chore: Rework mem2reg load & store removal

[jfecher]

Description

Problem*

Resolves

Summary*

Draft so that I can compare opcode differences as I work on this PR. Currently I've removed all the extra load/store removals but haven't added them back so initial differences are expected to be bad, similar to #6434.

The goal is to rework the additions to mem2reg (notably all the different ways loads & stores are removed) to be more unified and to handle aliases better.

Additional Context

The end goal of this PR is not necessarily to improve brillig bytecode size (although that'd be a nice bonus) but to rework how we handle load & store removal in mem2reg currently to be less buggy so that #6434 can re-enable this feature without bugs and fix the opcode counts in that PR.

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: 6cbc6ff3d5df777bb7a3b37d5cf3c955460a2aba, compared to commit: 35dedb54a0853ba0fa85038d832a520f9ba01a98

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
brillig_block_parameter_liveness	+12,000 ❌	+245.15%
regression_unsafe_no_predicates	+15 ❌	+50.00%
tuples	+17 ❌	+36.96%
conditional_regression_661	+35 ❌	+26.92%
references	+67 ❌	+22.95%
regression_5045	+16 ❌	+18.82%
derive	+13 ❌	+9.56%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
brillig_block_parameter_liveness	16,895 (+12,000)	+245.15%
regression_unsafe_no_predicates	45 (+15)	+50.00%
tuples	63 (+17)	+36.96%
conditional_regression_661	165 (+35)	+26.92%
references	359 (+67)	+22.95%
regression_5045	101 (+16)	+18.82%
derive	149 (+13)	+9.56%
hashmap	29,430 (+2,367)	+8.75%
6_array	475 (+37)	+8.45%
6	1,498 (+93)	+6.62%
conditional_regression_short_circuit	1,571 (+96)	+6.51%
nested_dyn_array_regression_5782	154 (+9)	+6.21%
uhashmap	17,447 (+921)	+5.57%
higher_order_functions	778 (+39)	+5.28%
brillig_slices	566 (+26)	+4.81%
fold_2_to_17	614 (+27)	+4.60%
brillig_sha256	852 (+37)	+4.54%
bench_2_to_17	352 (+15)	+4.45%
sha256	2,204 (+93)	+4.41%
array_dynamic_nested_blackbox_input	1,050 (+43)	+4.27%
7_function	666 (+27)	+4.23%
regression_4449	915 (+37)	+4.21%
sha256_var_witness_const_regression	1,498 (+56)	+3.88%
ecdsa_secp256k1	1,069 (+37)	+3.59%
poseidon2	356 (+12)	+3.49%
conditional_regression_421	93 (+3)	+3.33%
fold_numeric_generic_poseidon	793 (+24)	+3.12%
no_predicates_numeric_generic_poseidon	793 (+24)	+3.12%
brillig_cow	436 (+12)	+2.83%
conditional_1	1,387 (+37)	+2.74%
if_else_chain	117 (+3)	+2.63%
brillig_calls_conditionals	130 (+3)	+2.36%
array_to_slice	1,043 (+24)	+2.36%
regression_capacity_tracker	238 (+5)	+2.15%
sha256_var_padding_regression	5,488 (+114)	+2.12%
import	149 (+3)	+2.05%
regression	746 (+14)	+1.91%
slices	2,347 (+43)	+1.87%
ram_blowup_regression	1,049 (+19)	+1.84%
strings	1,058 (+19)	+1.83%
sha256_var_size_regression	2,127 (+38)	+1.82%
sha256_regression	7,543 (+132)	+1.78%
pedersen_check	757 (+12)	+1.61%
brillig_pedersen	757 (+12)	+1.61%
array_dynamic_blackbox_input	1,213 (+19)	+1.59%
pedersen_commitment	200 (+3)	+1.52%
simple_shield	882 (+12)	+1.38%
sha2_byte	3,384 (+37)	+1.11%
eddsa	11,281 (+119)	+1.07%
array_dynamic	331 (+3)	+0.91%
brillig_cow_regression	2,437 (+22)	+0.91%
slice_dynamic_index	2,894 (+25)	+0.87%
pedersen_hash	361 (+3)	+0.84%
merkle_insert	784 (+6)	+0.77%
slice_regex	2,589 (+12)	+0.47%
schnorr	1,466 (+6)	+0.41%
regression_5252	4,893 (+15)	+0.31%
poseidon_bn254_hash_width_3	5,698 (+8)	+0.14%
poseidon_bn254_hash	5,698 (+8)	+0.14%

[github-actions]

