mirror of
https://github.com/Druthulu/BFM-decomp
synced 2026-09-27 22:45:39 -04:00
Drew T
845df54ca1
One clean whole-EXE rebuild verified the batch (gate_main), and main re-checked BYTE-IDENTICAL against config/check.us.sha before anything was credited. AverageZ3 DpqColor3 InvSquareRoot MatrixNormal OuterProduct12 Square12 VM_NOWON_OBJ_2A0 VM_NOWON_OBJ_44C VectorNormalSS func_80013478 func_8001C044 func_80047EC8 func_80049440 func_80062434 func_8006252C func_800625DC gfx2D_BG0_OBJ_658 gfx2D_BG0_OBJ_698 gfx2D_BG1_OBJ_648
457 lines
14 KiB
C
457 lines
14 KiB
C
#include "common.h"
|
|
|
|
/*
|
|
* func_80047D3C == PsyQ SquareRoot0(long a): hand-written GTE-assisted integer
|
|
* square root (splat marks it "Handwritten function"). The argument goes
|
|
* straight into the GTE (mtc2 $a0,$30 = lzcs), the leading-zero count comes
|
|
* back (mfc2 $v0,$31 = lzcr), the mantissa is normalised, a table at
|
|
* D_8006D8EC is indexed and the half-word result shifted out. Un-compilable
|
|
* tells: direct cop2 transfers with hand-placed hazard nops, `sub`/`addi`
|
|
* where cc1 emits `subu`/`addiu`, a dead `andi $t0,$v0,1`, a branch target
|
|
* that doubles as the preceding `b`'s delay slot (0x80047D84), and a second
|
|
* exit whose return value sits in the final `jr`'s delay slot. No -O2 C can
|
|
* emit this, so per cookbook §265 the body is banked as a verbatim file-scope
|
|
* __asm__ transcription of the target .s, with the full form-1 directive
|
|
* scaffold (.ent/.frame/.end) used by the two gate-proven banked precedents.
|
|
*
|
|
* maspsx interaction: the leading ".set noreorder" is SPACE-formed so maspsx's
|
|
* is_reorder tracker stays True and IT fills every empty branch delay slot --
|
|
* so the nops after beq/bltz/b/jr are deliberately NOT written by hand. The
|
|
* two mtc2 hazard nops ARE hand-written (maspsx knows nothing about cop2
|
|
* latency), and so is the nop after `lh $t5,%lo(...)($t5)`: maspsx's
|
|
* load-delay pass does not fire on this lh/base-reuse shape (measured).
|
|
* Immediates are decimal (maspsx int()s them); %hi/%lo relocate correctly.
|
|
*
|
|
* Recovered C semantics:
|
|
* long SquareRoot0(long a) {
|
|
* gte_ldLZCS(a); u = gte_stLZCR();
|
|
* if (u == 0x20) return 0;
|
|
* e = u & ~1;
|
|
* sh = (0x1f - e) >> 1;
|
|
* m = (e - 0x18 < 0) ? (a >> (0x18 - e)) : (a << (e - 0x18));
|
|
* return ((int)*(short *)&D_8006D8EC[(m - 0x40)] << sh) >> 12;
|
|
* }
|
|
*/
|
|
__asm__(
|
|
".text\n"
|
|
".align 2\n"
|
|
".globl func_80047D3C\n"
|
|
".ent func_80047D3C\n"
