Dusk/WeirdUtils
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity
Files
08d1992eaf1d427b95ceb2fd27efc0b3a51d46ec
WeirdUtils/src/transform44/BONE_SSE_PROGRESS.md
T
MarcelineVQ 2b9b5bc043 bone_sse: assembly-verified reimplementation of transformMatrix4x4 
13 bugs fixed by comparing against full assembly dump (5317 instructions):
- Emitter check: this+0x188 -> this+0x1D8
- Animation time: added FILD*time_scale pattern for both primary (+0xB0)
  and secondary (+0xDC) slots
- Conditional multiply: bone_local *= *(bone_rt+0xF0) was missing
- Billboard post-processing: 4 switch cases (types 8/16/32/64) implemented
- Color animation loop bound: model_hdr+0x64 -> +0x6C
- Bone keyframe data stride: 0x24 -> 0x54
- Ribbon emitter output stride: 0x15C -> 0x170
- Particle data/output strides: 0x1FC/0x17C -> 0x1F8/0x16C
- Child SceneObject offsets: attach_idx +0x184->+0x1D4, next +0x190->+0x1E4
- Root bone parent: identity -> this+0xFC

New files:
- BONE_SSE_PROGRESS.md: section-by-section verification status
- t44_full_asm.txt: complete function assembly (ground truth)
- t44_helpers_asm.txt: all 12 helper function assemblies
- math_sse.zig: 18 x87->SSE polyfill stubs (VanillaFixes integration)

Also: OnWorldUpdate hook for true per-frame counting, DUMP_FRAMES=450.
SSE dispatch currently disabled while particle sections are being verified.
2026-03-14 17:34:37 -07:00

7.8 KiB
Raw Blame History

bone_sse.zig — Faithful Reproduction Progress

Principle

Exact reproduction of transformMatrix4x4 (0x714260, 17703 bytes) in Zig. Every section verified against assembly at src/transform44/decompiled/t44_full_asm.txt. The decompilation is NOT the primary source — assembly is authoritative.

Assembly References

  • decompiled/t44_full_asm.txt — 5317 instructions, complete function
  • decompiled/t44_helpers_asm.txt — 11 helper functions
  • SCENEOBJECT_OFFSETS.md — 51 [EBX+N] offsets

Section Status

Section 1: Entry checks (asm lines 5-16)

  • [EBX+0x10] NULL check, [EBX+0x40] vs [EAX+0x10] sync check
  • VERIFIED ✓

Section 2: Emitter setup (asm lines 17-36)

  • emitter_ctx+0x50 AND this+0x1D8 check (NOT +0x188!)
  • emitter_flag at this+0x50, copy emitter_ctx+0x17C to this+0x17C
  • VERIFIED ✓ — bug fixed: was using +0x188, now +0x1D8

Section 3: World position/scale (asm lines 37-74)

  • pos = mat2[i] * this[0x184/0x188/0x18C], stored at this+0x1A0
  • offset = mat3[i] + this[0x190/0x194/0x198], stored at this+0x1AC
  • render_scale_z = mat4 * this[0x180], stored at this+0x19C
  • VERIFIED ✓

Section 4: Global sequences (asm lines 75-95)

  • count/durations from model_hdr+0x14/0x18
  • timestamp from anim_ctx+0xC, base from this+0x68
  • values ptr at this+0x64, unsigned DIV for modulo
  • VERIFIED ✓

Section 5: initPPSG matrix multiply (asm lines 96-102)

  • output=this+0xFC, left=this+0xBC, right=mat1
  • Calling multiplyMatrix4x4_Basic (0x7507BB) directly
  • VERIFIED ✓

Section 6: child_objects_padding (asm lines 103-126)

  • emitter_ctx NULL or bit0 set: len_sq of world_xform[8,9,10]
  • else: copy from emitter_ctx+0x84
  • Stored at this+0x84
  • VERIFIED ✓