Changes to circuit sizes

Generated at commit: 6cbc6ff3d5df777bb7a3b37d5cf3c955460a2aba, compared to commit: 35dedb54a0853ba0fa85038d832a520f9ba01a98

🧾 Summary (10% most significant diffs)

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
sha256_var_size_regression	+20 ❌	+0.12%	+19 ❌	+0.03%

---

Full diff report 👇

Program	ACIR opcodes (+/-)	%	Circuit size (+/-)	%
sha256_var_size_regression	16,181 (+20)	+0.12%	70,980 (+19)	+0.03%

[github-actions]

Changes to number of Brillig opcodes executed

Generated at commit: 6cbc6ff3d5df777bb7a3b37d5cf3c955460a2aba, compared to commit: 35dedb54a0853ba0fa85038d832a520f9ba01a98

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
brillig_block_parameter_liveness	+6,000 ❌	+206.19%
tuples	+17 ❌	+53.13%
conditional_regression_661	+19 ❌	+16.10%
references	+67 ❌	+15.91%
6	+858 ❌	+11.63%
conditional_regression_short_circuit	+858 ❌	+11.52%
derive	+13 ❌	+9.77%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
brillig_block_parameter_liveness	8,910 (+6,000)	+206.19%
tuples	49 (+17)	+53.13%
conditional_regression_661	137 (+19)	+16.10%
references	488 (+67)	+15.91%
6	8,233 (+858)	+11.63%
conditional_regression_short_circuit	8,308 (+858)	+11.52%
derive	146 (+13)	+9.77%
regression_4449	240,692 (+20,790)	+9.45%
hashmap	61,304 (+4,113)	+7.19%
nested_dyn_array_regression_5782	135 (+9)	+7.14%
brillig_sha256	4,506 (+297)	+7.06%
array_dynamic_nested_blackbox_input	4,811 (+303)	+6.72%
sha256	11,477 (+619)	+5.70%
conditional_1	6,142 (+297)	+5.08%
higher_order_functions	1,590 (+72)	+4.74%
uhashmap	159,571 (+7,087)	+4.65%
sha256_var_witness_const_regression	7,363 (+322)	+4.57%
ecdsa_secp256k1	7,381 (+297)	+4.19%
6_array	1,922 (+77)	+4.17%
brillig_slices	745 (+26)	+3.62%
conditional_regression_421	90 (+3)	+3.45%
if_else_chain	102 (+3)	+3.03%
brillig_calls_conditionals	114 (+3)	+2.70%
pedersen_commitment	284 (+6)	+2.16%
import	147 (+3)	+2.08%
poseidon2	720 (+12)	+1.69%
brillig_pedersen	1,153 (+18)	+1.59%
pedersen_check	1,153 (+18)	+1.59%
simple_shield	2,618 (+39)	+1.51%
slices	3,925 (+43)	+1.11%
merkle_insert	3,392 (+36)	+1.07%
pedersen_hash	581 (+6)	+1.04%
array_to_slice	2,382 (+24)	+1.02%
strings	1,907 (+19)	+1.01%
7_function	2,839 (+27)	+0.96%
brillig_cow	1,294 (+12)	+0.94%
regression_capacity_tracker	576 (+4)	+0.70%
sha2_byte	47,500 (+297)	+0.63%
array_dynamic	502 (+3)	+0.60%
regression	2,714 (+14)	+0.52%
slice_dynamic_index	4,341 (+21)	+0.49%
fold_numeric_generic_poseidon	5,284 (+24)	+0.46%
no_predicates_numeric_generic_poseidon	5,284 (+24)	+0.46%
sha256_regression	126,775 (+422)	+0.33%
slice_regex	3,861 (+12)	+0.31%
sha256_var_size_regression	18,404 (+50)	+0.27%
array_dynamic_blackbox_input	19,437 (+50)	+0.26%
schnorr	10,202 (+18)	+0.18%
sha256_var_padding_regression	235,091 (+150)	+0.06%
brillig_cow_regression	583,804 (+31)	+0.01%
poseidon_bn254_hash	169,332 (+8)	+0.00%
poseidon_bn254_hash_width_3	169,332 (+8)	+0.00%
eddsa	737,929 (+31)	+0.00%
regression_5252	951,276 (+30)	+0.00%
ram_blowup_regression	877,687 (+25)	+0.00%
fold_2_to_17	131,469,850 (+27)	+0.00%
bench_2_to_17	144,690,675 (+15)	+0.00%

[jfecher]

