mirror/BFM-decomp
1
0
Fork 0
mirror of https://github.com/Druthulu/BFM-decomp synced 2026-09-27 05:56:00 -04:00
Code Issues Packages Projects Releases Wiki Activity

feat(decomp): parallel gate — 2 fns across 2 binaries (10 workers)

Browse Source 
ov_SC03_105    func_801803A0
  ov_SC06_033    func_80190E64
This commit is contained in:
Drew T
2026-09-02 10:00:36 -06:00
parent a2b0a6b98f
commit 38faa55a56
4 changed files with 302 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
config/overlays.mk
+1 -1
View File
@@ -2427,7 +2427,7 @@ build/src/ov_SC03_105/ov_SC03_105_jr_80159C84.o: JTBL_PADS := 0,4 # §8e pads (
build/src/ov_SC03_105/ov_SC03_105_jr_8015AE2C.o: JTBL_PADS := 0,4 # §8e pads (jtbl_carve.py) tables=+0x0,+0x20
build/src/ov_SC03_105/ov_SC03_105_jr_8016AB6C.o: JTBL_PADS := 0,0 # §8e pads (jtbl_carve.py) tables=+0x0,+0x20
build/src/ov_SC03_105/ov_SC03_105_jr_80178D40.o: JTBL_PADS := 0,0 # §8e pads (jtbl_carve.py) tables=+0x0,+0x178
build/src/ov_SC03_105/ov_SC03_105_jr_8017C8D0.o: JTBL_PADS := 0,0 # §8e pads (jtbl_carve.py) tables=+0x0,+0x20
build/src/ov_SC03_105/ov_SC03_105_jr_8017C8D0.o: JTBL_PADS := 0,0,0 # §8e pads (jtbl_carve.py) tables=+0x0,+0x20,+0x34
build/src/ov_SC03_105/ov_SC03_105_o0c.o: JTBL_PADS := 0,4,4,4 # §8e pads (jtbl_carve.py) tables=+0x0,+0x38,+0xa8,+0x118
ov_SC03_105_CHECK_SHA := config/check.ov_SC03_105.sha
ov_SC03_105_SYMBOLS := config/symbols.ov_SC03_105.txt
config/splat.ov_SC03_105.yaml
+1 -1
View File
@@ -167,7 +167,7 @@ segments:
- [0x8fe5c, .rodata, ov_SC03_105_jr_8017AE2C] # Phase-26 §8 jtbl-rodata carve (jtbl_carve.py)
- [0x8fe70, data, tail18]
- [0x8ffa8, .rodata, ov_SC03_105_jr_8017C8D0] # Phase-26 §8 jtbl-rodata carve (jtbl_carve.py)
- [0x8ffdc, data, tail19]
- [0x8fff4, data, tail19]
- [0x90138, .rodata, ov_SC03_105_jr_80186B24] # Phase-26 §8 jtbl-rodata carve (jtbl_carve.py)
- [0x90160, .rodata, ov_SC03_105_jr_80186DDC] # Phase-26 §8 jtbl-rodata carve (jtbl_carve.py)
- [0x90188, data, tail20]
src/ov_SC03_105/ov_SC03_105_jr_8017C8D0.c
+138 -1
View File
@@ -5046,7 +5046,144 @@ void func_801802E0(s32 a0) {
}
INCLUDE_ASM("asm/ov_SC03_105/nonmatchings/ov_SC03_105_jr_8017C8D0", func_801803A0);
#include "common.h"
extern s32 D_80126B58; /* TU:7037 house spelling; base taken as (u8 *)& like TU:7059 */
extern s32 D_801BCBBC; /* TU:5024 */
extern u16 D_801BA59A; /* sh/lhu width; no prior TU decl */
extern u16 D_801BA59C; /* sh/lhu width; no prior TU decl */
extern s32 D_8018E610; /* TU:4975 */
extern s32 func_80047948(s32 a0); /* TU:2206 */
extern s32 func_8004787C(s32 a0); /* TU:2207 */
extern s32 rand(void); /* TU:958 */
extern s32 func_8012BEE8(s32 a0); /* TU:3924 */
extern s32 func_8012CBF4(s32 a0); /* TU:3927 */
extern void func_8012AD80(s32 a0); /* TU:4390 */
extern s32 func_8018233C(s32 arg0); /* defined later in this TU (:5121) */