Section 7: Identity matrices + timestamp delta (asm lines 127-175)

  • Two 4x4 identity matrices (EBP-0x70 and EBP-0xE4)
  • Time delta from this+0x4C (search_data_base_ptr)
  • Bone loop setup: count from model_hdr+0x34, defs from +0x38
  • VERIFIED ✓

Section 8a: Bone loop start (asm lines 177-189)

  • bone_def = defs + idx0x6C, bone_rt = base + idx0x118
  • anim_slot at bone_rt+0xA4, checked against -1
  • VERIFIED ✓

Section 8b: Animation time computation (asm lines 190-262)

  • FILD * time_scale(+0xB0) → __ftol pattern for time conversion
  • Looping vs clamped via anim_entry+0x10 bit 0
  • VERIFIED ✓ — bug fixed: was missing time_scale multiplication

Section 8c: Secondary animation time (asm lines 284-364)

  • Same pattern with time_scale at +0xDC
  • Expiry check: timestamp - crossfade_end(+0x100) >= 0 → expire slot
  • VERIFIED ✓ — bug fixed: was "copy from primary", now full computation

Section 8d: Blend weight / crossfade (asm lines 382-447)

  • Hermite: (3.0 - 2*t) * t * t * crossfade_weight(+0x108)
  • t = (float)remaining * crossfade_inv(+0x104)
  • Clamped to [ZERO_THRESHOLD, 1.0]
  • PARTIALLY VERIFIED — formula matches, float comparison semantics unchecked

Section 8e: Parent inheritance + billboard PRE-processing (asm lines 448-478+)

  • Root: src = this+0xFC. Child: src = bone_out + parent*0x40
  • combined_flags = bone_rt[0xF4] | bone_def[0x04]
  • Billboard types 2/4/6 via (flags & 6)
  • PARTIALLY VERIFIED — offsets correct, billboard math from decompilation

Section 8f: Animation flags check (asm ~0x714D11)

  • (combined_flags & 0x280) == 0: copy parent directly
  • VERIFIED ✓

Section 8g: Rotation (asm 0x714D1F-0x714D52)

  • rot_nts at bone_def+0x34, rot AnimData at bone_def+0x28
  • buildRotMatFromQuat at 0x74B6B5 OVERWRITES matrix (does NOT multiply)
  • VERIFIED ✓

Section 8h: Scale + conditional multiply (asm 0x714DF7-0x714FA1)

  • scale_nts at bone_def+0x50, scale AnimData at bone_def+0x44
  • Conditional: if (char)flags < 0 AND bone_rt[0xF0] != 0: bone_local *= *(bone_rt+0xF0) via matmul 0x74A7C0
  • VERIFIED ✓ — bug fixed: conditional multiply was missing

Section 8i: Translation (asm 0x714FA1-0x7151BA)

  • trans_nts at bone_def+0x18, trans AnimData at bone_def+0x0C
  • offset = (pivot + interp) - matrix * pivot
  • Final multiply: output = bone_local × parent via matmul 0x74A7C0
  • PARTIALLY VERIFIED — structure matches, inline SSE multiply used

Section 8k: Billboard POST-processing (asm 0x7151F9-0x71594E)

  • flags & 0x78, switch on types 8/16/32/64
  • Scale lengths preserved, translation recomputed
  • IMPLEMENTED — from assembly, cross product formulas need double-check

Section 9: Texture animation (asm 0x715966-0x715C87)

  • count=model_hdr+0x54, data=+0x58, output=this+0xA0
  • Data stride 0x38, output stride 0x50
  • VERIFIED ✓ (strides confirmed from asm 0x715C70/0x715C73)

Section 10: Color animation (asm 0x715C87-0x715F1F)

  • Entry gate=model_hdr+0x64, loop bound=model_hdr+0x6C, data=+0x68
  • output=this+0xA8, data stride 0x1C, output stride 0x20
  • VERIFIED ✓ — bug fixed: loop bound was +0x64, now +0x6C

