feat(performance): brillig array set optimization by asterite · Pull Request #8550 · noir-lang/noir

asterite · 2025-05-16T13:32:18Z

Description

Problem

Follow up to #8536

Summary

Similar logic to the one of ArraySet.

While doing this I introduced a few refactors, for example just having one insert_array_get and insert_array_set on a FunctionBuilder.

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.

github-actions · 2025-05-16T13:52:24Z

Changes to number of Brillig opcodes executed

Generated at commit: 9d4524a9acd0f733e2bcf230d5c4408e5c107efc, compared to commit: 8f70d14a03186124e12d6bce9a8b58e30d515b22

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
fold_call_witness_condition_inliner_min	+1 ❌	+1.49%
fold_call_witness_condition_inliner_max	+1 ❌	+1.45%
fold_call_witness_condition_inliner_zero	+1 ❌	+1.45%
brillig_arrays_inliner_max	-2 ✅	-1.42%
brillig_arrays_inliner_zero	-2 ✅	-1.42%
array_rc_regression_7842_inliner_min	-3 ✅	-1.68%
mutate_array_copy_inliner_min	-3 ✅	-1.69%
fold_distinct_return_inliner_min	-2 ✅	-2.70%

Full diff report 👇

Program	Brillig opcodes (+/-)	%
fold_call_witness_condition_inliner_min	68 (+1)	+1.49%
fold_call_witness_condition_inliner_max	70 (+1)	+1.45%
fold_call_witness_condition_inliner_zero	70 (+1)	+1.45%
brillig_cow_regression_inliner_min	197,026 (-5)	-0.00%
brillig_cow_regression_inliner_max	194,345 (-5)	-0.00%
brillig_cow_regression_inliner_zero	194,345 (-5)	-0.00%
7_function_inliner_max	2,060 (-1)	-0.05%
conditional_1_inliner_min	1,922 (-2)	-0.10%
conditional_1_inliner_zero	1,922 (-2)	-0.10%
conditional_1_inliner_max	1,882 (-2)	-0.11%
poseidonsponge_x5_254_inliner_min	183,756 (-360)	-0.20%
regression_5252_inliner_min	917,134 (-1,812)	-0.20%
poseidonsponge_x5_254_inliner_zero	181,839 (-360)	-0.20%
regression_5252_inliner_zero	905,221 (-1,812)	-0.20%
poseidon_bn254_hash_width_3_inliner_min	168,130 (-351)	-0.21%
poseidon_bn254_hash_width_3_inliner_zero	166,274 (-351)	-0.21%
fold_numeric_generic_poseidon_inliner_min	4,721 (-10)	-0.21%
no_predicates_numeric_generic_poseidon_inliner_min	4,693 (-10)	-0.21%
fold_numeric_generic_poseidon_inliner_zero	4,453 (-10)	-0.22%
no_predicates_numeric_generic_poseidon_inliner_zero	4,453 (-10)	-0.22%
poseidonsponge_x5_254_inliner_max	150,769 (-360)	-0.24%
regression_5252_inliner_max	749,365 (-1,812)	-0.24%
global_slice_rc_regression_8259_inliner_min	393 (-1)	-0.25%
poseidon_bn254_hash_width_3_inliner_max	135,862 (-351)	-0.26%
fold_numeric_generic_poseidon_inliner_max	3,765 (-10)	-0.26%
no_predicates_numeric_generic_poseidon_inliner_max	3,765 (-10)	-0.26%
fold_2_to_17_inliner_min	1,069,076 (-3,106)	-0.29%
bench_2_to_17_inliner_min	562,066 (-1,633)	-0.29%
fold_2_to_17_inliner_zero	1,031,654 (-3,106)	-0.30%
bench_2_to_17_inliner_zero	468,908 (-1,633)	-0.35%
bench_2_to_17_inliner_max	429,693 (-1,633)	-0.38%
fold_2_to_17_inliner_max	789,307 (-3,106)	-0.39%
regression_struct_array_conditional_inliner_max	1,130 (-6)	-0.53%
regression_struct_array_conditional_inliner_min	1,130 (-6)	-0.53%
regression_struct_array_conditional_inliner_zero	1,130 (-6)	-0.53%
global_array_rc_regression_8259_inliner_min	188 (-1)	-0.53%
brillig_pedersen_inliner_min	790 (-5)	-0.63%
pedersen_check_inliner_min	790 (-5)	-0.63%
array_if_cond_simple_inliner_max	308 (-2)	-0.65%
array_if_cond_simple_inliner_min	308 (-2)	-0.65%
array_if_cond_simple_inliner_zero	308 (-2)	-0.65%
simple_shield_inliner_min	2,171 (-15)	-0.69%
simple_shield_inliner_zero	2,149 (-15)	-0.69%
simple_shield_inliner_max	2,117 (-15)	-0.70%
brillig_pedersen_inliner_zero	633 (-5)	-0.78%
pedersen_check_inliner_zero	633 (-5)	-0.78%
fold_complex_outputs_inliner_min	718 (-6)	-0.83%
brillig_pedersen_inliner_max	588 (-5)	-0.84%
pedersen_check_inliner_max	588 (-5)	-0.84%
fold_complex_outputs_inliner_zero	645 (-6)	-0.92%
merkle_insert_inliner_zero	3,084 (-30)	-0.96%
merkle_insert_inliner_max	2,948 (-30)	-1.01%
merkle_insert_inliner_min	2,848 (-30)	-1.04%
pedersen_hash_inliner_min	462 (-5)	-1.07%
brillig_arrays_inliner_min	167 (-2)	-1.18%
pedersen_hash_inliner_max	397 (-5)	-1.24%
pedersen_hash_inliner_zero	397 (-5)	-1.24%
fold_complex_outputs_inliner_max	440 (-6)	-1.35%
brillig_arrays_inliner_max	139 (-2)	-1.42%
brillig_arrays_inliner_zero	139 (-2)	-1.42%
array_rc_regression_7842_inliner_min	176 (-3)	-1.68%
mutate_array_copy_inliner_min	175 (-3)	-1.69%
fold_distinct_return_inliner_min	72 (-2)	-2.70%

