Mesh Formats
Tools
-
cmesh and smesh share the same structure, called static_mesh (not confusing). The cpu file gets a v_file_header, followed by:
- static_mesh
- Texture names
- Navpoints
- Collision spheres
- Collision cylinders
- Rig bone indices
- Mesh data
- Materials
- VIDs
- submesh LOD info
The relevant structures for static_mesh:
// Navigation reference point
struct static_mesh_navp {
char name[MAX_NAVP_NAME_LENGTH]; // name to reference nav point by
et_int vid; // vid this navp is attached to.
et_fl_vector pos; // position of navpoint in object coords
et_fl_quaternion orient; // quaternion representation of navpoint
};
struct cmesh_csphere {
et_uint32 body_part_id; // ID for the body part this represents
et_int parent_index; // index of parent bone (-1 if none)
et_fl_vector pos; // position of collision sphere in object coords
et_float radius; // radius of collision sphere
};
struct cmesh_ccylinder {
et_uint32 body_part_id; // ID for the body part this represents
et_int parent_index;
et_fl_vector axis;
et_fl_vector pos;
et_float radius;
et_float height;
};
class static_mesh {
public:
et_int signature;
et_short version;
et_short mesh_flags;
// grouping the various count fields together to minimize padding waste
et_short num_navpoints;
et_short num_rig_bones; // Number of bones this character mesh uses from the RIG file.
et_short num_materials; // number of entries in materials array
et_short num_material_maps; // number of entries in material_maps and material_map_name_crcs arrays
et_short num_lods_per_submesh; //number of lods per submesh. All submeshes must have same number of lods, currently.
et_uint16 num_submesh_vids; //this will be >= num_logical_submeshes (there can be additional vids tacked on the end purely for informational reasons)
V_FILE_PTR(static_mesh_navp *) navpoints; // prop points, I think.
V_FILE_PTR(et_uint *) rig_bone_indices; // List of RIG bone indices used by this character mesh.
et_fl_vector bounding_center; // for some reason, these are only in VIF meshes.
et_float bounding_radius;
// collision primitives for inter-character hit detection
et_short m_num_cspheres; // collider spheres used for hit detection
et_short m_num_ccylinders; // collider cylinders used for hit detection
// 4 BYTES PADDING IF V_FILE_PTR is 8 bytes
V_FILE_PTR(cmesh_csphere *) m_cspheres;
V_FILE_PTR(cmesh_ccylinder *) m_ccylinders;
// rendering mesh data
V_FILE_PTR(rl_mesh *) mesh;
//NOTE: It is possible for material_maps and materials to be NULL. That means the materials are expected to be provided externally.
V_FILE_PTR(V_FILE_PTR(rl_material_map*) *) material_maps; // num_material_maps sized array of material maps
V_FILE_PTR(et_uint32 *) material_map_name_crcs; // num_material_maps sized array of name crc for each material map
V_FILE_PTR(V_FILE_PTR(rl_static_material *) *) materials; // num_materials sized array of materials
//logical number of submeshes (same as actual number of submeshes only if we don't have lods)
et_uint32 num_logical_submeshes;
// 4 BYTES PADDING IF V_FILE_PTR is 8 bytes
//This pointer will either be null (no VID info) or will point to an array of ints of length num_logical_submeshes
//It will contain the VID corresponding to each logical submesh
V_FILE_PTR(et_uint32 *) submesh_vids;
//This pointer will either be null (no LOD info) or will point to a (flat) 2D array, representing the rl_mesh submesh indexes for each lod level of each "logical submesh"
//submesh_lod_info[logical_submesh_index*num_lods_per_submesh + lod_level] --> index into rl_mesh's flat array of submeshes.
V_FILE_PTR(et_short *) submesh_lod_info;
};
The CPU mesh data itself has a little header information: [LIST] []Version (uint16 + 4 byte pad) []CRC to validate mapping with the GPU file (uint32) []CPU size (uint32) []GPU size (uint32) [/LIST] Then we get the data:
struct rl_vertex_buffer_data
{
et_int32 num_verts; // number of vertices in the buffer
uint8 vert_stride_0; // stride (size) of each vertex (stream 0)
uint8 vertex_format; // vertex format
uint8 num_uv_channels; // number of uv channels (from 0 to 4)
uint8 vert_stride_1; // stride (size) of each vertex (stream 1)
V_FILE_PTR(void *) render_data; // pointer to platform specific data we need for this vertex buffer (created at run-time)
V_FILE_PTR(rl_mesh_generic_vertex *) verts;
};
struct rl_index_buffer_data
{
et_uint32 num_indices; // number of indices in the index buffer
// 4 BYTE PAD WHEN V_FILE_PTR IS 8 BYTES
V_FILE_PTR(rl_generic_index *) indices; // can be either et_uint16 or et_uint32
uint8 index_size; // size of indices in bytes (2 or 4)
uint8 prim_type; // keep it 4 byte aligned
et_uint16 num_blocks;
};
struct rl_bone_group_data
{
uint8 num_group_bones; // How many entries in the mapped_bone_list this group uses.
uint8 group_bone_offset; // Index in the mapped_bone_list of the first bone in this group.