|
|
"func_80047D3C:\n"
|
|
".frame $sp, 0, $31\n"
|
|
".set noreorder\n"
|
|
"mtc2 $4, $30\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"mfc2 $2, $31\n"
|
|
"addiu $1, $0, 32\n"
|
|
"beq $2, $1, 3f\n"
|
|
"andi $8, $2, 1\n"
|
|
"addiu $10, $0, -2\n"
|
|
"and $10, $2, $10\n"
|
|
"addiu $9, $0, 31\n"
|
|
"sub $9, $9, $10\n"
|
|
"sra $9, $9, 1\n"
|
|
"addi $11, $10, -24\n"
|
|
"bltz $11, 1f\n"
|
|
"sllv $12, $4, $11\n"
|
|
"b 2f\n"
|
|
"1:\n"
|
|
"addiu $11, $0, 24\n"
|
|
"sub $11, $11, $10\n"
|
|
"srav $12, $4, $11\n"
|
|
"2:\n"
|
|
"addi $12, $12, -64\n"
|
|
"sll $12, $12, 1\n"
|
|
"lui $13, %hi(D_8006D8EC)\n"
|
|
"addu $13, $13, $12\n"
|
|
"lh $13, %lo(D_8006D8EC)($13)\n"
|
|
"nop\n"
|
|
"sllv $13, $13, $9\n"
|
|
"srl $2, $13, 12\n"
|
|
"jr $31\n"
|
|
"nop\n"
|
|
"3:\n"
|
|
"jr $31\n"
|
|
"addiu $2, $0, 0\n"
|
|
".set reorder\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
".end func_80047D3C\n"
|
|
);
|
|
|
|
__asm__(
|
|
".text\n"
|
|
".align 2\n"
|
|
".globl InvSquareRoot\n"
|
|
".ent InvSquareRoot\n"
|
|
"InvSquareRoot:\n"
|
|
".frame $sp, 0, $31\n"
|
|
".set noreorder\n"
|
|
"mtc2 $4, $30\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"mfc2 $2, $31\n"
|
|
"addiu $1, $0, 32\n"
|
|
"beq $2, $1, 3f\n"
|
|
"beq $2, $0, 3f\n"
|
|
"andi $8, $2, 1\n"
|
|
"addiu $10, $0, -2\n"
|
|
"and $10, $2, $10\n"
|
|
"addiu $9, $0, 31\n"
|
|
"sub $9, $9, $10\n"
|
|
"sra $9, $9, 1\n"
|
|
"addi $11, $10, -24\n"
|
|
"bltz $11, 1f\n"
|
|
"sllv $12, $4, $11\n"
|
|
"b 2f\n"
|
|
"1:\n"
|
|
"addiu $11, $0, 24\n"
|
|
"sub $11, $11, $10\n"
|
|
"srav $12, $4, $11\n"
|
|
"2:\n"
|
|
"addi $12, $12, -64\n"
|
|
"sll $12, $12, 1\n"
|
|
"lui $13, %hi(D_8006DA80)\n"
|
|
"addu $13, $13, $12\n"
|
|
"lh $13, %lo(D_8006DA80)($13)\n"
|
|
"sw $9, 0($6)\n"
|
|
"sw $13, 0($5)\n"
|
|
"addiu $2, $0, 1\n"
|
|
"jr $31\n"
|
|
"nop\n"
|
|
"3:\n"
|
|
"jr $31\n"
|
|
"addiu $2, $0, -1\n"
|
|
".set reorder\n"
|
|
".end InvSquareRoot\n"
|
|
);
|
|
|
|
__asm__(".text\n.align 2\n.globl func_80047E58\n.ent\tfunc_80047E58\n"
|
|
"func_80047E58:\n.frame $sp,0,$31\n.mask 0x00000000,0\n.fmask 0x00000000,0\n"
|
|
".set\tnoreorder\n"
|
|
"lw $t0, 0($a0)\n"
|
|
"lw $t1, 4($a0)\n"
|
|
"lw $t2, 8($a0)\n"
|
|
"b .L80047EAC\n"
|
|
"addu $a3, $ra, $zero\n"
|
|
".set\treorder\n.end\tfunc_80047E58\n");
|
|
|
|
|
|
/* VectorNormal @ 0x80047E6C -- HANDWRITTEN PsyQ libgte-style wrapper (cookbook §179-B/§179-C,
|
|
* §265 form 1; byte-proven exemplars: func_80059234 in src/800c.c, func_800D0440 in
|
|
* src/md_MAIN_003/md_MAIN_003.c).
