Improve asm const propagation

[xunilrj]

Description

This PR optimises ASM const propagation. That means that it tries to use as few registers as possible, hoping that their initialisation can be removed later with DCE and others...

This is particularly useful on function calls, and indexing:

1 - On function calls we normally have something like

movi $r0 i0
move $$arg0 $r0
jal ...

We now replace this with:

movi $r0 i0
movi $$arg0 i0
jal ...

and let DCE kill the movi $r0 i0 because is no longer neeed;

2 - Constant indexing like something[0] normally generates addi and mul, now these will be coalesced into a simple movi

Gain on testing is huge because we tend to have tons of constants on tests... on real code like O2, we see more humble gains.

Checklist

• I have linked to any relevant issues.
• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation where relevant (API docs, the reference, and the Sway book).
  • If my change requires substantial documentation changes, I have requested support from the DevRel team
• I have added tests that prove my fix is effective or that my feature works.
• I have added (or requested a maintainer to add) the necessary Breaking* or New Feature labels where relevant.
• I have done my best to ensure that my PR adheres to the Fuel Labs Code Review Standards.
• I have requested a review from the relevant team or maintainers.

[codspeed-hq]

Merging this PR will not alter performance

✅ 25 untouched benchmarks

---

_{Comparing xunilrj/asm-const-prop-improv (f81f077) with master (0d8fbc8)}

[xunilrj]

Comparing 'e2e-gas-usages':

	Improvements	Regressions
Count	134	1
Average	3.03%	-0.08%
Median	0.77%	-0.08%
Max	9.96%	-0.08%
Min	0.01%	-0.08%

| should_pass/stdlib/u128_log_test (test.toml) | 11100 | 11109 | -0.08% |

Comparing 'e2e-bytecode-sizes':

	Improvements	Regressions
Count	123	1
Average	3.61%	-0.18%
Median	2.04%	-0.18%
Max	9.09%	-0.18%
Min	0.01%	-0.18%

| should_pass/stdlib/u128_log_test (test.toml) | 4512 | 4520 | -0.18% |

Comparing 'in-language-gas-usages':

	Improvements	Regressions
Count	341	2
Average	0.69%	-0.07%
Median	0.64%	-0.07%
Max	3.50%	-0.08%
Min	0.01%	-0.06%

| u128_inline_tests::parity_u128_log_with_ruint | 1756499 | 1757515 | -0.06% |
| u128_inline_tests::u128_log | 22656 | 22674 | -0.08% |

Comparing 'bytecode-sizes' for project 'o2':

	Improvements	Regressions
Count	11	—
Average	0.09%	—
Median	0.10%	—
Max	0.11%	—
Min	0.01%	—

Comparing 'gas-usages' for project 'o2':

	Improvements	Regressions
Count	65	—
Average	0.03%	—
Median	0.02%	—
Max	0.42%	—
Min	0.01%	—

[xunilrj]

e2e gas usage

	Improvements	Regressions
Count	271	—
Average	7.36%	—
Median	2.32%	—
Max	76.22%	—
Min	0.04%	—

e2e bytecode

	Improvements	Regressions
Count	304	—
Average	11.12%	—
Median	4.44%	—
Max	94.15%	—
Min	0.31%	—

in lang gas usage

	Improvements	Regressions
Count	599	—
Average	6.73%	—
Median	1.39%	—
Max	84.77%	—
Min	0.01%	—

Comparing 'gas-usages' for project 'o2':

	Improvements	Regressions
Count	65	—
Average	0.32%	—
Median	0.29%	—
Max	1.51%	—
Min	0.21%	—

Comparing 'bytecode-sizes' for project 'o2':

	Improvements	Regressions
Count	13	—
Average	1.55%	—
Median	1.67%	—
Max	2.13%	—
Min	0.06%	—

[cursor]

PR Summary

Medium Risk
Changes core ASM optimization logic and emitted bytecode, which can affect correctness and any tests/tools that depend on exact instruction sequences or offsets.

Overview
Improves sway-core’s ASM optimization by significantly expanding constant_propagate to fold constants and apply algebraic/bitwise simplifications (e.g., turning ops into movi/addi/noop, simplifying jnz when the condition is known, and better invalidating dependent known-values). The pass now accepts a logging callback (no-op in normal builds) and adds expect-test coverage for many transform cases.

To support these transforms, the ASM layer adds small ergonomics and representation tweaks (e.g., use_registers_mut now returns a Vec, VirtualOp convenience constructors, From<&str> for VirtualRegister, immediate conversions, and NOOP now renders as noop), plus AbstractInstructionSet display no longer prints blank lines.

Generated output changes ripple into updated integration/snapshot expectations: forc raw log PC/ptr values, multiple e2e stdout.snap files (bytecode/offsets/hashes/sizes), and ABI configurable offsets now reflect the new optimized bytecode.

^{Written by Cursor Bugbot for commit f81f077. This will update automatically on new commits. Configure here.}

[xunilrj]

e2e gas usage

	Improvements	Regressions
Count	272	—
Average	7.48%	—
Median	2.38%	—
Max	76.22%	—
Min	0.04%	—

e2e bytecode

	Improvements	Regressions
Count	305	—
Average	11.21%	—
Median	4.61%	—
Max	94.15%	—
Min	0.31%	—

inlang gas usage

	Improvements	Regressions
Count	606	—
Average	6.95%	—
Median	1.50%	—
Max	84.77%	—
Min	0.01%	—

Comparing 'bytecode-sizes' for project 'o2':

	Improvements	Regressions
Count	13	—
Average	1.56%	—
Median	1.67%	—
Max	2.13%	—
Min	0.07%	—

Comparing 'gas-usages' for project 'o2':

	Improvements	Regressions
Count	65	—
Average	0.32%	—
Median	0.29%	—
Max	1.51%	—
Min	0.21%	—

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}