feat(perf): Remove inc_rc instructions for arrays which are never mutably borrowed

[vezenovm]

Description

Problem*

Part of general effort to reduce Brillig bytecode sizes

Summary*

An array can be borrowed without being mutated. We code gen inc_rc instructions quite pessimistically on every borrow or reassignment of an array. If we know that the array is never mutated, it should be safe to remove these inc_rc instructions.

During DIE we now track all inc_rc instructions in a block and if that array is never used in an array set or a store, we mark its inc_rc instruction for removal.

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: 857c48c1088cf289f395256976e61f8c72d7fb39, compared to commit: d26d20fece6153e4c4f9e06578d0685623f0ec4a

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
regression_struct_array_conditional	-51 ✅	-8.04%
nested_array_in_slice	-123 ✅	-9.04%
simple_2d_array	-15 ✅	-10.64%
generics	-24 ✅	-11.32%
nested_dyn_array_regression_5782	-21 ✅	-14.19%
slice_regex	-432 ✅	-14.41%
fold_complex_outputs	-135 ✅	-15.98%
wildcard_type	-54 ✅	-16.56%
nested_array_dynamic_simple	-36 ✅	-29.03%
brillig_nested_arrays	-183 ✅	-54.79%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
regression_4709	133,782 (-780)	-0.58%
hashmap	26,585 (-167)	-0.62%
modulus	1,732 (-12)	-0.69%
brillig_sha256	755 (-6)	-0.79%
ecdsa_secp256k1	968 (-9)	-0.92%
conditional_1	1,264 (-12)	-0.94%
slice_coercion	314 (-3)	-0.95%
array_dynamic	307 (-3)	-0.97%
bench_2_to_17	294 (-3)	-1.01%
regression_4449	814 (-9)	-1.09%
sha2_byte	3,162 (-36)	-1.13%
fold_2_to_17	525 (-6)	-1.13%
databus_two_calldata	250 (-3)	-1.19%
fold_numeric_generic_poseidon	700 (-9)	-1.27%
no_predicates_numeric_generic_poseidon	700 (-9)	-1.27%
conditional_regression_short_circuit	1,350 (-18)	-1.32%
sha256_regression	6,950 (-96)	-1.36%
brillig_keccak	1,713 (-24)	-1.38%
keccak256	1,713 (-24)	-1.38%
u128	3,107 (-45)	-1.43%
brillig_cow	404 (-6)	-1.46%
slice_dynamic_index	2,587 (-39)	-1.49%
array_dynamic_nested_blackbox_input	980 (-15)	-1.51%
regression_capacity_tracker	192 (-3)	-1.54%
sha256	1,893 (-30)	-1.56%
6	1,776 (-30)	-1.66%
debug_logs	4,992 (-87)	-1.71%
array_sort	340 (-6)	-1.73%
to_be_bytes	169 (-3)	-1.74%
array_dynamic_blackbox_input	1,154 (-21)	-1.79%
brillig_oracle	325 (-6)	-1.81%
sha256_var_witness_const_regression	1,293 (-24)	-1.82%
merkle_insert	724 (-15)	-2.03%
brillig_cow_regression	1,980 (-45)	-2.22%
6_array	394 (-9)	-2.23%
7	127 (-3)	-2.31%
blake3	127 (-3)	-2.31%
brillig_blake3	127 (-3)	-2.31%
uhashmap	23,576 (-573)	-2.37%
to_le_bytes	123 (-3)	-2.38%
hash_to_field	117 (-3)	-2.50%
brillig_hash_to_field	117 (-3)	-2.50%
7_function	567 (-15)	-2.58%
sha256_var_size_regression	1,921 (-54)	-2.73%
regression_5252	5,180 (-165)	-3.09%
poseidonsponge_x5_254	4,343 (-150)	-3.34%
eddsa	11,230 (-390)	-3.36%
regression	690 (-24)	-3.36%
poseidon_bn254_hash	5,523 (-198)	-3.46%
poseidon_bn254_hash_width_3	5,523 (-198)	-3.46%
schnorr	877 (-33)	-3.63%
brillig_array_to_slice	79 (-3)	-3.66%
array_to_slice	910 (-36)	-3.81%
strings	898 (-36)	-3.85%
slice_loop	293 (-12)	-3.93%
to_bytes_consistent	141 (-6)	-4.08%
pedersen_commitment	67 (-3)	-4.29%
higher_order_functions	725 (-33)	-4.35%
import	59 (-3)	-4.84%
simple_shield	294 (-15)	-4.85%
aes128_encrypt	459 (-24)	-4.97%
side_effects_constrain_array	108 (-6)	-5.26%
global_consts	205 (-12)	-5.53%
bigint	2,040 (-123)	-5.69%
slices	2,026 (-123)	-5.72%
to_bytes_integration	48 (-3)	-5.88%
pedersen_check	191 (-12)	-5.91%
brillig_pedersen	191 (-12)	-5.91%
brillig_slices	461 (-30)	-6.11%
to_bits	180 (-12)	-6.25%
embedded_curve_ops	118 (-9)	-7.09%
brillig_unitialised_arrays	38 (-3)	-7.32%
regression_5045	76 (-6)	-7.32%
simple_radix	38 (-3)	-7.32%
nested_array_dynamic	2,461 (-195)	-7.34%
struct_inputs	278 (-24)	-7.95%
regression_struct_array_conditional	583 (-51)	-8.04%
nested_array_in_slice	1,238 (-123)	-9.04%
simple_2d_array	126 (-15)	-10.64%
generics	188 (-24)	-11.32%
nested_dyn_array_regression_5782	127 (-21)	-14.19%
slice_regex	2,565 (-432)	-14.41%
fold_complex_outputs	710 (-135)	-15.98%
wildcard_type	272 (-54)	-16.56%
nested_array_dynamic_simple	88 (-36)	-29.03%
brillig_nested_arrays	151 (-183)	-54.79%

[sirasistant]

Is this safe in the presence of calls to other functions? I guess yes since we increase the rc in the called fn?

[vezenovm]

Is this safe in the presence of calls to other functions? I guess yes since we increase the rc in the called fn?

Yes I believe it should be safe due to the fact that if the called function performs a mutation we will keep the rc on the array parameter inside the called fn.

[TomAFrench]

LGTM, couple of nits