|
|
*
|
|
* Tell-tales that this is NOT compiler output:
|
|
* - $ra is stashed in $a3 across the call and restored after
|
|
* (addu $a3,$ra,$zero / addu $ra,$a3,$zero) -- a hand-written leaf-caller idiom
|
|
* gcc-2.7.2 never emits;
|
|
* - the FINAL store (sw $t2,8($a1)) sits IN the jr $ra delay slot -- unreachable from a
|
|
* C function body, whose epilogue (jr $ra; nop) cc1 appends unconditionally;
|
|
* - callee func_80047EC8 is itself splat-marked "Handwritten function" (raw GTE sqr/gpf body).
|
|
*
|
|
* Hence the whole function is FILE-SCOPE basic asm (no enclosing C function => no generated
|
|
* prologue/epilogue), transcribed 1:1 from asm/nonmatchings/800b/VectorNormal.s:
|
|
* - ".ent\t"/".end\t" with literal TABS: maspsx process_line special-cases ".ent\t" and
|
|
* emits a fresh ".set noreorder"; a plain ".set noreorder" line is swallowed as state only.
|
|
* - every displacement/immediate DECIMAL (maspsx int(operand) is base-10).
|
|
* - delay-slot contents (the jal's nop, the jr's sw) written by hand, matching the target,
|
|
* same as the banked func_80059234 / func_80059FC0 bodies.
|
|
* - no .frame/.mask/.fmask: frameless (nothing saved, sp untouched); the banked
|
|
* func_80059234 passed the whole-binary gate without them, proving the gate ignores
|
|
* non-text frame metadata.
|
|
*
|
|
* Symbol audit vs the target's own relocation lines: the ONLY relocation in the target is
|
|
* the jal to func_80047EC8 (line 8 of the .s); spelled identically here. The function
|
|
* references no data symbols, no jumptables, and no hi-lo pairs.
|
|
*
|
|
* INTEGRATION NOTE: this REPLACES the INCLUDE_ASM("asm/nonmatchings/800b", VectorNormal)
|
|
* stub line in src/800b.c. The stub must be deleted, not kept beside this block -- a
|
|
* leftover stub defines the same symbol twice and the whole-binary gate fails at link
|
|
* scope (the §14b failure class match_one cannot see). Sibling VectorNormalSS (0x30,
|
|
* lh/sh variant, same handwritten idiom) banks with the identical template; its
|
|
* .L80047EAC label is dead text (single fall-through path).
|
|
*/
|
|
__asm__(
|
|
".text\n"
|
|
".align\t2\n"
|
|
".globl\tVectorNormal\n"
|
|
".ent\tVectorNormal\n"
|
|
"VectorNormal:\n"
|
|
".set\tnoreorder\n"
|
|
"lw $t0, 0($a0)\n"
|
|
"lw $t1, 4($a0)\n"
|
|
"lw $t2, 8($a0)\n"
|
|
"addu $a3, $ra, $zero\n"
|
|
"jal func_80047EC8\n"
|
|
"nop\n"
|
|
"addu $ra, $a3, $zero\n"
|
|
"sw $t0, 0($a1)\n"
|
|
"sw $t1, 4($a1)\n"
|
|
"jr $ra\n"
|
|
"sw $t2, 8($a1)\n"
|
|
