phase12: track verified func_80042E10 source

src/func_80042E10.c

@@ -0,0 +1,75 @@
+/*
+ * func_80042E10 — 88 bytes at 0x80042E10..0x80042E68
+ *
+ * PHASE 12 RESUME LANE C (autonomous matching). Row from the UN-ATTEMPTED band
+ * (negatives-d-unattempted.tsv) filtered through config/function_extents.tsv
+ * (size 88 grade exact term=jr_ra). NOT previously in the ledger.
+ *
+ * MATCH: candidate_bytes=88, differing_bytes=0, result=MATCH, exit 0 on the DEFAULT toolchain
+ * (tools/old-gcc/gcc-2.7.2-psx/cc1). No region overrides.
+ *
+ * Hypothesis, not a claim about meaning: `a2` is a BIT count; `n = a2 >> 5` is the word index at
+ * which the range ends. Walk words `n` down to `0` of two same-sized arrays and count how many
+ * satisfy `(x & y) == y`; report whether ALL `n + 1` of them do. Used with `n` = the last word the
+ * two bit maps share, this is a "is every remaining bit of x present in y" test.
+ *
+ * Original words (22):
+ *     00063143  sra   a2,a2,0x5          n = a2 >> 5
+ *     00C03821  move  a3,a2              i = n
+ *     00004021  move  t0,zero            count = 0
+ *     2409FFFF  li    t1,-1              ---- the end value lives in a REGISTER ----
+ *     00061080  sll   v0,a2,0x2
+ *     00A22821  addu  a1,v0,a1           a1 += n * 4
+ *     00A22021  addu  a0,v0,a0           a0 += n * 4
+ * $L: 8C820000  lw    v0,0(a0)
+ *     8CA30000  lw    v1,0(a1)
+ *     00000000  nop
+ *     00431024  and   v0,v0,v1
+ *     14430002  bne   v0,v1,0x80042e48   if ((x & y) != y) skip the count
+ *     00000000  nop
+ *     25080001  addiu t0,t0,1            count++
+ *     24A5FFFC  addiu a1,a1,-4
+ *     24E7FFFF  addiu a3,a3,-1           i--     <-- INSIDE the loop body
+ *     14E9FFF6  bne   a3,t1,0x80042e2c   <-- tested AFTER the decrement, against t1
+ *     2484FFFC  addiu a0,a0,-4           (delay slot)
+ *     24C20001  addiu v0,a2,1            n + 1
+ *     01021026  xor   v0,t0,v0
+ *     03E00008  jr    ra
+ *     2C420001  sltiu v0,v0,1            (count ^ (n + 1)) < 1  ==  count == n + 1
+ *
+ * THE LEVER: THE LOOP MUST BE A DO-WHILE WHOSE DECREMENT IS IN THE CONDITION, COUNTING DOWN TO -1.
+ * The natural `for (i = n; i >= 0; i--)` spelling is 88 bytes with 16 differing bytes: it emits
+ * `bltz a3,<exit>` and NEVER materialises -1, because gcc tests `i >= 0` with a register-immediate
+ * branch. Writing
+ *     i = n;  count = 0;
+ *     do { ... body ...; i--; } while (i != -1);
+ * makes gcc hoist `li t1,-1` BEFORE the loop and test `bne a3,t1` after the decrement, which is
+ * the original's form byte-for-byte. (The `for (i = n; i != -1; i--)` spelling is NOT equivalent
+ * here: it is 96 bytes, because a for-loop's test is placed at the top.) **The `li <reg>,-1` +
+ * `bne` pair is therefore a fingerprint of a do-while countdown to -1, not of a `!= -1` condition
+ * written in any loop form** -- the same fingerprint appears in 0x800A86B4, where the natural `i >= 0`
+ * spelling also emits `bgez` and one induction variable instead of the original's two pointers.
+ * SECOND, SMALLER: the statement order matters -- `i = n;` must precede `count = 0;` for the
+ * prologue (`sra` / `move a3,a2` / `move t0,zero` / `li t1,-1`) to line up.
+ *
+ * LIMITS: `a2` is treated as a bit count (`>> 5`) because only that makes the `* 4` byte scaling
+ * consistent, and the two arrays are assumed to be bit maps of the same size; neither is
+ * independently evidenced. The `==` is compiled to `(a ^ b) < 1` (gcc's equality idiom for this
+ * shape), which is why the source's `count == n + 1` does not appear as a `beq`.
+ */
+
+int func_80042E10(int a0, int a1, int a2)
+{
+    int n = a2 >> 5;
+    int count;
+    int i;
+
+    i = n;
+    count = 0;
+    do {
+        if ((*(int *)(a0 + i * 4) & *(int *)(a1 + i * 4)) == *(int *)(a1 + i * 4))
+            count++;
+        i--;
+    } while (i != -1);
+    return count == n + 1;
+}