};
struct rl_bone_map_data
{
et_uint16 num_mapped_bones;
// 2 BYTES OR 6 BYTES PAD DEPENDING ON V_FILE_PTR SIZE
V_FILE_PTR(uint8 *) mapped_bone_list;
et_uint16 num_bone_groups;
// 2 BYTES OR 6 BYTES PAD DEPENDING ON V_FILE_PTR SIZE
V_FILE_PTR(rl_bone_group_data *) bone_group_list;
};
struct rl_mesh_data
{
et_uint32 mesh_flags;
et_int32 num_sub_meshes;
V_FILE_PTR(rl_submesh_data *) sub_meshes;
et_uint32 num_vertex_buffers;
// 4 BYTE PAD WHEN V_FILE_PTR IS 8 BYTES
V_FILE_PTR(rl_vertex_buffer_data *) vertex_buffers;
rl_index_buffer_data index_buffer;
rl_bone_map_data bone_map;
// these were moved out of vertex buffer data structure when we put in
// multiple vertex formats, because it does not make sense to have different
// scales/offsets for different vertex buffers on the same mesh.
et_fl_vector position_scale; // scale to apply to vertex positions
et_fl_vector position_offset; // offset to apply to vertex positions
};
Vertex format will be one of these (most likely float3 or xcposition):
enum rl_vertex_attribute_type {
RLVA_INVALID_TYPE = -1,
////////////////////////
// Floating Point Types
RLVAT_FLOAT1 = 0,
RLVAT_FLOAT2,
RLVAT_FLOAT3,
RLVAT_FLOAT4,
////////////////////////
// Half Float Types
RLVAT_HALF2,
RLVAT_HALF4,
////////////////////////
// Byte Types
RLVAT_UBYTE4,
RLVAT_UBYTE4N,
////////////////////////
// Short Types
RLVAT_SHORT2N,
RLVAT_SHORT4N,
RLVAT_SHORT2,
RLVAT_SHORT4,
////////////////////////
// Compressed Normal Meta Types
RLVAT_CNORMAL,
RLVAT_CTANGENT,
////////////////////////
// Color Meta Types
RLVAT_COLOR4,
////////////////////////
// Compressed Position Meta Types
RLVAT_CPOSITION,
RLVAT_XCPOSITION,
NUM_RL_VERTEX_ATTRIBUTE_TYPES
};
lmesh files are for level meshes. They also get a v file header, then this header structure:
struct level_mesh_aux_data {
et_fl_bbox m_bbox;
// these are flags stored about the entries. There are num_prims/2 bytes in this array.
// Each nibble has the information for a single primitive with the least
// significant bit having the first primitive's flags, ie, ((m_flags[i]) & 0xF) = first primitive
// (flag & 0xF0) >> 4 = second primitive
uint8 m_flags[0];
};
struct level_mesh_subinstance {
public:
et_fl_matrix43 m_transform;
et_uint16 m_submesh_index;
et_uint16 m_flags;
};
struct level_mesh_zbias_decal {
public:
et_fl_matrix43 m_transform;
et_int m_decal_layer;
et_uint32 m_mesh_index;
};
struct level_mesh_stitch_edge {
public:
et_uint32 m_num_knots;
V_FILE_PTR(et_fl_vector *) m_knots;
};
struct level_mesh_posdummy {
et_int m_type; // use level_mesh_posdummy_types
et_fl_matrix43 m_transform;
};
struct level_mesh_occluder {
et_vm_matrix44 m_transform;
et_vm_bbox m_aabb;
};
struct level_mesh_lod_array {
static const int max_lods = 3;
level_mesh_lod_array()
{
for (int i = 0; i < max_lods; i++) {
m_lods[i] = NULL;
m_variant_values[i] = NULL;
}
m_level_mesh_name[0]=0;
}
rl_multi_mesh_instance::lod* m_lods[max_lods];
level_mesh_variant_value *m_variant_values[max_lods];
linked_list_p<level_mesh_header> m_level_meshes[max_lods];
// debug cursor (which also works in release) name
static const int max_name_len = 48;
char m_level_mesh_name[max_name_len];
};
struct level_mesh_material_variant {
et_uint32 m_material_name_checksum;
et_uint16 m_variant_0_id;
et_uint16 m_variant_1_id;
};
class level_mesh_header {
public:
et_uint32 m_signature;
et_uint16 m_version;
// 2 BYTES PADDING
et_uint32 m_flags;
et_fl_vector m_bounding_center;
et_float m_bounding_radius;
et_vm_bbox m_local_aabb;
// :ANDYC:TODO: remove when zbias decals are part of multi mesh instance
// :BEGIN REMOVE LATER:
struct {
// rendering mesh data
// 4 BYTES PADDING IF V_FILE_PTRS are 8 bytes
