chore: Try to optimize compilation memory

[aakoshh]

Description

Problem*

The rollup-checkpoint-root and rollup-checkpoint-root-single-block contracts need almost 10GB memory during compilation, see here.

Summary*

Improves compilation memory usage in the heaviest aztec-protocol circuits:

contract	new	old	ratio
private-kernel-reset	498.22 MB	562.99 MB	0.88
rollup-checkpoint-root-single-block	6820 MB	9710 MB	0.70
rollup-checkpoint-root	6820 MB	9720 MB	0.70
rollup-tx-base-private	1050 MB	1360 MB	0.77
rollup-tx-base-public	2940 MB	6960 MB	0.42
sha512_100_bytes	185.39 MB	251.6 MB	0.74

Reinsertion

Adds an allow_reinsert flag to FunctionInserter::push_instruction_value; when true it passes the instruction ID to to DataFlowGraph::insert_instruction_and_results_if_simplified to try to avoid storing a new copy of an instruction when we can reinsert it as-is.

The flag gets different values in different passes:

• flatten_cfg: Allows reinsertion as we are mostly focussing on sprinkling around the enable_side_effect instructions and weaving them into instructions, but whatever isn't affected can be reused. Not using OTOH duplicates all instructions.
• mem2reg: We can only reuse instructions we haven't seen before, because we are tracking which instruction to remove by ID, and if we reinsert multiple times then we might just want to remove the last one, yet we end up removing all occurrences.
• unrolling: Not using because we are importing between blocks
• loop_invariant: Using when we reinsert into the same block, not when hoisting into the pre-header.

Inline order

Changed function inlining to go in bottom-up order and overwrite functions as soon as they have been inlined, hoping to reduce the outstanding memory. Unfortunately it had no effect on the peak memory.

Remove unused instructions

Added a Ssa::remove_unused_instructions pass after some of the DIE and mem2reg passes, hoping to free some memory from the DenseMap by making the content smaller. It had negligible effect, maybe 300MB. Potentially the Instruction::Noop is as large as the largest Instruction anyway, except for MakeArray which has a Vec in it, and maybe some of the others that contain String; the results were SmallVec, so they don't consume more or less if there are just 1-2 results.

Instruction::map_values

Based on profiling, the unconditional creation of a new im::Vec for MakeArray::elements consumed a lot of memory. Since im supports sharing via Arc, I changed this to try to just .clone() and not modify unless an item actually changed.

Additional Context

Profiled command:

./target/release/nargo compile \
      --program-dir /Users/aakoshh/Work/aztec/aztec-packages/noir-projects/noir-protocol-circuits \
      --package rollup_checkpoint_root \
      --silence-warnings --skip-underconstrained-check --skip-brillig-constraints-check --force

Code to compile and sign nargo for Instruments:

#!/usr/bin/env bash

set -e

DIR=$(dirname $0)

cat > $DIR/debug.plist <<EOF
<?xml version="1.0" encoding="UTF-8"?><!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "https://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0"><dict><key>com.apple.security.get-task-allow</key><true/></dict></plist>
EOF

cd $DIR/noir && cargo build -p nargo_cli --release && cd -

codesign -s - -v -f --entitlements $DIR/debug.plist $DIR/noir/target/release/nargo

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]

ACVM Benchmarks

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
purely_sequential_opcodes	267650 ns/iter (± 1320)	252791 ns/iter (± 1014)	1.06
perfectly_parallel_opcodes	232700 ns/iter (± 4288)	224580 ns/iter (± 683)	1.04
perfectly_parallel_batch_inversion_opcodes	2784739 ns/iter (± 7785)	2258668 ns/iter (± 4124)	1.23

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

[github-actions]

Test Suite Duration

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
test_report_AztecProtocol_aztec-packages_noir-projects_aztec-nr	119 s	119 s	1
test_report_AztecProtocol_aztec-packages_noir-projects_noir-contracts	135 s	135 s	1
test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_blob	261 s	268 s	0.97
test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_private-kernel-lib	207 s	212 s	0.98
test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_reset-kernel-lib	35 s	35 s	1
test_report_AztecProtocol_aztec-packages_noir-projects_noir-protocol-circuits_crates_types	155 s	132 s	1.17
test_report_noir-lang_noir-bignum_	162 s	162 s	1
test_report_noir-lang_noir_bigcurve_	342 s	421 s	0.81
test_report_noir-lang_sha256_	15 s	16 s	0.94
test_report_noir-lang_sha512_	14 s	13 s	1.08
test_report_zkpassport_noir-ecdsa_	3 s	1 s	3
test_report_zkpassport_noir_rsa_	1 s	1 s	1

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

[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: 909e22b	Previous: da35eca	Ratio
rollup-block-root-single-tx	0.003 s	0.002 s	1.50

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

CC: @TomAFrench

[github-actions]

Execution Time

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	0.012 s	0.012 s	1
private-kernel-reset	0.157 s	0.157 s	1
private-kernel-tail	0.01 s	0.01 s	1
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	0.004 s	0.004 s	1
rollup-checkpoint-merge	0.003 s	0.003 s	1
rollup-checkpoint-root-single-block	12.3 s	12.8 s	0.96
rollup-checkpoint-root	12.5 s	12.4 s	1.01
rollup-root	0.004 s	0.004 s	1
rollup-tx-base-private	0.299 s	0.299 s	1
rollup-tx-base-public	0.236 s	0.238 s	0.99
rollup-tx-merge	0.002 s	0.002 s	1
semaphore-depth-10	0.009 s	0.008 s	1.13
sha512-100-bytes	0.069 s	0.071 s	0.97

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

[github-actions]

