mirror of
https://github.com/Druthulu/BFM-decomp
synced 2026-09-27 05:56:00 -04:00
Drew T
0eadf313ee
- tools/permuter_weights.py (NEW): classify(klass,where) -> regalloc|schedule|cse|None + render_settings_toml emitting the [weight_overrides] table decomp-permuter merges over the gcc defaults (main.py:336 / helpers.py per-key replace). Biases pass-selection toward each class's §31 levers (perm_reorder_decls RC-1/3, perm_reorder_stmts RC-2/S1, perm_temp_for_expr S2, perm_commutative cse); value/type noise -> ~0.1. NO submodule edit. - p16_permute.setup(..., klass=, where=) writes it (+--klass, backlog auto-lookup); grinder.py auto-threads klass/where_stuck. klass=None -> gcc defaults (undirected superset). - grinder fix (R14): replace the blind tried.clear() idle churn with input-changed gating (draft_sig=(best_draft mtime, closeness) — re-open a fn only when the worker improved it). - VALIDATED: regalloc profile drove func_8014E048 masked 36->29 (match_one -drz 35->28) where the undirected search stalled; re-logged (28, improved draft promoted). Flagship S11/RC-6 count-exact seeds improve but do not gate (intrinsic, §31 "two probes dont grind") -> T5b fuel. - distill: cookbook §3b + SETUP inventory (R16/R30). No build-input changed -> 136/136 untouched.