feat: use constraint information to perform constant folding #4060

TomAFrench · 2024-01-16T23:41:43Z

Description

Problem*

Resolves #4070

Summary*

This PR uses Instruction::Constrain to perform simplifications in SSA where one value is swapped for another simpler value due to the fact that the circuit will be unsatisfiable should the two values not be equivalent (because of the Instruction::Constrain.

A previous implementation of this resulted in the bug #2645. This was because constraints which were only active in certain areas of the code were applied across all of the SSA. This implementation avoids this by using a separate cache for the constrained values for each value of side_effects_enabled_var. This means that passing a Instruction::EnableSideEffects instruction will "forget" any constraints which aren't currently active.

One trouble from this PR is that we now require multiple passes in order to fully simplify the circuit. I've simply added another set of the final 3 passes but we could in future perform extra passes until the SSA stabilises.

Additional Context

Documentation*

Check one:

No documentation needed.
Documentation included in this PR.
[Exceptional Case] 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 · 2024-01-16T23:45:02Z

Changes to circuit sizes

Generated at commit: 2e85a543c1d304fffea55ca6fa98253b39238b11, compared to commit: fd150521a480c04ff64f84e3c1a2faf1e8394516

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
slices	-63 ✅	-9.95%	-264 ✅	-5.83%
slice_dynamic_index	-337 ✅	-13.95%	-1,265 ✅	-11.63%

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
operator_overloading	592 (-4)	-0.67%	8,214 (0)	0.00%
strings	31 (-38)	-55.07%	14,373 (0)	0.00%
6_array	524 (-11)	-2.06%	4,198 (-19)	-0.45%
slices	570 (-63)	-9.95%	4,264 (-264)	-5.83%
slice_dynamic_index	2,079 (-337)	-13.95%	9,611 (-1,265)	-11.63%

…ossible. (#4065) # Description ## Problem\* Resolves  ## Summary\* This PR adds a very simple pass which moves any `Instruction::Constrain`s to immediately after when all of their inputs are available. This allows the ACIR gen pass to use this information earlier as well as simplifying using these constraints in order to inform constant folding in future. ## Additional Context Pulled out from #4060 ## Documentation\* Check one: - [x] No documentation needed. - [ ] Documentation included in this PR. - [ ] **[Exceptional Case]** Documentation to be submitted in a separate PR. # PR Checklist\* - [x] I have tested the changes locally. - [x] I have formatted the changes with [Prettier](https://prettier.io/) and/or `cargo fmt` on default settings.

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

kevaundray · 2024-01-22T13:08:33Z

For the protocol-circuits what is the diff in the compile times?

TomAFrench · 2024-01-22T13:16:58Z

// nightly
nargo compile --force  18.11s user 4.98s system 181% cpu 12.700 total

// this PR
nargo compile --force  27.53s user 3.65s system 303% cpu 10.269 total

github-actions · 2024-02-01T14:05:48Z

🚀 Deployed on https://65bba53aef5d6f0e51c2352d--noir-docs.netlify.app

TomAFrench · 2024-02-07T12:05:20Z

This is ready for review now. We no longer have a large number of rounds of constant folding, instead we perform a single round which doesn't use the constraint information and then we have another which applies constraints.

jfecher

LGTM. Just one nitpick

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

#4303) # Description ## Problem\* Followup to #4302. ## Summary\* This PR unwraps `constrain cast(v0, typ) == cast(v1, typ)` into `constrain v0 == v1`. This will potentially improve #4060 effectiveness as we're moving this constraint information up the DFG. ## Additional Context ## Documentation\* Check one: - [x] No documentation needed. - [ ] Documentation included in this PR. - [ ] **[Exceptional Case]** Documentation to be submitted in a separate PR. # PR Checklist\* - [x] I have tested the changes locally. - [x] I have formatted the changes with [Prettier](https://prettier.io/) and/or `cargo fmt` on default settings. --------- Co-authored-by: jfecher <jake@aztecprotocol.com>

