BFM-decomp/docs/lever-progress.tsv at e76cb1c89e094e0bcecf3050d91623bbcadaaf2b

mirror of https://github.com/Druthulu/BFM-decomp synced 2026-10-02 16:00:27 -04:00

Files

T

Drew T fc7d8c4019 phase-36: T7 burst — a22 banked (126 bodies), a14 refused as an invented zero term, and the struct question answered on the record

- a22: func_8017B238 closed by giving the if-arm's pointer and the else arm's first table address one function-scope
  local. Three decisions turn on that edit, all dump-proven: make_regs_eqv's head rule (cse.c:840-857), set_preference
  stripping one RTX level so an arithmetic set inherits its operand's register (global.c:1535), and combine_regs with
  birthing_insn_p's reg_n_sets == 1 (local-alloc.c:1765-1788, sched.c:2469) — which is why the merge must be with the
  else arm's non-call-crossing temp. 26,714 -> 26,462 sites.
- a14 reaches score 0 on func_80139BE0 but only by an INVENTED identically-zero term whose sole purpose is to keep a
  value live. NOT BANKED: an invented no-op expression is a compiler-forcing construct in C clothing, and worse than the
  marked launder it replaces because the launder is counted and this would be silent. The phase's own rule is ban the
  silence, not the lever. Parked for the structs/types phase with its reading; the tree comment above it ('no pure-C
  spelling survives that fold') is refuted.
- a6 corrects cookbook 455: cse1 (cse.c:7439-7502) rewrites the producer's destination to the copy's whenever the
  producer is the immediately preceding insn; combine only finishes the job once that adjacency is broken. It enumerated
  the three lever-free blockers can_combine_p admits and showed the body can pay for none, then scanned all 4,284 built
  objects for the shape — 101 hits, every lever-free precedent paying with a genuine second use, a narrow local with two
  consumers, or a join label.
- a24's residual was one absent load: cse forwards a just-stored halfword so no lh is emitted, and seven branch
  displacement mismatches were downstream of it. Its method gap is the instrument to fix next — three agents have now
  asked for local-alloc's quantity table in the pack, and alloc_table.py prints an empty or one-row table because it
  keys on dump lines that are often absent.
- Drew's struct question answered on the record: structs are not in the binary (types are erased; no metadata in a
  retail build), what is there is base + offset + width + stride, so a struct is an inference across every function
  touching a base — which makes per-function struct invention the wrong unit and is the failure P35 already recorded.
  Recommendation: keep pins as the main lane, build a zero-token struct evidence census beside it, park stuck pins with
  their evidence, and fix call signatures first because they are the bigger and cheaper blocker (471 narrow call sites
  in 323 pinned bodies). Measured: 372,224 raw cast dereferences against 92,624 struct member accesses.