Commenting to preserve the changes to brillig bytecode sizes comment, this is everything we lose from not doing load & store removal:

Changes to Brillig bytecode sizes

Generated at commit: da19e6da5118bf8acdf435a33ff45aef478d100c, compared to commit: 35dedb54a0853ba0fa85038d832a520f9ba01a98

🧾 Summary (10% most significant diffs)

Program Brillig opcodes (+/-) %
brillig_block_parameter_liveness +16,000 ❌ +326.86%
regression_unsafe_no_predicates +28 ❌ +93.33%
tuples +28 ❌ +60.87%
regression_5045 +38 ❌ +44.71%
conditional_regression_661 +56 ❌ +43.08%
brillig_slices +207 ❌ +38.33%
references +95 ❌ +32.53%
bench_2_to_17 +56 ❌ +16.62%
Full diff report 👇

Program Brillig opcodes (+/-) %
brillig_block_parameter_liveness 20,895 (+16,000) +326.86%
regression_unsafe_no_predicates 58 (+28) +93.33%
tuples 74 (+28) +60.87%
regression_5045 123 (+38) +44.71%
conditional_regression_661 186 (+56) +43.08%
brillig_slices 747 (+207) +38.33%
references 387 (+95) +32.53%
bench_2_to_17 393 (+56) +16.62%
fold_2_to_17 683 (+96) +16.35%
hashmap 31,411 (+4,348) +16.07%
poseidon2 397 (+53) +15.41%
6_array 505 (+67) +15.30%
derive 156 (+20) +14.71%
no_predicates_numeric_generic_poseidon 862 (+93) +12.09%
fold_numeric_generic_poseidon 862 (+93) +12.09%
uhashmap 18,318 (+1,792) +10.84%
6 1,501 (+96) +6.83%
conditional_regression_short_circuit 1,575 (+100) +6.78%
nested_array_in_slice 1,431 (+87) +6.47%
nested_dyn_array_regression_5782 154 (+9) +6.21%
7_function 677 (+38) +5.95%
higher_order_functions 781 (+42) +5.68%
regression 771 (+39) +5.33%
slices 2,417 (+113) +4.90%
brillig_sha256 853 (+38) +4.66%
array_dynamic_nested_blackbox_input 1,053 (+46) +4.57%
sha256 2,207 (+96) +4.55%
regression_4449 916 (+38) +4.33%
regression_capacity_tracker 243 (+10) +4.29%
sha256_var_witness_const_regression 1,500 (+58) +4.02%
eddsa 11,592 (+430) +3.85%
brillig_cow 440 (+16) +3.77%
ecdsa_secp256k1 1,070 (+38) +3.68%
brillig_rc_regression_6123 179 (+6) +3.47%
conditional_regression_421 93 (+3) +3.33%
slice_dynamic_index 2,963 (+94) +3.28%
conditional_1 1,392 (+42) +3.11%
array_to_slice 1,048 (+29) +2.85%
import 150 (+4) +2.74%
fold_complex_outputs 581 (+15) +2.65%
if_else_chain 117 (+3) +2.63%
brillig_calls_conditionals 130 (+3) +2.36%
sha256_var_padding_regression 5,494 (+120) +2.23%
brillig_pedersen 761 (+16) +2.15%
pedersen_check 761 (+16) +2.15%
pedersen_commitment 201 (+4) +2.03%
ram_blowup_regression 1,050 (+20) +1.94%
strings 1,059 (+20) +1.92%
sha256_var_size_regression 2,129 (+40) +1.91%
sha256_regression 7,549 (+138) +1.86%
simple_shield 886 (+16) +1.84%
sha2_byte 3,406 (+59) +1.76%
array_dynamic_blackbox_input 1,215 (+21) +1.76%
regression_5252 4,941 (+63) +1.29%
array_dynamic 332 (+4) +1.22%
brillig_cow_regression 2,444 (+29) +1.20%
pedersen_hash 362 (+4) +1.12%
merkle_insert 786 (+8) +1.03%
regression_struct_array_conditional 550 (+4) +0.73%
array_sort 298 (+2) +0.68%
schnorr 1,468 (+8) +0.55%
slice_regex 2,589 (+12) +0.47%
brillig_keccak 1,774 (+4) +0.23%
keccak256 1,774 (+4) +0.23%
poseidon_bn254_hash_width_3 5,702 (+12) +0.21%
poseidon_bn254_hash 5,702 (+12) +0.21%
poseidonsponge_x5_254 4,504 (+4) +0.09%
u128 2,918 (+1) +0.03%