chore: Separate unconstrained functions during monomorphization

[jfecher]

Description

Problem*

Resolves #6841
Resolves #6831

Summary*

Adds a boolean is_unconstrained as an extra key during monomorphization. This should separate constrained from unconstrained functions. This key is infectious so every function called from an unconstrained function will also be unconstrained.

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]

Changes to Brillig bytecode sizes

Generated at commit: d0748c8ba7ff5ec6ee62cdcf940712bf0ebfc6e8, compared to commit: 326173d3b7fa19297baac7017f0c518e495696a4

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
sha256_regression	-13 ✅	-0.19%
higher_order_functions	-598 ✅	-88.33%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
sha256_var_padding_regression	5,123 (-6)	-0.12%
sha256	2,404 (-4)	-0.17%
sha256_regression	6,904 (-13)	-0.19%
higher_order_functions	79 (-598)	-88.33%

[github-actions]

Peak Memory Sample

Program	Peak Memory	%
keccak256	78.48M	0%
workspace	123.58M	0%
regression_4709	422.91M	0%
ram_blowup_regression	1.58G	0%
rollup-base-public	10.47G	0%
rollup-base-private	6.57G	0%
private-kernel-tail	202.04M	0%
private-kernel-reset	716.67M	-1%
private-kernel-inner	292.31M	0%
parity-root	172.14M	0%

[github-actions]

Compilation Report

Program	Compilation Time	%
sha256_regression	1.327s	-3%
regression_4709	0.788s	-2%
ram_blowup_regression	15.145s	-1%
rollup-root	4.202s	7%
rollup-block-merge	4.244s	10%
rollup-base-public	115.827s	4%
rollup-base-private	95.274s	-2%
private-kernel-tail	1.036s	-7%
private-kernel-reset	7.052s	-5%
private-kernel-inner	2.228s	1%
parity-root	0.759s	-7%
noir-contracts	83.448s	-6%

[github-actions]

Execution Report

Program	Execution Time	%
sha256_regression	0.591s	-2%
regression_4709	0.385s	-4%
ram_blowup_regression	3.881s	-1%
rollup-root	2.155s	-1%
rollup-block-merge	2.159s	2%
rollup-base-public	34.950s	-1%
rollup-base-private	31.626s	0%
private-kernel-tail	0.749s	-4%
private-kernel-reset	2.158s	-1%
private-kernel-inner	1.248s	1%
parity-root	0.547s	2%

[github-actions]

Changes to number of Brillig opcodes executed

Generated at commit: d0748c8ba7ff5ec6ee62cdcf940712bf0ebfc6e8, compared to commit: 326173d3b7fa19297baac7017f0c518e495696a4

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
sha256_regression	-141 ✅	-0.12%
higher_order_functions	-1,144 ✅	-91.16%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
sha256_var_padding_regression	222,210 (-6)	-0.00%
sha256	14,992 (-4)	-0.03%
sha256_regression	118,563 (-141)	-0.12%
higher_order_functions	111 (-1,144)	-91.16%

[github-actions]

Execution Memory Report

Program	Peak Memory
keccak256	74.61M
workspace	123.34M
regression_4709	315.93M
ram_blowup_regression	512.47M
rollup-base-public	3.05G
rollup-base-private	2.97G
private-kernel-tail	182.06M
private-kernel-reset	255.64M
private-kernel-inner	215.24M
parity-root	155.86M

[github-actions]

Compilation Memory Report

Program	Peak Memory	%
keccak256	78.50M	0%
workspace	123.63M	0%
regression_4709	422.91M	0%
ram_blowup_regression	1.58G	0%
rollup-base-public	10.47G	0%
rollup-base-private	6.57G	0%
private-kernel-tail	202.02M	0%
private-kernel-reset	716.65M	-1%
private-kernel-inner	292.29M	0%
parity-root	172.13M	0%

[jfecher]

This should be ready for review now. There were some issues with unused functions before that have since been resolved. I've tried a few combinations including updating the defunctionalization pass to not touch acir functions, but I found that it was easiest to just remove unused functions again after inlining to avoid triggering a constant_assert in an unreachable function in the hashmap example.

[TomAFrench]

This causes some regressions on circuit sizes which is not being shown due to a bug in the noir-gates-diff action. The fix is in #6944.

[github-actions]

Changes to circuit sizes

Generated at commit: d0748c8ba7ff5ec6ee62cdcf940712bf0ebfc6e8, compared to commit: 326173d3b7fa19297baac7017f0c518e495696a4

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
array_dynamic_nested_blackbox_input	-2 ✅	-0.79%	-13 ✅	-0.18%
nested_array_dynamic	-156 ✅	-4.44%	-278 ✅	-2.13%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
array_dynamic_nested_blackbox_input	251 (-2)	-0.79%	7,313 (-13)	-0.18%
nested_array_dynamic	3,358 (-156)	-4.44%	12,755 (-278)	-2.13%

[TomAFrench]

This causes some regressions on circuit sizes

Ah nice, looks like I was reading things backwards.