[codex] Lower loop-bound array indices as i32

[andrewtdiz]

Summary

• expose trusted local loop bounds as temporary i32 slots while lowering loop bodies
• lower known-array computed get indices through existing i32 expression lowering when all operands are trusted
• keep the numeric array typed-feedback guard and fallback intact, converting the final i32 index back to double only for the guard argument
• add an IR regression test for xs[i * size + 1] inside an i < size loop

Why

After the numeric array guard precheck stack, matrix multiplication still spent work rebuilding hot indices like i * size + k through double arithmetic before the typed-feedback get guard. Loop-bound analysis had already hoisted size as an i32 bound, but that slot was not visible to computed index lowering.

Benchmark Notes

Stacked on codex/perry-numeric-array-guard-precheck / PR #5309.

Baseline #5309:

• direct matrix samples: 400ms, 398ms, 398ms, 385ms, 386ms; median 398ms
• quick matrix: 407ms
• perf direct matrix: 1.443B cycles, 7.034B instructions, 1.568B branches, 241K branch misses

This branch:

• direct matrix 10-sample set: 397ms, 396ms, 390ms, 392ms, 392ms, 393ms, 383ms, 389ms, 386ms, 384ms; median 391ms
• quick matrix: 387ms
• perf direct matrix: 1.457B cycles, 7.018B instructions, 1.568B branches, 249K branch misses
• typed-feedback trace preserved 33,619,968 get passes, 65,536 set passes, 0 get/set failures, and 39 push fallbacks

The wall-time signal is positive and the generated IR removes the double multiply/add index chain from the two hot matmul get sites. The single perf sample is mixed: instructions are slightly lower, while cycles and branch misses are slightly higher.

Validation

• cargo fmt --check
• git diff --check
• cargo test -p perry-codegen --test typed_feedback
• cargo test -p perry-codegen --test typed_shape_descriptors
• PERRY_BIN=target/release/perry python3 tests/test_typed_feedback_runtime_evidence.py
• tests/test_benchmark_output_verifier.sh
• cargo build --release
• target/release/perry compile --no-cache benchmarks/suite/16_matrix_multiply.ts -o /tmp/perry-matrix-i32-index-proto --trace llvm --quiet
• benchmarks/quick.sh
• benchmarks/compare.sh --quick --runs 3 --warn-only --json-out /tmp/perry-i32-index-proto.json

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: fa63c38e-89ef-4925-ae7c-ab4e775996b6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/perry-i32-array-index-lowering

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}