".set\treorder\n"
|
|
".end\tVectorNormal\n"
|
|
);
|
|
|
|
__asm__(".text\n.align 2\n.globl VectorNormalSS\n.ent\tVectorNormalSS\n"
|
|
"VectorNormalSS:\n.frame $sp,0,$31\n.mask 0x00000000,0\n.fmask 0x00000000,0\n"
|
|
".set\tnoreorder\n"
|
|
"lh $t0, 0($a0)\n"
|
|
"lh $t1, 2($a0)\n"
|
|
"lh $t2, 4($a0)\n"
|
|
"nop\n"
|
|
"addu $a3, $ra, $zero\n"
|
|
".L80047EAC:\n"
|
|
"jal func_80047EC8\n"
|
|
"nop\n"
|
|
"addu $ra, $a3, $zero\n"
|
|
"sh $t0, 0($a1)\n"
|
|
"sh $t1, 2($a1)\n"
|
|
"jr $ra\n"
|
|
"sh $t2, 4($a1)\n"
|
|
".set\treorder\n.end\tVectorNormalSS\n");
|
|
|
|
__asm__(
|
|
".text\n"
|
|
".align\t2\n"
|
|
".globl\tfunc_80047EC8\n"
|
|
".ent\tfunc_80047EC8\n"
|
|
"func_80047EC8:\n"
|
|
".set\tnoreorder\n"
|
|
"mtc2 $t0, $9\n"
|
|
"mtc2 $t1, $10\n"
|
|
"mtc2 $t2, $11\n"
|
|
"nop\n"
|
|
"sqr 0\n"
|
|
"mfc2 $t3, $25\n"
|
|
"mfc2 $t4, $26\n"
|
|
"mfc2 $t5, $27\n"
|
|
"add $t3, $t3, $t4\n"
|
|
"add $v0, $t3, $t5\n"
|
|
"mtc2 $v0, $30\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"mfc2 $v1, $31\n"
|
|
"addiu $at, $zero, -2\n"
|
|
"and $v1, $v1, $at\n"
|
|
"addiu $t6, $zero, 31\n"
|
|
"sub $t6, $t6, $v1\n"
|
|
"addi $t3, $v1, -24\n"
|
|
"bltz $t3, 1f\n"
|
|
"sra $t6, $t6, 1\n"
|
|
"b 2f\n"
|
|
"sllv $t4, $v0, $t3\n"
|
|
"1:\n"
|
|
"addiu $t3, $zero, 24\n"
|
|
"sub $t3, $t3, $v1\n"
|
|
"srav $t4, $v0, $t3\n"
|
|
"2:\n"
|
|
"addi $t4, $t4, -64\n"
|
|
"sll $t4, $t4, 1\n"
|
|
"lui $t5, %hi(D_8006DA80)\n"
|
|
"addu $t5, $t5, $t4\n"
|
|
"lh $t5, %lo(D_8006DA80)($t5)\n"
|
|
"nop\n"
|
|
"mtc2 $t5, $8\n"
|
|
"mtc2 $t0, $9\n"
|
|
"mtc2 $t1, $10\n"
|
|
"mtc2 $t2, $11\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"gpf 0\n"
|
|
"mfc2 $t0, $25\n"
|
|
"mfc2 $t1, $26\n"
|
|
"mfc2 $t2, $27\n"
|
|
"srav $t0, $t0, $t6\n"
|
|
"srav $t1, $t1, $t6\n"
|
|
"srav $t2, $t2, $t6\n"
|
|
"jr $ra\n"
|
|
"nop\n"
|
|
".set\treorder\n"
|
|
".end\tfunc_80047EC8\n"
|
|
);
|
|
|
|
|
|
/*
|
|
* MatrixNormal (0x80047F84) -- HANDWRITTEN libgte-style GTE routine
|
|
* (splat marks every cop2 instruction "handwritten"): six lh loads of the source
|
|
* MATRIX row pairs, cfc2/ctc2 rotation-control save/set, mtc2 vector loads, two
|
|
* identical GTE compute ops (word 0x4B78000C -> cop2 0x178000C, same encoding
|
|
* rule as the banked GsTMDfastG3GL: emitted word = 0x4A000000 | imm), three tail
|
|
* calls to func_80047EC8 interleaved with sh runs to the out-matrix, and a
|
|
* return sequence `jr $ra` whose DELAY SLOT holds a live `sh $t2,0x10($a1)`
|
|
* plus one literal trailing nop.