extern void func_80183768(void *a0, s32 a1, s32 a2); /* defined later in this TU (:5669) */
extern void func_80183D38(void *a0, s16 a1, s32 a2); /* defined later in this TU (:5930) */
extern s32 func_80183A64(s32 *a0, s32 a1, s32 a2); /* defined later in this TU (:5775) */
extern s32 func_8018388C(void *a0, s32 a1); /* defined later in this TU (:5714) */
extern void func_80183DA0(s32 a0); /* defined later in this TU (:5953) */
extern void func_801824CC(s32, s32, s32, s32, s32); /* TU:5009 spelling; K&R s16 def at :5191 */
void func_801803A0(s32 a0)
{
s32 s0 = a0;
register s32 s3 __asm__("$19");
u8 *b58;
s32 v;
u32 v1;
/* LOAD-BEARING: ONE pointer temp per switch arm. A single temp assigned in both arms has
* reg_n_sets==2, which kills sched1's birthing_insn_p launch boost (§350): its `lw 0x20`
* floats to the block top, overlaps the `li $v1,0x40` range, and the allocno lands in $a0
* (measured 9 rows). Set once each, both loads sit after the sw and colour $v1. */
s32 pA;
s32 pB;
s32 i;
s32 r;
/* LOAD-BEARING: dead 16-byte aggregate gives the 0x40 frame (0x30 without it); gcc-2.7.2
* allocates an aggregate local at declaration even when unreferenced (§226/§333). */
s32 pad[4];
v = func_80047948((*(s32 *)(s0 + 0xEC) << 5) & 0xFE0);
*(u16 *)(s0 + 0x50) = ((u32)(v * 3)) >> 9;
v = func_8004787C((*(s32 *)(s0 + 0xEC) << 5) & 0x7E0);
*(u16 *)(s0 + 0x52) = ((u32)(v * 3)) >> 9;
/* LOAD-BEARING: `u8 *` base + `*(s16 *)(b58 + 6)` (the TU:7059 idiom), NOT an s16 array
* `tbl[3]`. The ARRAY_REF is a /s MEM and true_dependence's struct-vs-fixed-scalar escape
* lets the `lh 0x6($s1)` hoist above the `sh D_801BA59C` store; the cast INDIRECT_REF is
* not /s, so the load stays pinned after both global stores as in the target (§351; 4 rows).
* Keep it set ONCE — a §194-K re-tie moves the `la $s1` to the prologue top (§350). */
b58 = (u8 *)&D_80126B58;
switch (*(u16 *)(s0 + 0x34)) {
case 0:
v1 = *(u32 *)*(s32 *)(s0 + 0xCC);
s3 = 0x1800;
if (v1 == -2) goto lA;
if ((u32)v1 >= 0xFFFFFFFFu) goto lB;
if (v1 == -3) goto lC;
goto lD;
lA:
*(s32 *)(s0 + 0x1C) = 0x40;
pA = *(s32 *)(s0 + 0x20);
*(u16 *)(s0 + 0x34) = *(u16 *)(s0 + 0x34) + 1;
*(u16 *)(pA + 0x1A) = 0x2000;
*(u16 *)(pA + 0x18) = 0x2000;
goto lE;
lC:
*(u16 *)(s0 + 0x34) = 4;
*(s32 *)(s0 + 0x1C) = 0x3C;
for (i = 0; i < 3; i++) {
r = rand();
func_80183D38((void *)s0, i + 1, 0xFFFC0000 - ((r % 5) << 16));
}
goto lE;
lB:
*(s32 *)(s0 + 0xCC) = *(s32 *)((s32)&D_8018E610 + D_801BCBBC * 4);
return;
lD:
*(u16 *)(s0 + 2) = v1;
*(s32 *)(s0 + 0xCC) = *(s32 *)(s0 + 0xCC) + 4;
return;
lE:
*(s32 *)(s0 + 0xCC) = *(s32 *)(s0 + 0xCC) + 4;
break;
case 1:
s3 = 0x2000;
if (func_8012BEE8(s0) != 0) {
*(u16 *)(s0 + 0x34) = *(u16 *)(s0 + 0x34) + 1;
D_801BA59A = *(u16 *)(s0 + 0xA);
D_801BA59C = *(u16 *)(s0 + 0xE);
*(s16 *)(s0 + 0xE8) = (*(s16 *)(s0 + 6) - *(s16 *)(b58 + 6)) >> 1;
*(s32 *)(s0 + 0x1C) = 1;
