1a`	Ratio
`rollup-block-root-single-tx`	`231` s	`189` s	`1.22`

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

CC: @TomAFrench

vezenovm · 2025-07-31T21:47:51Z

I ended up making a fixed point loop in constant_folding. The loop is contained in a top-level method in constant_folding and fold_constant_brillig_calls is now a submodule of constant_folding.

ugh, we seem to be getting large time regressions in the sha512 library https://github.com/noir-lang/noir/actions/runs/16660549208/job/47156670574?pr=9365. Perhaps we do just have to bite the bullet and change the interpreter interface.

vezenovm · 2025-07-31T22:07:33Z

ugh, we seem to be getting large time regressions in the sha512 library https://github.com/noir-lang/noir/actions/runs/16660549208/job/47156670574?pr=9365. Perhaps we do just have to bite the bullet and change the interpreter interface.

If we want to get this fix in and are ok with a small circuit size regression I will just revert to e83fe33 where everything is green. Otherwise, I will take a look at how we can change the interpreter interface to avoid borrowing the entire Ssa for this case.

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: `6ccd460`	Previous: `d7e2c17`	Ratio
`private-kernel-inner`	`0.02` s	`0.016` s	`1.25`

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

CC: @TomAFrench

TomAFrench · 2025-08-14T10:57:28Z

compiler/noirc_evaluator/src/ssa/opt/constant_folding/brillig_calls.rs

+//! This improves constant folding across function boundaries, particularly for small utility
+//! Brillig functions that can be inlined
+//!
+//! Note: Only calls to Brillig functions from ACIR functions (Brillig entry points) are evaluated here.


This doesn't seem correct based on the changes to brillig bytecode sizes shown in the benchmarls.

vezenovm · 2025-09-02T13:47:05Z

Superseded by #9655

vezenovm added 4 commits July 30, 2025 22:08

initial changes and new constant_evaluation pass

b6e1863

snaps

ad05885

cleanup and rename

2be2dec

Merge branch 'master' into mv/const-eval-with-ssa-interpreter

6ecb539

vezenovm marked this pull request as draft July 30, 2025 23:41

github-actions bot reviewed Jul 30, 2025

View reviewed changes

vezenovm added 4 commits July 30, 2025 23:51

simplify constant folding pass, and cleanup

5bba10f

one more clippy

69ce350

more snaps

6dfeae5

fold constants after folding brillig calls

1fc3288

vezenovm added the bench-show Display benchmark results on PR label Jul 31, 2025