🤖 I have created a release *beep* *boop* --- <details><summary>0.24.0</summary> ## [0.24.0](v0.23.0...v0.24.0) (2024-02-12) ### ⚠ BREAKING CHANGES * rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) * Add expression width into acir (AztecProtocol/aztec-packages#4014) * init storage macro (AztecProtocol/aztec-packages#4200) * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) * Sync commits from `aztec-packages` ([#4144](#4144)) ### Features * Add bit size to const opcode (AztecProtocol/aztec-packages#4385) ([158c8ce](158c8ce)) * Add brillig array index check ([#4127](#4127)) ([c29f85f](c29f85f)) * Add definitions for From and Into traits to Noir prelude ([#4169](#4169)) ([4421ce4](4421ce4)) * Add expression width into acir (AztecProtocol/aztec-packages#4014) ([158c8ce](158c8ce)) * Add instrumentation for tracking variables in debugging ([#4122](#4122)) ([c58d691](c58d691)) * Add option to print monomorphized program ([#4119](#4119)) ([80f7e29](80f7e29)) * Add support for overriding expression width ([#4117](#4117)) ([c8026d5](c8026d5)) * Add warnings for usage of restricted bit sizes ([#4234](#4234)) ([0ffc38b](0ffc38b)) * Allow bitshifts to be represented in SSA for brillig ([#4301](#4301)) ([d86ff1a](d86ff1a)) * Allow brillig to read arrays directly from memory (AztecProtocol/aztec-packages#4460) ([158c8ce](158c8ce)) * Allow globals to refer to any expression ([#4293](#4293)) ([479330e](479330e)) * Allow nested arrays and vectors in Brillig foreign calls (AztecProtocol/aztec-packages#4478) ([158c8ce](158c8ce)) * Allow variables and stack trace inspection in the debugger ([#4184](#4184)) ([bf263fc](bf263fc)) * **avm:** Back in avm context with macro - refactor context (AztecProtocol/aztec-packages#4438) ([158c8ce](158c8ce)) * **aztec-nr:** Initial work for aztec public vm macro (AztecProtocol/aztec-packages#4400) ([158c8ce](158c8ce)) * Deallocate stack items at the instruction level ([#4339](#4339)) ([8f024a8](8f024a8)) * Disable constraint bubbling pass ([#4131](#4131)) ([9ba2de6](9ba2de6)) * Disable unused variable checks on low-level and oracle functions ([#4179](#4179)) ([8f70e57](8f70e57)) * Evaluation of dynamic assert messages ([#4101](#4101)) ([c284e01](c284e01)) * Improve Error Handling for Cargo in Bootstrap Script ([#4211](#4211)) ([3a90849](3a90849)) * Init storage macro (AztecProtocol/aztec-packages#4200) ([158c8ce](158c8ce)) * **lsp:** Goto type reference for Struct ([#4091](#4091)) ([d56cac2](d56cac2)) * Move bounded_vec into the noir stdlib ([#4197](#4197)) ([c50621f](c50621f)) * Multiply first to allow more ACIR gen optimizations ([#4201](#4201)) ([882639d](882639d)) * Option expect method ([#4219](#4219)) ([8e042f2](8e042f2)) * Perform constraints on uncasted values if they are the same type ([#4303](#4303)) ([816fa85](816fa85)) * Remove predicate from `sort` intrinsic function ([#4228](#4228)) ([d646243](d646243)) * Remove replacement of boolean range opcodes with `AssertZero` opcodes ([#4107](#4107)) ([dac0e87](dac0e87)) * Replace bitwise ANDs used for truncation with `Instruction::Truncate` ([#4327](#4327)) ([eb67ff6](eb67ff6)) * Replace modulo operations with truncations where possible ([#4329](#4329)) ([70f2435](70f2435)) * Separate compilation and expression narrowing in `nargo` interface ([#4100](#4100)) ([62a4e37](62a4e37)) * Simplify