*(u16 *)(s0 + 0xEA) = *(u16 *)(s0 + 0xA) + 0x102;
}
break;
case 2:
func_80183768((void *)s0, *(s16 *)(s0 + 0xE8), *(s16 *)(s0 + 0xEA));
func_8012CBF4(s0);
v = *(s32 *)(s0 + 0x1C) + 1;
*(s32 *)(s0 + 0x1C) = v;
s3 = 0x2000;
if (v >= 0x41) {
pB = *(s32 *)(s0 + 0x20);
*(s32 *)(s0 + 0x1C) = 0x1E;
*(u16 *)(s0 + 0x34) = 0;
*(u16 *)(pB + 0x1A) = 0x1800;
*(u16 *)(pB + 0x18) = 0x1800;
*(u16 *)(s0 + 0xA) = D_801BA59A;
*(u16 *)(s0 + 0xE) = D_801BA59C;
}
break;
case 4:
s3 = 0x1800;
if (func_8012BEE8(s0) != 0) {
*(u16 *)(s0 + 0x34) = 0;
*(s32 *)(s0 + 0x1C) = 0x1E;
}
break;
case 5:
func_8012AD80(s0);
s3 = 0x1800;
if (func_8012BEE8(s0) != 0) {
*(s32 *)(s0 + 0x1C) = 0x1E;
*(u16 *)(s0 + 0x34) = 0;
*(s32 *)(s0 + 0xC) = 0xFE380000;
}
break;
}
func_80183A64((s32 *)s0, 0x1800000, 0xFE800000);
func_8018388C((void *)s0, 0x30);
func_8018233C(s0);
func_80183DA0(s0);
func_801824CC(s0, (s16)s3, 0, -0x20, 0x800000);
*(s32 *)(s0 + 0xEC) += 1;
}
INCLUDE_ASM("asm/ov_SC03_105/nonmatchings/ov_SC03_105_jr_8017C8D0", func_801806F8);
src/ov_SC06_033/ov_SC06_033_jr_8018D98C.c
+162 -1
View File
@@ -4993,7 +4993,168 @@ void func_80190CE0(s32 a0) {
}
INCLUDE_ASM("asm/ov_SC06_033/nonmatchings/ov_SC06_033_jr_8018D98C", func_80190E64);
/* func_80190E64 — ov_SC06_033 — 3-state (0/1/2 on +0x34) boss-intro/spawn driver.
match_one MATCH 302/302, all 53 relocations audited against the .s (law 1c).
Levers, in the order they mattered (each byte-proven on this function):
1. `first = D_801CB512;` — the 16-entry u16 rotate (house style of the already-banked
func_801916BC in this TU) must load `first` through the SYMBOL, not `*q1`: the target
has `lui/lhu %lo(D_801CB512)`, and the load must be written BEFORE `k = 0` (§3-T2).
2. `*(u16 *)(e + 6) = *(s16 *)(e + 6) - 0x40 + (rand() & 0x7F);` — the s16 LOAD is the
dial. Read as u16 the fold+narrowing produces a u16 `0xFFC0` constant that local-alloc
parks in $s0 (a 1-block pseudo crossing one call), which then bars `e` from $s0 in BOTH
cases and cascades through the whole callee-saved file (75 residual rows). Read as s16
the narrowed HImode arithmetic is signed, the negated constant sign-extends to -64 and
becomes `addiu $v1,$v1,-0x40` on the FIELD side, then `addu` with the rand term.
(Ablated 19 spellings; `(u16)(f - K)` cast and `+= (r) - K` both keep the 0xFFC0 reg.)
3. ONE `p` variable for `*(s32 *)(a0 + 0xCC)` in case 0 AND case 1 — a global pseudo that
does not cross calls gets the lowest used-so-far reg not conflicting with the local
RMW temps: $a0 in both blocks (a per-block temp gets $v1).
4. `i = rand() % 100` reuses the case-0 counter — that pseudo must cross calls to land in
$s1 (an 8-ref/12-insn temp takes $a1); it also lifts `i` above the D_801CB52C giv in
allocno order so the giv falls through to $s3 (pass 1: s0..s2 taken), and `tbl` then
shares $s3 with it in pass 0.