github-actions · 2025-05-16T13:52:25Z

Changes to Brillig bytecode sizes

Generated at commit: 9d4524a9acd0f733e2bcf230d5c4408e5c107efc, compared to commit: 8f70d14a03186124e12d6bce9a8b58e30d515b22

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
array_if_cond_simple_inliner_max	-2 ✅	-1.54%
array_if_cond_simple_inliner_min	-2 ✅	-1.54%
array_if_cond_simple_inliner_zero	-2 ✅	-1.54%
fold_complex_outputs_inliner_max	-6 ✅	-1.69%
pedersen_hash_inliner_min	-5 ✅	-1.71%
brillig_arrays_inliner_max	-2 ✅	-1.72%
brillig_arrays_inliner_zero	-2 ✅	-1.72%
merkle_insert_inliner_max	-10 ✅	-1.80%
pedersen_hash_inliner_max	-5 ✅	-1.87%
pedersen_hash_inliner_zero	-5 ✅	-1.87%
fold_distinct_return_inliner_min	-2 ✅	-2.13%

Full diff report 👇

Program	Brillig opcodes (+/-)	%
hashmap_inliner_min	8,841 (-1)	-0.01%
hashmap_inliner_zero	7,828 (-1)	-0.01%
uhashmap_inliner_min	7,561 (-1)	-0.01%
uhashmap_inliner_zero	6,897 (-1)	-0.01%
poseidonsponge_x5_254_inliner_min	3,020 (-3)	-0.10%
poseidonsponge_x5_254_inliner_zero	2,893 (-3)	-0.10%
regression_5252_inliner_min	3,396 (-4)	-0.12%
poseidon_bn254_hash_width_3_inliner_max	4,856 (-6)	-0.12%
regression_5252_inliner_zero	3,226 (-4)	-0.12%
poseidon_bn254_hash_width_3_inliner_min	4,762 (-6)	-0.13%
poseidon_bn254_hash_width_3_inliner_zero	4,443 (-6)	-0.13%
poseidonsponge_x5_254_inliner_max	3,680 (-6)	-0.16%
regression_5252_inliner_max	3,966 (-7)	-0.18%
7_function_inliner_max	526 (-1)	-0.19%
fold_2_to_17_inliner_min	363 (-1)	-0.27%
fold_2_to_17_inliner_zero	355 (-1)	-0.28%
bench_2_to_17_inliner_min	332 (-1)	-0.30%
fold_numeric_generic_poseidon_inliner_min	597 (-2)	-0.33%
no_predicates_numeric_generic_poseidon_inliner_min	577 (-2)	-0.35%
bench_2_to_17_inliner_max	276 (-1)	-0.36%
bench_2_to_17_inliner_zero	258 (-1)	-0.39%
brillig_cow_regression_inliner_min	1,267 (-5)	-0.39%
no_predicates_numeric_generic_poseidon_inliner_zero	503 (-2)	-0.40%
fold_numeric_generic_poseidon_inliner_zero	503 (-2)	-0.40%
brillig_cow_regression_inliner_max	1,202 (-5)	-0.41%
brillig_cow_regression_inliner_zero	1,202 (-5)	-0.41%
fold_2_to_17_inliner_max	478 (-2)	-0.42%
global_slice_rc_regression_8259_inliner_min	234 (-1)	-0.43%
reference_counts_inliner_max_inliner_zero	847 (-4)	-0.47%
reference_counts_inliner_0_inliner_zero	847 (-4)	-0.47%