V_FILE_PTR(V_FILE_PTR(rl_mesh *)*) m_base_meshes;
et_uint32 m_num_base_meshes;
V_FILE_PTR(V_FILE_PTR(rl_material_map *) *) m_base_material_maps;
et_uint32 m_num_base_material_maps;
// 4 BYTES PADDING IF V_FILE_PTRS are 8 bytes
V_FILE_PTR(const char *) m_texture_names;
V_FILE_PTR(level_mesh_zbias_decal *) m_zbias_decals;
et_uint32 m_num_zbias_decals;
} m_decals;
// :END REMOVE LATER:
et_int32 m_cm_index;
et_int32 m_corpse_cm_index;
V_FILE_PTR(hkpShape*) m_static_collision_model;
et_int32 m_num_stitch_edges;
V_FILE_PTR(level_mesh_stitch_edge *) m_stitch_edges;
et_int32 m_num_posdummies;
// 4 BYTES PADDING IF V_FILE_PTRS are 8 bytes
V_FILE_PTR(level_mesh_posdummy *) m_posdummies;
et_int32 m_num_occluders;
// 4 BYTES PADDING
V_WIDE_PTR(level_mesh_occluder *) m_occluders; // Using a wide ptr here to make decrementing pad bytes simpler.
V_FILE_PTR(rl_multi_mesh_data*) m_multi_mesh_lod;
et_fl_vector m_collision_bmin;
et_fl_vector m_collision_bmax;
// default values for when no variants are provided
et_uint m_num_material_variants;
V_WIDE_PTR(level_mesh_material_variant *) m_material_variants;
et_uint m_num_variant_values;
V_WIDE_PTR(level_mesh_variant_value *) m_variant_values;
// crc maps for each lod
et_uint m_this_lod_map_index;
et_uint m_num_variant_crc_maps;
V_WIDE_PTR(et_uint*) m_variant_crc_maps_counts;
V_WIDE_PTR(V_WIDE_PTR(et_uint32 *)*) m_variant_crc_maps;
// for keeping track to duplicate references in the level_mesh_lod_array code
V_WIDE_PTR(level_mesh_header*) next;
V_WIDE_PTR(level_mesh_header*) prev;
volatile uint32 m_preload_done;
uint8 m_pad_bytes[16]; // will always be zeroed. Use this space for easier versioning of header.
};
matlib files have a v file header, then this header:
struct material_library {
et_uint32 m_signature;
et_uint32 m_version;
// RL data
V_FILE_PTR(rl_material_map *) m_material_map;
V_FILE_PTR(V_FILE_PTR(rl_static_material *) *) m_materials;
et_uint m_num_materials;
// 4 BYTES PAD IF V_FILE_PTRS ARE 8 BYTES
V_FILE_PTR(char const *) m_texture_names;
// in the 1:1 file the following are tagged on
// << material data
};
The actual material data structure looks like this:
struct rl_material_const {
et_float elem[4];
};
// 1:1 data structure for writing texture descs to disk
//
struct rl_material_texture_desc {
rl_texture_handle texture_handle;
et_uint32 name_checksum; //the crc_stri() of the name of the texture semantic in the shader file, not the texture filename
et_uint16 texture_stage;
et_uint16 texture_flags;
};
// 1:1 data structure for writing materials to disk
//
struct rl_material_data {
rl_shader_handle shader_handle;
et_uint32 name_checksum;
// flags for the material see the RLMF_* defines
et_uint32 mat_flags;
// packed into 4 bytes
et_uint16 num_textures;
uint8 num_constants; // the number of quadwords in the constant block
uint8 max_constants; // the actual space allocated (extra space for adding constants in shaders at runtime)
V_FILE_PTR(rl_material_texture_desc *) textures;
V_FILE_PTR(et_uint32 *) constant_name_checksum; // array of crc`s of the constant names in the shader file
V_FILE_PTR(rl_material_const *) constant_block; // the pointer to the block of data that get's upload to the user constants
// There is no need to store the alpha version of a shader in a material... a material is either alpha or not alpha. So in the case of an alpha
// material, the shader handle will point to the alpha version of the inferred lighting shader. Only keeping the below field around until the next
// global recrunch.
rl_shader_handle alpha_shader_handle; // shader handle for alpha version of shader (only used with inferred rendering)
};