2026-09-10 11:02:14 -06:00

4.2 KiB

Raw Blame History

1	date	milestone	head	sites_AB	pins	asm	bodies_AB	distinct_AB	marked	gte_levers	per_tu_asm_macros	class_C	class_D	class_E	class_F	class_G
2	2026-09-09	T6 r1 (rung R's first class)	1761a3125	33957	20512	13445	12578	1758	33957	462	314	1722	50	7428	445	76
3	2026-09-09	T6 d1 (rung D's four exemplars)	8fac96651	33953	20508	13445	12574	1758	33953	462	314	1722	50	7428	445	76
4	2026-09-09	T6 p1 (four classes propagated)	a2da99b13	33427	19982	13445	12048	1755	33427	462	314	1722	50	7428	445	76
5	2026-09-09	S101 rung G g1 (the guided search's first run: 1 of 16 exemplars, 132 bodies)	623e55340	33295	19982	13313	11916	1754	33295	462	314	1590	50	7428	445	76
6	2026-09-09	S101 rung G g2 (wider beam + R10/R12/R13/R8-cse: 1 of 16, 132 bodies)	584031c71	33031	19718	13313	11784	1753	33031	462	314	1590	50	7428	445	76
7	2026-09-09	S101 rung G g3 (the next 64 classes: 13 of 64, 1,516 bodies)	5ae0b5ceb	31025	18442	12583	10268	1740	31025	462	314	1590	50	7428	445	76
8	2026-09-09	S101 rung G g4s (the 13 explained names, 38 classes: 3 of 38, 133 bodies)	068781ae0	30892	18309	12583	10135	1737	30892	462	314	1590	50	7428	445	76
9	2026-09-09	S101 rung G g5 (the head re-drawn wide: 5 of 70, 62 bodies)	1ec67f667	30806	18261	12545	10073	1732	30806	462	314	1590	50	7428	445	76
10	2026-09-09	S101 rung G g6 + g6b (the tail: 47 + 63 of 400 small classes, 93 + 186 + 47 bodies)	2f3ce92a1	30358	17954	12404	9747	1624	30358	462	314	1590	50	7428	445	76
11	2026-09-10	S102 T7 a1 — func_80156044 closed by an agent (merge-point expression sunk into the if/else arms, both $2/$3 pins gone); 130 bodies banked (1 + 129 propagated)	5a8bd1021	30098	17694	12404	9617	1623	30098	462	314	1590	50	7428	445	76
12	2026-09-10	S102 T7 s1 — the R15 sweep of the head: 6 exemplars closed by the sink alone (267 bodies), no drafting tokens	861dd0651	29697	17427	12270	9350	1617	29697	462	314	1590	50	7428	445	76
13	2026-09-10	S102 T7 a2 — func_80168828: the constant-holder pin deleted and one adjacent swap closes it (125 bodies, 1 of 2 pins off; the $4 pin is forced by the header's (void) declaration)	c00ffe5d4	29572	17302	12270	9350	1617	29572	462	314	1590	50	7428	445	76
14	2026-09-10	S102 T7 s3 (partial, the run crashed in propagate) — the constant-holder draw: 10 bodies closed	975850ff8	29533	17266	12267	9326	1594	29533	462	314	1590	50	7428	445	76
15	2026-09-10	S102 T7 s3b — the constant-holder draw re-run after the propagate fix: 1 of 80, 6 bodies; 59 of 80 hit the compile budget	5603a114c	29527	17260	12267	9320	1593	29527	462	314	1590	50	7428	445	76
16	2026-09-10	S102 T7 s5 — the first broad TAIL sweep with R15/R16/R17/R18: 285 of 800 bodies closed lever-free, no drafting tokens	449acd8cb	29204	17029	12175	9035	1308	29204	462	314	1590	50	7428	445	76
17	2026-09-10	S102 T7 s6 — the last never-attempted exemplars: 25 of 337; the fresh pool is now empty	67e91ad4d	29148	16996	12152	9010	1283	29148	462	314	1590	50	7428	445	76
18	2026-09-10	S102 T7 s7 (stopped early to restore agent cadence) — 5 of 207 judged at the wider width	724c68859	29140	16989	12151	9005	1280	29140	462	314	1590	50	7428	445	76
19	2026-09-10	S102 T7 a7 — func_801287B8: a truncated local extern was faking a pin; the missing call argument restored through a function-pointer cast (127 bodies)	bb0e95c7c	29013	16862	12151	8878	1279	29013	462	314	1590	50	7428	445	76
20	2026-09-10	S102 T7 a4 — func_8016C49C: one do-while on the function's LAST statement doubles a loop-note reference weight and flips allocno_compare (126 bodies)	9f5b22176	28887	16736	12151	8752	1278	28887	462	314	1590	50	7428	445	76
21	2026-09-10	S102 T7 burst — a8/a12/a18/a19 banked (132+118+119+134 bodies): the call-arity class twice, a MEM_IN_STRUCT_P alias flip, and a sequence-point split	fb6c857f7	27984	15965	12019	8249	1275	27984	462	314	1590	50	7428	445	76
22	2026-09-10	S102 T7 burst — a5/a9/a11/a16/a25 banked; the MISSING CALL ARGUMENT class confirmed six times independently (m2c dropped arguments at unprototyped and indirect call sites)	c6b380fcd	26714	15324	11390	7612	1271	26714	462	314	1590	50	7428	445	76
23	2026-09-10	S102 T7 burst — a22 banked (126); a14 REFUSED as an invented zero term; a6/a10/a15 read without closing	374c8a0e9	26462	15198	11264	7486	1270	26462	462	314	1590	50	7428	445	76

4.2 KiB Raw Blame History

4.2 KiB

Raw Blame History