reference_counts_inliner_min_inliner_zero	847 (-4)	-0.47%
no_predicates_numeric_generic_poseidon_inliner_max	635 (-3)	-0.47%
fold_numeric_generic_poseidon_inliner_max	635 (-3)	-0.47%
regression_struct_array_conditional_inliner_max	384 (-2)	-0.52%
regression_struct_array_conditional_inliner_min	384 (-2)	-0.52%
regression_struct_array_conditional_inliner_zero	384 (-2)	-0.52%
reference_counts_slices_inliner_0_inliner_zero	956 (-6)	-0.62%
reference_counts_inliner_max_inliner_min	928 (-6)	-0.64%
reference_counts_inliner_0_inliner_min	928 (-6)	-0.64%
reference_counts_inliner_min_inliner_min	928 (-6)	-0.64%
simple_shield_inliner_min	708 (-5)	-0.70%
simple_shield_inliner_max	678 (-5)	-0.73%
fold_call_witness_condition_inliner_min	127 (-1)	-0.78%
simple_shield_inliner_zero	616 (-5)	-0.81%
fold_call_witness_condition_inliner_max	113 (-1)	-0.88%
fold_call_witness_condition_inliner_zero	113 (-1)	-0.88%
conditional_1_inliner_max	557 (-5)	-0.89%
conditional_1_inliner_min	531 (-5)	-0.93%
conditional_1_inliner_zero	531 (-5)	-0.93%
global_array_rc_regression_8259_inliner_min	106 (-1)	-0.93%
reference_counts_slices_inliner_0_inliner_min	1,059 (-10)	-0.94%
pedersen_check_inliner_min	460 (-5)	-1.08%
brillig_pedersen_inliner_min	460 (-5)	-1.08%
fold_complex_outputs_inliner_min	506 (-6)	-1.17%
brillig_pedersen_inliner_zero	418 (-5)	-1.18%
pedersen_check_inliner_zero	418 (-5)	-1.18%
brillig_pedersen_inliner_max	406 (-5)	-1.22%
pedersen_check_inliner_max	406 (-5)	-1.22%
merkle_insert_inliner_min	403 (-5)	-1.23%
merkle_insert_inliner_zero	385 (-5)	-1.28%
fold_complex_outputs_inliner_zero	435 (-6)	-1.36%
array_rc_regression_7842_inliner_min	139 (-2)	-1.42%
mutate_array_copy_inliner_min	138 (-2)	-1.43%
brillig_arrays_inliner_min	136 (-2)	-1.45%
array_if_cond_simple_inliner_max	128 (-2)	-1.54%
array_if_cond_simple_inliner_min	128 (-2)	-1.54%
array_if_cond_simple_inliner_zero	128 (-2)	-1.54%
fold_complex_outputs_inliner_max	349 (-6)	-1.69%
pedersen_hash_inliner_min	287 (-5)	-1.71%
brillig_arrays_inliner_max	114 (-2)	-1.72%
brillig_arrays_inliner_zero	114 (-2)	-1.72%
merkle_insert_inliner_max	545 (-10)	-1.80%
pedersen_hash_inliner_max	263 (-5)	-1.87%
pedersen_hash_inliner_zero	263 (-5)	-1.87%
fold_distinct_return_inliner_min	92 (-2)	-2.13%