Free from the shape: `while (*tp <= i) tp += 3;` (duplicate_loop_exit_test + reorg's
delay-slot steal with the `addiu -6` undo), `-(cond) & 0x1E` (store_flag normalised to -1,
L9163), u16 dividend -> `multu 0xAAAAAAAB` (§201-D), state in $s0 reused for the `== 2`
compare via record_jump_equiv. */
extern u8 D_80078E78[];
extern u16 D_801CB512;
extern u16 D_801CB514;
extern u16 D_801CB4F4;
extern s32 D_801CB52C[];
extern u8 D_801CB540[];
extern u8 D_801CB4CC[];
extern s16 D_801CB398[];
extern u8 D_801CD99C[];
extern void func_8002D4C8(s32 a0, s32 a1);
extern void func_800599B8();
extern s32 func_8012C658(s32 arg0, s32 arg1, s32 arg2);
extern void func_8001C214(s32 a0, s32 a1);
extern s32 rand(void);
extern void func_8001D0E8(s32 a0, s32 a1, s32 a2);
extern int func_80143C74(short *, int);
extern void func_80128EA8();
extern void func_80016714(void *a0, s32 a1);
extern void func_8002A04C(s32 a0);
void func_80190E64(s32 a0)
{
u16 *tbl = (u16 *)D_80078E78;
s32 e;
s32 i;
s32 child;
s32 off;
s16 *tp;
s32 t;
s32 p;
s16 cnt;
u16 *q1, *q2;
u16 first, v;
s32 k;
switch (*(u16 *)(a0 + 0x34)) {
case 0:
func_8002D4C8(0xB33, 0);
q1 = &D_801CB512;
q2 = &D_801CB512 - 1;
first = D_801CB512;
k = 0;
do {
v = *q2;
q2--;
k++;
*q1 = v;
q1--;
} while (k < 0xF);
*q1 = first;
func_800599B8(&D_801CB514, &D_801CB4F4);
p = *(s32 *)(a0 + 0xCC);
*(u16 *)(p + 0x2C) |= 0x10;
for (i = 0; i < 5; i++) {
e = func_8012C658(0x33C, 1, a0);
if (e != 0) {
func_8001C214(*(s32 *)(e + 0x20), D_801CB52C[i]);
*(u16 *)(e + 0x12) = (rand() & 0xF) - 8;
*(u16 *)(e + 0x1A) = (rand() & 0xF) - 8;
*(u16 *)(e + 0x16) = -((rand() & 7) + 7);
*(u16 *)(e + 0xFC) = (rand() & 0x3F) + 4;
*(u16 *)(e + 0xFE) = (rand() & 0x3F) + 4;
*(u16 *)(e + 0x100) = (rand() & 0x3F) + 4;
*(u16 *)(e + 0xA) -= 0x30;
}
}
func_8001C214(*(s32 *)(a0 + 0x20), (s32)D_801CD99C);
*(s32 *)(*(s32 *)(a0 + 0x20) + 0x20) = (s32)D_801CB4CC;
func_8001D0E8(*(s32 *)(a0 + 0x20), 0x280, 0x1E0);
*(u16 *)(a0 + 0x34) += 1;
break;
case 1:
q1 = &D_801CB512;
q2 = &D_801CB512 - 1;
first = D_801CB512;
k = 0;
do {
v = *q2;
q2--;
k++;
*q1 = v;
q1--;
} while (k < 0xF);
*q1 = first;
func_800599B8(&D_801CB514, &D_801CB4F4);
e = ((s32 (*)(s32, s32))func_80143C74)(a0, 0);
if (e != 0) {
child = *(s32 *)(e + 0xCC);
*(u16 *)(child + 0x18) = rand() % 0x2000 + 0x3000;
*(u16 *)(child + 0x1A) = rand() % 0x2000 + 0x3000;
func_80128EA8(child, e + 0xD0, (s32)D_801CB540);
*(u16 *)(e + 0x6) = *(s16 *)(e + 0x6) - 0x40 + (rand() & 0x7F);
*(u16 *)(e + 0xE) = *(s16 *)(e + 0xE) - 0x40 + (rand() & 0x7F);
*(u16 *)(e + 0xA) = *(u16 *)(e + 0xA) - (rand() & 0x7F);
}
p = *(s32 *)(a0 + 0xCC);
*(u16 *)(p + 0x18) -= 0x88;
*(u16 *)(p + 0x1A) -= 0x88;
*(u16 *)(p + 0x1C) -= 0x88;
if (*(s16 *)(p + 0x18) < 0x89) {
*(u16 *)(a0 + 0x34) += 1;
func_80016714(*(void **)(a0 + 0xCC), 0x84);
*(s32 *)(a0 + 0xCC) = 0;
if (*(s16 *)(a0 + 0x70) == 1) {
off = -((u16)(tbl[0x1D] / 3) >= tbl[0x1E]) & 0x1E;
if ((u16)(tbl[0x1F] / 3) >= tbl[0x20]) {
off += 0x3C;
}
tp = &D_801CB398[off];
i = rand() % 100;
while (*tp <= i) {
tp += 3;
}
t = func_8012C658(tp[1], tp[2], a0);
if (t != 0) {
*(u16 *)(t + 0xA) -= 0x40;
}
}
}
break;
case 2:
*(u16 *)(a0 + 0xFC) = cnt = *(u16 *)(a0 + 0xFC) + 1;
if (cnt >= 0x20) {
func_80016714(*(void **)(a0 + 0xD0), 0x84);
*(s32 *)(a0 + 0xD0) = 0;
*(u16 *)(a0 + 0x34) += 1;
if (*(s16 *)(a0 + 0x70) == 2) {
*(u16 *)(a0 + 0x34) = 0;
*(u16 *)(a0 + 2) = 4;
}
func_8002A04C(a0);
}
break;
}
}
extern void func_8012BE54(s32);
extern s32 func_8012B8A4(s32);