|
|
*
|
|
* gcc-2.7.2 cannot express any of this from a C function body: the cop2 traffic
|
|
* is unreachable, and the live delay-slot return is impossible because cc1
|
|
* unconditionally appends its own `jr $ra; nop` epilogue (probed in this
|
|
* project: neither noreturn nor an asm-jump tail suppresses it). So the whole
|
|
* function is FILE-SCOPE __asm__ -- opaque text cc1 copies verbatim with no RTL
|
|
* function wrapped around it, exactly like func_80059234 (src/800c.c) and
|
|
* GsSortClear (src/gsgap3.c). The literal ".ent\tMatrixNormal" line is
|
|
* load-bearing: maspsx special-cases ".ent<TAB>" to emit a fresh
|
|
* ".set noreorder" (an ordinary .set line is swallowed, tracked only); writing
|
|
* glabel/endlabel instead would leave gas in reorder mode, auto-nopping every
|
|
* branch/jal slot and displacing the hand-placed delay-slot instructions.
|
|
* `jal func_80047EC8` carries the only relocations; the symbol is named exactly
|
|
* as the target's own relocation lines spell it (each jal encodes 0x80047EC8).
|
|
*/
|
|
__asm__(
|
|
".text\n"
|
|
".align\t2\n"
|
|
".globl\tMatrixNormal\n"
|
|
".ent\tMatrixNormal\n"
|
|
"MatrixNormal:\n"
|
|
".set\tnoreorder\n"
|
|
"lh $t0, 0($a0)\n"
|
|
"lh $t1, 2($a0)\n"
|
|
"lh $t2, 4($a0)\n"
|
|
"lh $t3, 6($a0)\n"
|
|
"lh $t4, 8($a0)\n"
|
|
"lh $t5, 10($a0)\n"
|
|
"cfc2 $v0, $0\n"
|
|
"cfc2 $v1, $2\n"
|
|
"cfc2 $a2, $4\n"
|
|
"ctc2 $t0, $0\n"
|
|
"ctc2 $t1, $2\n"
|
|
"ctc2 $t2, $4\n"
|
|
"mtc2 $t5, $11\n"
|
|
"mtc2 $t3, $9\n"
|
|
"mtc2 $t4, $10\n"
|
|
"nop\n"
|
|
"cop2 0x178000C\n"
|
|
"mfc2 $t7, $25\n"
|
|
"mfc2 $t8, $26\n"
|
|
"mfc2 $t9, $27\n"
|
|
"ctc2 $t3, $0\n"
|
|
"ctc2 $t4, $2\n"
|
|
"ctc2 $t5, $4\n"
|
|
"nop\n"
|
|
"cop2 0x178000C\n"
|
|
"mtc2 $t3, $0\n"
|
|
"mtc2 $t4, $1\n"
|
|
"mtc2 $t5, $2\n"
|
|
"mfc2 $t0, $25\n"
|
|
"mfc2 $t1, $26\n"
|
|
"mfc2 $t2, $27\n"
|
|
"ctc2 $v0, $0\n"
|
|
"ctc2 $v1, $2\n"
|
|
"ctc2 $a2, $4\n"
|
|
"addu $a3, $ra, $zero\n"
|
|
"jal func_80047EC8\n"
|
|
"nop\n"
|
|
"sh $t0, 0($a1)\n"
|
|
"sh $t1, 2($a1)\n"
|
|
"sh $t2, 4($a1)\n"
|
|
"mfc2 $t0, $0\n"
|
|
"mfc2 $t1, $1\n"
|
|
"mfc2 $t2, $2\n"
|
|
"jal func_80047EC8\n"
|
|
"nop\n"
|
|
"sh $t0, 6($a1)\n"
|
|
"sh $t1, 8($a1)\n"