vezenovm

Looks good just some nits

compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs

compiler/noirc_evaluator/src/ssa.rs

compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs

asterite · 2025-05-16T14:27:58Z

I looked at the regression in fold_call_witness_condition, which isn't a big deal (just one more instruction). It seems there's a new constant needed now, with value 2, where previously there was no such constant. Maybe because of the offset it appears now. But I guess in most cases this PR will make things better.

vezenovm · 2025-05-16T14:29:48Z

I looked at the regression in fold_call_witness_condition, which isn't a big deal (just one more instruction). It seems there's a new constant needed now, with value 2, where previously there was no such constant. Maybe because of the offset it appears now. But I guess in most cases this PR will make things better.

Yeah I noticed the same thing. For most cases this optimization should make things better and this regression is tiny so it is fine.

compiler/noirc_evaluator/src/ssa/ir/instruction.rs

Automated pull of nightly from the [noir](https://github.com/noir-lang/noir) programming language, a dependency of Aztec. BEGIN_COMMIT_OVERRIDE fix(licm): Account for nested loops being control dependent when analyzing outer loops (noir-lang/noir#8593) chore(refactor): Switch unreachable function removal to use centralized call graph (noir-lang/noir#8578) chore(test): Allow lambdas in fuzzing (noir-lang/noir#8584) chore: use insta for execution_success stdout (noir-lang/noir#8576) chore: use generator instead of zero for ec-add predicate (noir-lang/noir#8552) fix: use predicate expression as binary result (noir-lang/noir#8583) fix(ssa): Do not generate apply functions when no lambda variants exist (noir-lang/noir#8573) chore: put `nargo expand` snapshosts in the same directory (noir-lang/noir#8577) chore: Use FxHashMap for TypeBindings (noir-lang/noir#8574) chore(experimental): use larger stack size for parsing (noir-lang/noir#8347) chore: use insta snapshots for compile_failure stderr (noir-lang/noir#8569) chore(inlining): Mark functions with <= 10 instructions and no control flow as inline always (noir-lang/noir#8533) chore(ssa): Add weighted edges to call graph, move callers and callees methods to call graph (noir-lang/noir#8513) fix(frontend): Override to allow empty array input (noir-lang/noir#8568) fix: avoid logging all unused params in DIE pass (noir-lang/noir#8566) chore: bump external pinned commits (noir-lang/noir#8562) chore(deps): bump base-x from 3.0.9 to 3.0.11 (noir-lang/noir#8555) chore(fuzz): Call function pointers (noir-lang/noir#8531) feat: C++ codegen for msgpack (noir-lang/noir#7716) feat(performance): brillig array set optimization (noir-lang/noir#8550) chore(fuzz): AST fuzzer to use function valued arguments (Part 1) (noir-lang/noir#8514) fix(licm): Check whether the loop is executed when hoisting with a predicate (noir-lang/noir#8546) feat: Implement $crate (noir-lang/noir#8537) fix: add offset to ArrayGet (noir-lang/noir#8536) chore: remove some unused enum variants and functions (noir-lang/noir#8538) fix: disallow `()` in entry points (noir-lang/noir#8529) chore: Remove println in ssa interpreter (noir-lang/noir#8528) fix: don't overflow when casting signed value to u128 (noir-lang/noir#8526) chore(performance): Enable hoisting pure with predicate calls (noir-lang/noir#8522) feat(fuzz): AST fuzzing with SSA interpreter (noir-lang/noir#8436) chore: Add u1 ops to interpreter, convert Value panics to errors (noir-lang/noir#8469) chore: Release Noir(1.0.0-beta.6) (noir-lang/noir#8438) chore(fuzz): AST generator to add `ctx_limit` to all functions (noir-lang/noir#8507) fix(inlining): Use centralized CallGraph structure for inline info computation (noir-lang/noir#8489) fix: remove private builtins from `Field` impl (noir-lang/noir#8496) feat: primitive types are no longer keywords (noir-lang/noir#8470) fix: parenthesized pattern, and correct 1-element tuple printing (noir-lang/noir#8482) fix: fix visibility of methods in `std::meta` (noir-lang/noir#8497) fix: Change `can_be_main` to be recursive (noir-lang/noir#8501) chore: add SSA interpreter test for higher order functions (noir-lang/noir#8486) fix(frontend)!: Ban zero sized arrays and strings as program input (noir-lang/noir#8491) fix!: remove `to_be_radix` and `to_le_radix` from stdlib interface (noir-lang/noir#8495) END_COMMIT_OVERRIDE --------- Co-authored-by: AztecBot <tech@aztecprotocol.com> Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>

