Matteo Bruni (@Mystral) commented about dlls/d3dx9_36/skin.c:
const D3DXMATRIX *bone_inv_transpose_transforms, const void *src_vertices, void *dst_vertices){
- FIXME("iface %p, bone_transforms %p, bone_inv_transpose_transforms %p, src_vertices %p, dst_vertices %p stub!\n",
iface, bone_transforms, bone_inv_transpose_transforms, src_vertices, dst_vertices);
- struct d3dx9_skin_info *skin = impl_from_ID3DXSkinInfo(iface);
- DWORD vertex_size = D3DXGetDeclVertexSize(skin->vertex_declaration, 0);
- return E_NOTIMPL;
- TRACE("iface %p, bone_transforms %p, bone_inv_transpose_transforms %p, src_vertices %p, dst_vertices %p.\n",
skin, bone_transforms, bone_inv_transpose_transforms, src_vertices, dst_vertices);- if (bone_inv_transpose_transforms)
- {
FIXME("Skinning vertices with two position elements is not supported.\n");return E_NOTIMPL;- }
Pretty sure that's not what the inverse of the transpose is for, regardless of what the documentation says.
Usually you want to use the inverse of the transpose of a transformation matrix specifically for normals, to make sure that they remain orthogonal with the transformed geometry. Not sure what's up with vertices with two position elements (if supported at all?), I guess it's possible that the comment from https://learn.microsoft.com/en-us/windows/win32/direct3d9/id3dxskininfo--upd... is also correct.
I tweaked a bit the test ([usm-test.txt](/uploads/27c8e1a1fed580424ee913b067bdda89/usm-test.txt)) to check my hunch and at least it looks like the inverse transpose matrix is used for normals when present.