feat(perf): Optimize array set from get

[vezenovm]

Description

Problem*

Part of general effort to reduce Brillig bytecode sizes

Summary*

We often times (usually when working with nested arrays) run into patterns where we fetch an inner nested array dynamically, array set a different array in the nested array, and the reset the other unchanged nested array at the same index. These array sets can be optimized out.

Here is an example of what happens on master in SSA. Looking at nested_array_dynamic with --force-brillig. Inside b14 you will see the following:

    v114 = array_get v110, index v113
    v115 = add v113, u32 1
    v116 = add v113, u32 2
    v117 = array_get v110, index v116
    v118 = array_set v110, index v113, value v114

After this PR v118 = array_set v110, index v113, value v114 will simplify to v110. If v114 is unused anywhere else in the function (which in this case it is) it will be removed by DIE.

    v114 = add v113, u32 1
    v115 = add v113, u32 2
    v116 = array_get v110, index v115

Additional Context

Similarly to try_optimize_array_get_from_previous_set we should be able to follow past array_sets when optimizing these array_sets.
e.g. if we see a pattern like this:

    v116 = array_get v110, index v115
    v120 = array_set v110, index v114, value [Field 100, Field 101, Field 102]
    v122 = array_set mut v120, index v115, value v116

We know that we performed array_set on the same v110 array, however, it was at a different index. So even though v120 is different than v110 we can simplify v122 = array_set mut v120, index v115, value v116 to v120.

I am working on this in a follow-up, as the simple change in this PR provides decent benefits on its own.

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: 14cfd451c95d56f02af37b713a38c85537a4ae84, compared to commit: 2eb4a2cc782e333be1fcdb3e2b3e6540505a2d18

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
fold_complex_outputs	-59 ✅	-8.31%
regression_struct_array_conditional	-83 ✅	-14.09%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
nested_array_in_slice	1,200 (-38)	-3.07%
array_dynamic_nested_blackbox_input	931 (-30)	-3.12%
nested_array_dynamic	2,302 (-180)	-7.25%
fold_complex_outputs	651 (-59)	-8.31%
regression_struct_array_conditional	506 (-83)	-14.09%

[github-actions]

Changes to circuit sizes

Generated at commit: 14cfd451c95d56f02af37b713a38c85537a4ae84, compared to commit: 2eb4a2cc782e333be1fcdb3e2b3e6540505a2d18

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
nested_array_in_slice	-52 ✅	-4.92%	-147 ✅	-2.63%
nested_array_dynamic	-337 ✅	-8.10%	-834 ✅	-6.06%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
array_dynamic_nested_blackbox_input	496 (-38)	-7.12%	38,904 (0)	0.00%
hashmap	93,383 (-191)	-0.20%	150,661 (-339)	-0.22%
nested_array_in_slice	1,005 (-52)	-4.92%	5,449 (-147)	-2.63%
nested_array_dynamic	3,822 (-337)	-8.10%	12,922 (-834)	-6.06%

[asterite]

Looks great!