all unsigned constant NOT instructions ([#4230](#4230)) ([fab4a6e](fab4a6e)) * Sync commits from `aztec-packages` ([#4144](#4144)) ([0205d3b](0205d3b)) * Use constraint information to perform constant folding ([#4060](#4060)) ([9a4bf16](9a4bf16)) ### Bug Fixes * Accurate tracking of slice capacities across blocks ([#4240](#4240)) ([7420dbb](7420dbb)) * Allow function calls in global definitions ([#4320](#4320)) ([0dc205c](0dc205c)) * Allow performing bitwise NOT on unsigned integers ([#4229](#4229)) ([b3ddf10](b3ddf10)) * Apply generic arguments from trait constraints before instantiating identifiers ([#4121](#4121)) ([eb6fc0f](eb6fc0f)) * Apply range constraints to return values from unconstrained functions ([#4217](#4217)) ([3af2a89](3af2a89)) * Apply trait constraints from method calls ([#4152](#4152)) ([68c5486](68c5486)) * Better errors for missing `fn` keyword ([#4154](#4154)) ([057c208](057c208)) * Check for tests in all packages before failing due to an unsatisfied test filter ([#4114](#4114)) ([1107373](1107373)) * Clean error when attemping to return a slice from Brillig to ACIR ([#4280](#4280)) ([bcad4ec](bcad4ec)) * Correct result when assigning shared arrays in unconstrained code ([#4210](#4210)) ([bdd8a96](bdd8a96)) * **docs:** Codegen docs before cutting a new version ([#4183](#4183)) ([2914310](2914310)) * Ensure that destination register is allocated when moving between registers in brillig gen ([#4316](#4316)) ([ca0a56e](ca0a56e)) * Ensure that unconstrained entrypoint functions don't generate constraints ([#4292](#4292)) ([fae4ead](fae4ead)) * From field with constant values ([#4226](#4226)) ([593916b](593916b)) * **lsp:** Crash when file not in workspace ([#4146](#4146)) ([cf7130f](cf7130f)) * **lsp:** Replace panics with errors ([#4209](#4209)) ([26e9618](26e9618)) * Maintain correct type when simplifying `x ^ x` ([#4082](#4082)) ([9d83c2b](9d83c2b)) * Message formatting for assert statement ([#4323](#4323)) ([3972ead](3972ead)) * Prevent debugger crashing on circuits with no opcodes ([#4283](#4283)) ([2e32845](2e32845)) * Prevent declarations of blackbox functions outside of the stdlib ([#4177](#4177)) ([9fb6b09](9fb6b09)) * Remove panic from `init_log_level` in `acvm_js` ([#4195](#4195)) ([2e26530](2e26530)) * Respect order in bubble up for redundant asserts ([#4109](#4109)) ([189aa48](189aa48)) * Revert "correct result when assigning shared arrays" and added regression test ([#4333](#4333)) ([05e78b3](05e78b3)) * Save the data bus to the current function before generating others ([#4047](#4047)) ([0a5bd4f](0a5bd4f)) * Simplify constant assert messages into `ConstrainError::Static` ([#4287](#4287)) ([fd15052](fd15052)) * Ssa typing for array & slice indexes ([#4278](#4278)) ([4074bab](4074bab)) * Ssa typing for assign_lvalue_index ([#4289](#4289)) ([37f149c](37f149c)) * SSA typing for right shifts ([#4302](#4302)) ([41ee1aa](41ee1aa)) * Ssa typing of make_offset ([#4277](#4277)) ([e4378ee](e4378ee)) * Track graphs of item dependencies to find dependency cycles ([#4266](#4266)) ([61eabf1](61eabf1)) * Type check ACIR mutable reference passed to brillig ([#4281](#4281)) ([7e139de](7e139de)) * Update array method type signatures in the docs ([#4178](#4178)) ([7c0a955](7c0a955)) * Zero out input to `to_radix` calls if inactive ([#4116](#4116)) ([3f5bad3](3f5bad3)) ### Miscellaneous Chores * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) ([158c8ce](158c8ce)) * Rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) ([158c8ce](158c8ce)) </details> <details><summary>0.40.0</summary> ## [0.40.0](v0.39.0...v0.40.0) (2024-02-12) ### ⚠ BREAKING CHANGES * rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) * Add expression width into acir (AztecProtocol/aztec-packages#4014) * init storage macro (AztecProtocol/aztec-packages#4200) * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) * Sync commits from `aztec-packages` ([#4144](#4144)) * Breaking changes from aztec-packages ([#3955](#3955)) * Rename Arithmetic opcode to AssertZero ([#3840](#3840)) * Remove unused methods on ACIR opcodes ([#3841](#3841)) * Remove partial backend feature ([#3805](#3805)) ### Features * Add bit size to const opcode (AztecProtocol/aztec-packages#4385) ([158c8ce](158c8ce)) * Add expression width into acir (AztecProtocol/aztec-packages#4014) ([158c8ce](158c8ce)) * Add instrumentation for tracking variables in debugging ([#4122](#4122)) ([c58d691](c58d691)) * Add support for overriding expression width ([#4117](#4117)) ([c8026d5](c8026d5)) * Allow brillig to read arrays directly from memory (AztecProtocol/aztec-packages#4460) ([158c8ce](158c8ce)) * Allow nested arrays and vectors in Brillig foreign calls (AztecProtocol/aztec-packages#4478) ([158c8ce](158c8ce)) * Allow variables and stack trace inspection in the debugger ([#4184](#4184)) ([bf263fc](bf263fc)) * **avm:** Back in avm context with macro - refactor context (AztecProtocol/aztec-packages#4438) ([158c8ce](158c8ce)) * **aztec-nr:** Initial work for aztec public vm macro (AztecProtocol/aztec-packages#4400) ([158c8ce](158c8ce)) * Aztec-packages ([#3754](#3754)) ([c043265](c043265)) * Breaking changes from aztec-packages ([#3955](#3955)) ([5be049e](5be049e)) * Evaluation of dynamic assert messages ([#4101](#4101)) ([c284e01](c284e01)) * Init storage macro (AztecProtocol/aztec-packages#4200) ([158c8ce](158c8ce)) * Remove range constraints from witnesses which are constrained to be constants ([#3928](#3928)) ([afe9c7a](afe9c7a)) * Remove replacement of boolean range opcodes with `AssertZero` opcodes ([#4107](#4107)) ([dac0e87](dac0e87)) * Speed up transformation of debug messages ([#3815](#3815)) ([2a8af1e](2a8af1e)) * Sync `aztec-packages` ([#4011](#4011)) ([fee2452](fee2452)) * Sync commits from `aztec-packages` ([#4068](#4068)) ([7a8f3a3](7a8f3a3)) * Sync commits from `aztec-packages` ([#4144](#4144)) ([0205d3b](0205d3b)) ### Bug Fixes * Deserialize odd length hex literals ([#3747](#3747)) ([4000fb2](4000fb2)) * Remove panic from `init_log_level` in `acvm_js` ([#4195](#4195)) ([2e26530](2e26530)) * Return error rather instead of panicking on invalid circuit ([#3976](#3976)) ([67201bf](67201bf)) ### Miscellaneous Chores * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) ([158c8ce](158c8ce)) * Remove partial backend feature ([#3805](#3805)) ([0383100](0383100)) * Remove unused methods on ACIR opcodes ([#3841](#3841)) ([9e5d0e8](9e5d0e8)) * Rename Arithmetic opcode to AssertZero ([#3840](#3840)) ([836f171](836f171)) * Rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) ([158c8ce](158c8ce)) </details> --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please).