Section 11: Bone keyframe processing (asm 0x715F25-0x7163BC)

  • count=model_hdr+0x74, data=+0x78
  • Data stride 0x54, output stride 0x98, matrix stride 0x40
  • VERIFIED ✓ — bug fixed: data stride was 0x24, now 0x54

Section 12a: Ribbon emitters (asm 0x7163BC-0x716AD9)

  • count=model_hdr+0x11C, data=+0x120, output=this+0x200
  • Data stride 0xD4, output stride 0x170
  • VERIFIED ✓ — bug fixed: output stride was 0x15C, now 0x170

Section 12b: Particle emitters (asm 0x716AD9-0x71763E)

  • count=model_hdr+0x124, data=+0x128, output=this+0x3C4
  • Data stride 0x7C, output stride 0x84
  • VERIFIED ✓

Section 12c: Additional particles — model_hdr+0x134 (asm 0x71763E-0x717D6A)

  • count=model_hdr+0x134, data=+0x138, output=this+0x3C8
  • Data stride 0xDC, output stride 0xD0
  • Tracks: visibility(+0xC0/+0xCC), position(+0x24/+0x30), alpha(+0x40/+0x4C), speed(+0x5C/+0x68), emission(+0x78/+0x84), scale(+0xA4/+0xB0)
  • IMPLEMENTED — visibility + 5 tracks from assembly

Section 12d: additional_remaining reset (asm 0x717D6A-0x717D6F)

  • [EBX+0x3D8] = 0 between sections 12c and 12e
  • FIXED — now correctly placed between 0x134 and 0x13C sections

Section 12e: Final particle section — model_hdr+0x13C (asm 0x717D6A-0x7185E3)

  • count=model_hdr+0x13C, data=+0x140
  • output1=this+0x3D0, output2=this+0x3D4
  • Data stride 0x1F8, output stride 0x16C
  • All 10 tracks + visibility implemented from assembly: visibility(+0x1DC/+0x1E8), emission(+0x34/+0x40), speed(+0x50/+0x5C), color(+0x6C/+0x78), track4(+0x88/+0x94), Vec3_spline(+0xA4/+0xB0), track6(+0xC0/+0xCC), track7(+0xDC/+0xE8), track8(+0xF8/+0x104), track9(+0x114/+0x120), track10(+0x130/+0x13C)
  • IsParticleBufferEmpty call at 0x7B5F60
  • additional_remaining OR at [EBX+0x3D8]
  • VERIFIED ✓ (strides and track offsets from assembly)

Section 13: Attachment recursion (asm 0x7185E3-0x718784)

  • Child attach_idx at child+0x1D4, next_sibling at child+0x1E4
  • VERIFIED ✓

Section 14: Sync update (asm 0x718775-0x718784)

  • this+0x40 = *(anim_ctx + 0x10)
  • VERIFIED ✓

Bug Fix Log

# Bug Wrong Correct Assembly ref
1 Emitter check field this+0x188 this+0x1D8 line 28
2 Primary anim time no time_scale FILD*[brt+0xB0] 0x7145AB
3 Secondary anim time copy from primary full w/ [brt+0xDC] 0x714711
4 Conditional multiply missing bone_local*=*(brt+0xF0) 0x714F8D
5 Billboard post-proc TODO/missing 4 switch cases 0x7151F9
6 Color loop bound model_hdr+0x64 +0x6C 0x715F0A
7 Bone KF data stride 0x24 0x54 0x7163A2
8 Ribbon output stride 0x15C 0x170 0x716AC2
9 Final particle data 0x1FC 0x1F8 0x7185CD
10 Final particle output 0x17C 0x16C 0x7185BA
11 Child attach_idx +0x184 +0x1D4 0x718668
12 Child next_sibling +0x190 +0x1E4 0x718764
13 Root bone parent identity this+0xFC 0x714945
Reference in New Issue View Git Blame Copy Permalink