|
|
"sh $t2, 10($a1)\n"
|
|
"addu $t0, $t7, $zero\n"
|
|
"addu $t1, $t8, $zero\n"
|
|
"jal func_80047EC8\n"
|
|
"addu $t2, $t9, $zero\n"
|
|
"addu $ra, $a3, $zero\n"
|
|
"sh $t0, 12($a1)\n"
|
|
"sh $t1, 14($a1)\n"
|
|
"jr $ra\n"
|
|
"sh $t2, 16($a1)\n"
|
|
"nop\n"
|
|
".set\treorder\n"
|
|
".end\tMatrixNormal\n"
|
|
);
|
|
|
|
void gteMIMefunc()
|
|
{
|
|
__asm__ __volatile__(
|
|
".set\tnoreorder\n"
|
|
"sll $6, $6, 3\n"
|
|
"add $6, $6, $5\n"
|
|
"mtc2 $7, $8\n"
|
|
".L80048078:\n"
|
|
"lw $8, 0($5)\n"
|
|
"lh $10, 4($5)\n"
|
|
"sra $9, $8, 16\n"
|
|
"andi $8, $8, 65535\n"
|
|
"mtc2 $8, $9\n"
|
|
"mtc2 $9, $10\n"
|
|
"mtc2 $10, $11\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"cop2 0x198003D\n"
|
|
"lw $2, 0($4)\n"
|
|
"lh $7, 4($4)\n"
|
|
"sra $3, $2, 16\n"
|
|
"andi $2, $2, 65535\n"
|
|
"addi $5, $5, 8\n"
|
|
"mfc2 $8, $9\n"
|
|
"mfc2 $9, $10\n"
|
|
"mfc2 $10, $11\n"
|
|
"add $2, $2, $8\n"
|
|
"add $3, $3, $9\n"
|
|
"add $7, $7, $10\n"
|
|
"sh $2, 0($4)\n"
|
|
"sh $3, 2($4)\n"
|
|
"sh $7, 4($4)\n"
|
|
"addi $4, $4, 8\n"
|
|
"bne $5, $6, .L80048078\n"
|
|
"nop\n"
|
|
".set\treorder\n"
|
|
);
|
|
}
|
|
|
|
__asm__(
|
|
".text\n"
|
|
".align 2\n"
|
|
".globl SquareRoot12\n"
|
|
"SquareRoot12:\n"
|
|
".set\tnoreorder\n"
|
|
"mtc2 $4, $30\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"mfc2 $2, $31\n"
|
|
"addiu $1, $0, 32\n"
|
|
".word 0x1041001d\n"
|
|
"nop\n"
|
|
"andi $8, $2, 1\n"
|
|
"addiu $10, $0, -2\n"
|
|
"and $10, $2, $10\n"
|
|
"addiu $9, $0, 19\n"
|
|
"sub $9, $9, $10\n"
|
|
"sra $9, $9, 1\n"
|
|
"addi $11, $10, -24\n"
|
|
".word 0x05600003\n"
|
|
"nop\n"
|
|
"sllv $12, $4, $11\n"
|
|
".word 0x10000003\n"
|
|
"addiu $11, $0, 24\n"
|
|
"sub $11, $11, $10\n"
|
|
"srav $12, $4, $11\n"
|
|
"addi $12, $12, -64\n"
|
|
"sll $12, $12, 1\n"
|
|
"lui $13, %hi(D_8006D8EC)\n"
|
|
"addu $13, $13, $12\n"
|
|
"lh $13, %lo(D_8006D8EC)($13)\n"
|
|
"nop\n"
|
|
".word 0x05200004\n"
|
|
"nop\n"
|
|
"sllv $2, $13, $9\n"
|
|
".word 0x03e00008\n"
|
|
"nop\n"
|
|
"neg $9, $9\n"
|
|
".word 0x03e00008\n"
|
|
"srlv $2, $13, $9\n"
|
|
".word 0x03e00008\n"
|
|
"addiu $2, $0, 0\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
"nop\n"
|
|
".set\treorder\n"
|
|
);
|