asterite added 2 commits May 16, 2025 10:34

Renames

5e9522a

Add offset to ArraySet

c768688

asterite force-pushed the ab/brillig-array-get-and-set-optimization branch from 8f2912e to c768688 Compare May 16, 2025 13:47

Merge branch 'master' into ab/brillig-array-get-and-set-optimization

87bd939

vezenovm reviewed May 16, 2025

View reviewed changes

Apply suggestions from code review

e57cd7a

vezenovm approved these changes May 16, 2025

View reviewed changes

compiler/noirc_evaluator/src/ssa/ir/instruction.rs Outdated Show resolved Hide resolved

Fix typo

418699b

asterite enabled auto-merge May 16, 2025 14:49

asterite added this pull request to the merge queue May 16, 2025

Merged via the queue into master with commit 9ec22c1 May 16, 2025
118 checks passed

asterite deleted the ab/brillig-array-get-and-set-optimization branch May 16, 2025 15:22

noirwhal mentioned this pull request May 16, 2025

chore: Release Noir(1.0.0-beta.7) #8520

Merged

AztecBot mentioned this pull request May 22, 2025

chore(next): release 1.0.0 AztecProtocol/aztec-packages#14467

Closed

AztecBot mentioned this pull request May 23, 2025

chore(next): release 0.88.0 AztecProtocol/aztec-packages#14495

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(performance): brillig array set optimization#8550

feat(performance): brillig array set optimization#8550
asterite merged 5 commits intomasterfrom
ab/brillig-array-get-and-set-optimization

asterite commented May 16, 2025

Uh oh!

github-actions bot commented May 16, 2025 •

edited

Loading

Uh oh!

github-actions bot commented May 16, 2025 •

edited

Loading

Uh oh!

vezenovm left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

asterite commented May 16, 2025

Uh oh!

vezenovm commented May 16, 2025

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

asterite commented May 16, 2025

Description

Problem

Summary

Additional Context

Documentation

PR Checklist

Uh oh!

github-actions bot commented May 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes to number of Brillig opcodes executed

🧾 Summary (10% most significant diffs)

Uh oh!

github-actions bot commented May 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes to Brillig bytecode sizes

🧾 Summary (10% most significant diffs)

Uh oh!

vezenovm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

asterite commented May 16, 2025

Uh oh!

vezenovm commented May 16, 2025

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions bot commented May 16, 2025 •

edited

Loading

github-actions bot commented May 16, 2025 •

edited

Loading