perf(codegen): inline births are black and seeded; omit redundant root-store shading (#11929)

[proggeramlug]

Closes #11929. Re-lands #12037 (reverted by #12038) with the inline-birth hole closed structurally.

What

Codegen no longer emits incremental-mark shading on stores into GENERATED GC roots: compiler-managed locals, temporary roots, module globals, persistent scalar/array root aliases. Root bindings/registrations, heap/runtime barriers (closure captures stay heap slots with js_write_barrier_slot) and all GC gates are unchanged.

Removing root-store shading is sound because of two things TOGETHER:

1. FinalRootRemark rescans every mutable root, including the perf(codegen): use stable homes for root-heavy functions (#11926) #11960 stable homes (cycle.rs → roots.rs → bound_ptr()).
2. Every inline birth is black and seeded.

Neither alone suffices.

Inline births (the hole that caused the revert)

Both inline codegen allocators, new and small array literals/bundles, now go through ONE shared mechanism that mirrors the runtime allocator:

1. read the LIVE per-thread birth flags at each birth;
2. write the header with the flags in one store;
3. initialize every slot (undefined / +0.0 F64 birth lanes / elements);
4. seed non-leaf objects through the existing mark-seed queue (js_gc_note_black_birth, which rechecks MARKED and pointer-free itself);
5. publish.

There is no safepoint inside that window. The array literal's re-entering numeric-layout normalizer runs after it, with the array rooted. During a budgeted sweep the flags are 0, so there is no MARKED bit and no seed, exactly as in the runtime allocators. Those births fall past the sweep snapshot's recorded fill (#11842).

There is no exact gc_flags == GC_FLAG_ARENA precheck in the source or the emitted IR, so black births don't fall to the slow path.

The audit covers every allocation path:

• runtime objects, arrays, closures, strings, typed arrays and containers already use the runtime allocator (flags + type-dependent seeding);
• static templates, stack storage and foreign backing are not collectible births.

Results (qb6, 5 interleaved runs, medians; main → head)

	instructions	CPU s	RSS KiB	binary
tsc	10,198,988,430 → 10,131,745,462 (−0.66%)	2.29 → 2.05	221,660 → 215,996	−5.71 MB
Zod	785,455,172 → 783,296,310 (−0.27%)	0.13 → 0.12	56,772 → 56,620	−213 KB
hello	1,223,220 → 1,222,913	—	15,832 → 15,664	unchanged; relocations unchanged

Root-shading calls: generic witness 26 → 0, q200 2,608 → 0.

Verification

• Runtime suite 5,105 passed; codegen 2,544 passed; gc-root-dominance and fmt pass.
• 60 GC proof/sabotage executions: the white-birth sabotages for new and array literals, and seed-before-init under a forced step, all fail as intended.
• 18 fixture runs match node across all GC knob sets, including instrumented PERRY_GC_BUDGETED_OLD_RECLAIM=1 + PERRY_GC_PROTECT_OLD_SWEEP=1 (quarantine banner confirmed).
• The proof tests were adapted to main's removed RememberedSetRebuild phase; no gate was weakened.
• GC subsets and lint show only failures main already has.

Reviewed by the GC owner before opening: seed ordering for new across the constructor, the array-literal ordering, sweep-phase flags = 0, and seed lifetime across DisableBarrier → sweep snapshot.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 08878487-4f09-4e7e-826f-172fb8a3d865

📥 Commits

Reviewing files that changed from the base of the PR and between 44de841 and 2793669.

⛔ Files ignored due to path filters (4)

• crates/perry-codegen/src/gc_effects/linux-x86_64.tsv is excluded by !**/*.tsv
• crates/perry-codegen/src/gc_effects/macos-aarch64.tsv is excluded by !**/*.tsv
• crates/perry-codegen/src/gc_effects/windows-x86_64.tsv is excluded by !**/*.tsv
• crates/perry-codegen/src/wasm32/runtime_abi.tsv is excluded by !**/*.tsv

📒 Files selected for processing (34)

• changelog.d/PENDING-11929r-inline-birth.md
• crates/perry-codegen/src/expr/array_literal.rs
• crates/perry-codegen/src/expr/array_push_guard_tests.rs
• crates/perry-codegen/src/expr/generic_overhead_tests.rs
• crates/perry-codegen/src/expr/hit_path_access_tests.rs
• crates/perry-codegen/src/expr/hit_path_tests.rs
• crates/perry-codegen/src/expr/inline_birth.rs
• crates/perry-codegen/src/expr/mod.rs
• crates/perry-codegen/src/expr/packed_loop_shadow_barrier_tests.rs
• crates/perry-codegen/src/expr/scalar_slot_root.rs
• crates/perry-codegen/src/expr/shadow_inline.rs
• crates/perry-codegen/src/expr/shadow_slot.rs
• crates/perry-codegen/src/expr/write_barrier.rs
• crates/perry-codegen/src/lower_call/alloc_hot_tests.rs
• crates/perry-codegen/src/lower_call/ctor_prologue_stores.rs
• crates/perry-codegen/src/lower_call/inline_birth_invariant.rs
• crates/perry-codegen/src/lower_call/new_alloc.rs
• crates/perry-codegen/src/rooting/temp_root.rs
• crates/perry-codegen/src/runtime_decls/strings.rs
• crates/perry-codegen/tests/scalar_replaced_slot_roots.rs
• crates/perry-codegen/tests/shadow_slot_hygiene.rs
• crates/perry-codegen/tests/typed_feedback.rs
• crates/perry-runtime/src/arena/block.rs
• crates/perry-runtime/src/arena/inline.rs
• crates/perry-runtime/src/gc/barrier/mod.rs
• crates/perry-runtime/src/gc/tests/cycle_state.rs
• crates/perry-runtime/src/gc/tests/cycle_state/birth_color_controls.rs
• crates/perry-runtime/src/gc/tests/cycle_state/inline_birth.rs
• crates/perry-runtime/src/gc/tests/cycle_state/root_remark.rs
• crates/perry-runtime/src/gc/trace.rs
• test-files/test_gap_gc_11929_inline_births.ts
• test-files/test_gap_gc_11929_stable_home_marking.ts
• test-files/test_gap_gc_generated_root_final_remark.ts
• test-parity/gc_repsel_corpus.txt

 __________________________________________________
< Are you not entertained? By the bugs I've found? >
 --------------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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