feat(mem2reg): address last known value is independent of its aliases (take three)

[asterite]

Description

Problem

Fixes #9468

Summary

Trying to apply #9613 once more, but fixing an issue.

First issue

I think the refactor in #9613 was fine except: get_known_value was changed to get an address' value regardless of whether the address has aliases.

This get_known_value method is only used in one place: when analyzing Load instructions. There, if get_known_value returned None we did this:

// We don't know the exact value of the address, so we must keep the stores to it.
references.mark_value_used(address, self.inserter.function);

That was actually "because the address might be aliased we better keep the stores to it". That is, because get_known_value implied both that the address didn't have aliases while having a known value, it returning None potentially meant that it was aliased. Now that get_known_value is not related to aliases it's not okay to do that if the value is unknown.

Instead, we must preserve stores to the address if the address is potentially aliased, which is what this PR does (so it's further separating the two concepts).

Second issue

get_known_value retrieving a known value even if an address is aliases, combined with issue #9638, triggered a bug. That was then fixed by #9640

Third issue

The fix above worked because get_known_value returned nothing if an address has an alias. So in this code:

acir(inline) fn create_note f0 {
  b0(v0: &mut Field, v1: &mut Field):
    jmp b1(v0, v1)
  b1(v2: &mut Field, v3: &mut Field):
    store Field 2 at v2
    store Field 3 at v3
    v4 = load v2 -> Field
    return v4
}

with the second issue fix we end up with v2 being aliased to v0, and v3 being aliased to v1.
Then load v2 cannot be simplified because it doesn't have a known value (well, it has, but the code in master returns None if it has aliases).

Now that get_known_value does return a value even if it has aliases, the code above breaks.
The issue is that we set v0 as an alias of v2, but we don't also set the aliases of v0 to be aliases of v2. That is, v2 should be aliased to v0 and v1. That's what the last commit does. It doesn't seem to introduce regressions, and in fact this PR is an optimization.

Additional Context

With these three fixes, CI on Aztec-Packages is green: AztecProtocol/aztec-packages#16576 🎉

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 number of Brillig opcodes executed

Generated at commit: 9e3a8684f63f4a5d7c9d89e5e085691e9039d814, compared to commit: 7c31a63b4688b4bb22e83cdb9639881119343264

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
regression_9455_inliner_max	-3 ✅	-6.67%
regression_9455_inliner_min	-3 ✅	-6.67%
regression_9455_inliner_zero	-3 ✅	-6.67%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
uhashmap_inliner_min	174,325 (-97)	-0.06%
uhashmap_inliner_zero	164,271 (-97)	-0.06%
hashmap_inliner_min	79,931 (-57)	-0.07%
hashmap_inliner_zero	71,462 (-57)	-0.08%
regression_6674_2_inliner_min	863 (-1)	-0.12%
regression_6674_3_inliner_min	1,531 (-2)	-0.13%
regression_9303_inliner_max	159 (-2)	-1.24%
regression_9303_inliner_min	159 (-2)	-1.24%
regression_9303_inliner_zero	159 (-2)	-1.24%
regression_9439_inliner_max	116 (-3)	-2.52%
regression_9439_inliner_min	116 (-3)	-2.52%
regression_9439_inliner_zero	116 (-3)	-2.52%
regression_9455_inliner_max	42 (-3)	-6.67%
regression_9455_inliner_min	42 (-3)	-6.67%
regression_9455_inliner_zero	42 (-3)	-6.67%

[github-actions]

Changes to Brillig bytecode sizes

Generated at commit: 9e3a8684f63f4a5d7c9d89e5e085691e9039d814, compared to commit: 7c31a63b4688b4bb22e83cdb9639881119343264

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
regression_9439_inliner_max	-5 ✅	-3.36%
regression_9439_inliner_min	-5 ✅	-3.36%
regression_9439_inliner_zero	-5 ✅	-3.36%
regression_9455_inliner_max	-5 ✅	-9.62%
regression_9455_inliner_min	-5 ✅	-9.62%
regression_9455_inliner_zero	-5 ✅	-9.62%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
hashmap_inliner_min	9,334 (-1)	-0.01%
hashmap_inliner_zero	7,976 (-1)	-0.01%
uhashmap_inliner_min	7,431 (-3)	-0.04%
uhashmap_inliner_zero	6,945 (-3)	-0.04%
reference_counts_inliner_0_inliner_min	1,041 (-3)	-0.29%
reference_counts_inliner_min_inliner_min	1,041 (-3)	-0.29%
reference_counts_inliner_max_inliner_min	1,041 (-3)	-0.29%
regression_6674_3_inliner_min	672 (-2)	-0.30%
reference_counts_inliner_max_inliner_zero	890 (-3)	-0.34%
reference_counts_inliner_0_inliner_zero	890 (-3)	-0.34%
reference_counts_inliner_min_inliner_zero	890 (-3)	-0.34%
regression_6674_2_inliner_min	243 (-1)	-0.41%
reference_counts_slices_inliner_0_inliner_zero	994 (-7)	-0.70%
reference_counts_slices_inliner_0_inliner_min	1,201 (-11)	-0.91%
regression_9303_inliner_max	105 (-1)	-0.94%
regression_9303_inliner_min	105 (-1)	-0.94%
regression_9303_inliner_zero	105 (-1)	-0.94%
regression_9439_inliner_max	144 (-5)	-3.36%
regression_9439_inliner_min	144 (-5)	-3.36%
regression_9439_inliner_zero	144 (-5)	-3.36%
regression_9455_inliner_max	47 (-5)	-9.62%
regression_9455_inliner_min	47 (-5)	-9.62%
regression_9455_inliner_zero	47 (-5)	-9.62%

[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: e5894e6	Previous: c6835b5	Ratio
test_report_zkpassport_noir-ecdsa_	3 s	1 s	3

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

CC: @TomAFrench

[asterite]

It still fails on Aztec Packages... I'll maybe try again tomorrow.

[asterite]

Reopening because I'll try to fix the issue in this PR

[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: afb9ae0	Previous: 3629a25	Ratio
private-kernel-reset	10.284 s	7.78 s	1.32

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

CC: @TomAFrench

[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: afb9ae0	Previous: 3629a25	Ratio
sha512-100-bytes	0.133 s	0.1 s	1.33

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

CC: @TomAFrench

[asterite]

This is now ready for review. CI is green and this PR in Aztec-Packages is green too (AztecProtocol/aztec-packages#16576).