TomAFrench force-pushed the tf/bring-back-constraint-cache branch 2 times, most recently from ec8cfd2 to c1f6191 Compare January 17, 2024 01:04

This was referenced Jan 17, 2024

feat: bubble up Instruction::Constrains to be applied as early as possible. #4065

Merged

Apply knowledge of constraints to SSA constant folding #4070

Closed

feat: rewrite constraints on binary instructions in terms of their inputs #4064

Draft

TomAFrench force-pushed the tf/bring-back-constraint-cache branch 2 times, most recently from 80baa8f to ef2273c Compare January 17, 2024 17:39

TomAFrench force-pushed the tf/bring-back-constraint-cache branch from 10b7f20 to 399a3b2 Compare January 17, 2024 19:06

TomAFrench marked this pull request as ready for review January 17, 2024 19:52

TomAFrench requested review from jfecher and vezenovm and removed request for jfecher January 17, 2024 23:16

TomAFrench commented Jan 17, 2024

View reviewed changes

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

TomAFrench mentioned this pull request Jan 22, 2024

chore: Update noir AztecProtocol/aztec-packages#4168

Merged

TomAFrench force-pushed the tf/bring-back-constraint-cache branch 6 times, most recently from 8313eed to dd450c3 Compare February 1, 2024 13:55

github-actions bot added the documentation Improvements or additions to documentation label Feb 1, 2024

TomAFrench changed the base branch from master to tf/side-effects-pass February 1, 2024 13:55

TomAFrench marked this pull request as draft February 1, 2024 13:55

TomAFrench removed the documentation Improvements or additions to documentation label Feb 1, 2024

TomAFrench mentioned this pull request Feb 2, 2024

Instruction::has_side_effects and Instrinsic::has_side_effects are not accurate. #4237

Closed

feat: replace constrained values with simpler equivalents

fac2a3e

TomAFrench force-pushed the tf/bring-back-constraint-cache branch from dd450c3 to fac2a3e Compare February 7, 2024 10:34

TomAFrench changed the base branch from tf/side-effects-pass to master February 7, 2024 10:34

TomAFrench removed request for vezenovm and jfecher February 7, 2024 10:34

TomAFrench and others added 2 commits February 7, 2024 10:42

Merge branch 'master' into tf/bring-back-constraint-cache

ff649ad

feat: add a constant folding pass which is unaware of constraints

c81f9d3

TomAFrench marked this pull request as ready for review February 7, 2024 12:04

TomAFrench requested review from vezenovm and jfecher February 7, 2024 12:04

jfecher approved these changes Feb 7, 2024

View reviewed changes

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

TomAFrench added 2 commits February 7, 2024 17:06

chore: documentation

a779693

chore: cargo fmt

9a2bf9b

TomAFrench enabled auto-merge February 7, 2024 17:11

TomAFrench added this pull request to the merge queue Feb 7, 2024

Merged via the queue into master with commit 9a4bf16 Feb 7, 2024
33 checks passed

TomAFrench deleted the tf/bring-back-constraint-cache branch February 7, 2024 17:56

kevaundray mentioned this pull request Feb 7, 2024

chore: Release Noir(0.24.0) #4113

Merged

TomAFrench mentioned this pull request Feb 8, 2024

feat: perform constraints on uncasted values if they are the same type #4303

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: use constraint information to perform constant folding #4060

feat: use constraint information to perform constant folding #4060

TomAFrench commented Jan 16, 2024 •

edited

Loading

github-actions bot commented Jan 16, 2024 •

edited

Loading

kevaundray commented Jan 22, 2024

TomAFrench commented Jan 22, 2024

github-actions bot commented Feb 1, 2024

TomAFrench commented Feb 7, 2024

jfecher left a comment

feat: use constraint information to perform constant folding #4060

feat: use constraint information to perform constant folding #4060

Conversation

TomAFrench commented Jan 16, 2024 • edited Loading

Description

Problem*

Summary*

Additional Context

Documentation*

PR Checklist*

github-actions bot commented Jan 16, 2024 • edited Loading

Changes to circuit sizes

🧾 Summary (10% most significant diffs)

kevaundray commented Jan 22, 2024

TomAFrench commented Jan 22, 2024

github-actions bot commented Feb 1, 2024

TomAFrench commented Feb 7, 2024

jfecher left a comment

Choose a reason for hiding this comment

TomAFrench commented Jan 16, 2024 •

edited

Loading

github-actions bot commented Jan 16, 2024 •

edited

Loading