Opcode count

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	15876 opcodes	15876 opcodes	1
private-kernel-reset	78291 opcodes	78291 opcodes	1
private-kernel-tail	12195 opcodes	12195 opcodes	1
rollup-block-root-first-empty-tx	1365 opcodes	1365 opcodes	1
rollup-block-root-single-tx	1049 opcodes	1049 opcodes	1
rollup-block-root	2410 opcodes	2410 opcodes	1
rollup-checkpoint-merge	2330 opcodes	2330 opcodes	1
rollup-checkpoint-root-single-block	962022 opcodes	962022 opcodes	1
rollup-checkpoint-root	963382 opcodes	963382 opcodes	1
rollup-root	2830 opcodes	2830 opcodes	1
rollup-tx-base-private	263910 opcodes	263910 opcodes	1
rollup-tx-base-public	245349 opcodes	245349 opcodes	1
rollup-tx-merge	1486 opcodes	1486 opcodes	1
semaphore-depth-10	5700 opcodes	5700 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: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	736.2 KB	736.2 KB	1
private-kernel-reset	2111.5 KB	2111.5 KB	1
private-kernel-tail	549.6 KB	549.6 KB	1
rollup-block-root-first-empty-tx	179.9 KB	179.9 KB	1
rollup-block-root-single-tx	178.2 KB	178.2 KB	1
rollup-block-root	258.2 KB	258.2 KB	1
rollup-checkpoint-merge	373.8 KB	373.8 KB	1
rollup-checkpoint-root-single-block	27645.4 KB	27645.4 KB	1
rollup-checkpoint-root	27690.6 KB	27690.6 KB	1
rollup-root	414.3 KB	414.3 KB	1
rollup-tx-base-private	4920.2 KB	4920.2 KB	1
rollup-tx-base-public	4571.5 KB	4571.5 KB	1
rollup-tx-merge	186.2 KB	186.2 KB	1
semaphore-depth-10	570.7 KB	570.7 KB	1
sha512-100-bytes	506.3 KB	506.3 KB	1

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

[github-actions]

Compilation Time

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	1.68 s	1.722 s	0.98
private-kernel-reset	7.846 s	7.59 s	1.03
private-kernel-tail	1.338 s	1.538 s	0.87
rollup-block-root-first-empty-tx	1.422 s	1.382 s	1.03
rollup-block-root-single-tx	1.47 s	1.38 s	1.07
rollup-block-root	1.49 s	1.46 s	1.02
rollup-checkpoint-merge	1.526 s	1.478 s	1.03
rollup-checkpoint-root-single-block	223 s	210 s	1.06
rollup-checkpoint-root	219 s	213 s	1.03
rollup-root	1.514 s	1.528 s	0.99
rollup-tx-base-private	17.54 s	17.92 s	0.98
rollup-tx-base-public	131.4 s	80.42 s	1.63
rollup-tx-merge	1.348 s	1.384 s	0.97
semaphore-depth-10	0.799 s	0.786 s	1.02
sha512-100-bytes	1.587 s	2.05 s	0.77

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

[github-actions]

Compilation Memory

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	266.47 MB	273.53 MB	0.97
private-kernel-reset	498.21 MB	562.99 MB	0.88
private-kernel-tail	248.46 MB	256.79 MB	0.97
rollup-block-root-first-empty-tx	341.01 MB	343.28 MB	0.99
rollup-block-root-single-tx	338.46 MB	340.51 MB	0.99
rollup-block-root	341.44 MB	344.33 MB	0.99
rollup-checkpoint-merge	340.78 MB	344.37 MB	0.99
rollup-checkpoint-root-single-block	6820 MB	9710 MB	0.70
rollup-checkpoint-root	6820 MB	9720 MB	0.70
rollup-root	345.05 MB	346.97 MB	0.99
rollup-tx-base-private	1050 MB	1360 MB	0.77
rollup-tx-base-public	2890 MB	6960 MB	0.42
rollup-tx-merge	338.65 MB	339.37 MB	1.00
semaphore_depth_10	92.18 MB	97.55 MB	0.94
sha512_100_bytes	185.49 MB	251.64 MB	0.74

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

[github-actions]

Execution Memory

Details

Benchmark suite	Current: 0c5c7da	Previous: 845a8e4	Ratio
private-kernel-inner	255.57 MB	255.57 MB	1
private-kernel-reset	293.11 MB	293.11 MB	1
private-kernel-tail	244 MB	244 MB	1
rollup-block-root	338.92 MB	338.92 MB	1
rollup-checkpoint-merge	337.79 MB	337.79 MB	1
rollup-checkpoint-root-single-block	1020 MB	1020 MB	1
rollup-checkpoint-root	1020 MB	1020 MB	1
rollup-root	338.99 MB	338.99 MB	1
rollup-tx-base-private	452.47 MB	452.47 MB	1
rollup-tx-base-public	467.36 MB	467.36 MB	1
rollup-tx-merge	337.18 MB	337.18 MB	1
semaphore_depth_10	73.69 MB	73.69 MB	1
sha512_100_bytes	71.96 MB	71.96 MB	1

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

[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: 0c5c7da	Previous: 845a8e4	Ratio
rollup-tx-base-public	131.4 s	80.42 s	1.63

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

CC: @TomAFrench

[jfecher]

Looks great! Showing some good improvements in memory usage in CI.

[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: 0c5c7da	Previous: 845a8e4	Ratio
test_report_zkpassport_noir-ecdsa_	3 s	1 s	3

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: 0c5c7da	Previous: 845a8e4	Ratio
perfectly_parallel_batch_inversion_opcodes	2784739 ns/iter (± 7785)	2258668 ns/iter (± 4124)	1.23

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

CC: @TomAFrench

[jfecher]

Nice 👍