625 lines
15 KiB
C
625 lines
15 KiB
C
#include "model.h"
|
|
|
|
#include "log.h"
|
|
#include "resources.h"
|
|
#include "stb_ds.h"
|
|
#include "gameobject.h"
|
|
|
|
#include "render.h"
|
|
|
|
#include "HandmadeMath.h"
|
|
|
|
#include "math.h"
|
|
#include "time.h"
|
|
|
|
#include <cgltf.h>
|
|
|
|
#include <limits.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include "yugine.h"
|
|
|
|
#include "jsffi.h"
|
|
|
|
#include "texture.h"
|
|
|
|
#include "sokol/sokol_gfx.h"
|
|
|
|
static void processnode();
|
|
static void processmesh();
|
|
static void processtexture();
|
|
|
|
static cgltf_data *cdata;
|
|
static char *cpath;
|
|
|
|
cgltf_attribute *get_attr_type(cgltf_primitive *p, cgltf_attribute_type t)
|
|
{
|
|
for (int i = 0; i < p->attributes_count; i++) {
|
|
if (p->attributes[i].type == t)
|
|
return &p->attributes[i];
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
unsigned short pack_short_tex(float c) { return c * USHRT_MAX; }
|
|
|
|
|
|
void mesh_add_material(primitive *prim, cgltf_material *mat)
|
|
{
|
|
if (!mat) return;
|
|
|
|
prim->mat = calloc(sizeof(*prim->mat), 1);
|
|
material *pmat = prim->mat;
|
|
|
|
if (mat->has_pbr_metallic_roughness && mat->pbr_metallic_roughness.base_color_texture.texture) {
|
|
cgltf_image *img = mat->pbr_metallic_roughness.base_color_texture.texture->image;
|
|
if (img->buffer_view) {
|
|
cgltf_buffer_view *buf = img->buffer_view;
|
|
pmat->diffuse = texture_fromdata(buf->buffer->data, buf->size);
|
|
} else {
|
|
// char *path = makepath(dirname(cpath), img->uri);
|
|
// pmat->diffuse = texture_from_file(path);
|
|
// free(path);
|
|
}
|
|
} else
|
|
pmat->diffuse = texture_from_file("icons/moon.gif");
|
|
}
|
|
|
|
sg_buffer texcoord_floats(float *f, int n)
|
|
{
|
|
unsigned short packed[n];
|
|
for (int i = 0; i < n; i++) {
|
|
float v = f[i];
|
|
if (v < 0) v = 0;
|
|
if (v > 1) v = 1;
|
|
packed[i] = pack_short_tex(v);
|
|
}
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){
|
|
.data = SG_RANGE(packed),
|
|
.label = "tex coord vert buffer",
|
|
});
|
|
}
|
|
|
|
sg_buffer par_idx_buffer(uint32_t *p, int v)
|
|
{
|
|
uint16_t idx[v];
|
|
for (int i = 0; i < v; i++) idx[i] = p[i];
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){
|
|
.data = SG_RANGE(idx),
|
|
.type = SG_BUFFERTYPE_INDEXBUFFER
|
|
});
|
|
}
|
|
|
|
sg_buffer float_buffer(float *f, int v)
|
|
{
|
|
return sg_make_buffer(&(sg_buffer_desc){
|
|
.data = (sg_range){
|
|
.ptr = f,
|
|
.size = sizeof(*f)*v
|
|
}
|
|
});
|
|
}
|
|
|
|
sg_buffer index_buffer(float *f, int verts)
|
|
{
|
|
uint16_t idxs[verts];
|
|
for (int i = 0; i < verts; i++)
|
|
idxs[i] = f[i];
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){
|
|
.data = SG_RANGE(idxs),
|
|
.type = SG_BUFFERTYPE_INDEXBUFFER,
|
|
});
|
|
}
|
|
|
|
uint32_t pack_int10_n2(float *norm)
|
|
{
|
|
uint32_t ret = 0;
|
|
for (int i = 0; i < 3; i++) {
|
|
int n = (norm[i]+1.0)*511;
|
|
ret |= (n & 0x3ff) << (10*i);
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
sg_buffer normal_floats(float *f, int n)
|
|
{
|
|
return float_buffer(f, n);
|
|
uint32_t packed_norms[n/3];
|
|
for (int v = 0, i = 0; v < n/3; v++, i+= 3)
|
|
packed_norms[v] = pack_int10_n2(f+i);
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){
|
|
.data = SG_RANGE(packed_norms),
|
|
.label = "normal vert buffer",
|
|
});
|
|
}
|
|
|
|
sg_buffer ubyten_buffer(float *f, int v)
|
|
{
|
|
unsigned char b[v];
|
|
for (int i = 0; i < (v); i++)
|
|
b[i] = f[i]*255;
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){.data=SG_RANGE(b)});
|
|
}
|
|
|
|
sg_buffer ubyte_buffer(float *f, int v)
|
|
{
|
|
unsigned char b[v];
|
|
for (int i = 0; i < (v); i++)
|
|
b[i] = f[i];
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){.data=SG_RANGE(b)});
|
|
}
|
|
|
|
sg_buffer joint_buf(float *f, int v)
|
|
{
|
|
char joints[v];
|
|
for (int i = 0; i < (v); i++)
|
|
joints[i] = f[i];
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){ .data = SG_RANGE(joints)});
|
|
}
|
|
|
|
sg_buffer weight_buf(float *f, int v)
|
|
{
|
|
unsigned char weights[v];
|
|
for (int i = 0; i < (v); i++)
|
|
weights[i] = f[i]*255;
|
|
|
|
return sg_make_buffer(&(sg_buffer_desc){ .data = SG_RANGE(weights)});
|
|
}
|
|
|
|
HMM_Vec3 index_to_vert(uint32_t idx, float *f)
|
|
{
|
|
return (HMM_Vec3){f[idx*3], f[idx*3+1], f[idx*3+2]};
|
|
}
|
|
|
|
void primitive_gen_indices(primitive *prim)
|
|
{
|
|
if (prim->idx_count != 0) return;
|
|
|
|
uint16_t *idxs = malloc(sizeof(*idxs)*prim->idx_count);
|
|
|
|
for (int z = 0; z < prim->idx_count; z++)
|
|
idxs[z] = z;
|
|
|
|
prim->index = sg_make_buffer(&(sg_buffer_desc){
|
|
.data.ptr = idxs,
|
|
.data.size = sizeof(uint16_t) * prim->idx_count,
|
|
.type = SG_BUFFERTYPE_INDEXBUFFER});
|
|
|
|
free(idxs);
|
|
}
|
|
|
|
struct primitive mesh_add_primitive(cgltf_primitive *prim)
|
|
{
|
|
primitive retp = (primitive){0};
|
|
|
|
uint16_t *idxs;
|
|
if (prim->indices) {
|
|
int n = cgltf_accessor_unpack_floats(prim->indices, NULL, 0);
|
|
float fidx[n];
|
|
cgltf_accessor_unpack_floats(prim->indices, fidx, n);
|
|
idxs = malloc(sizeof(*idxs)*n);
|
|
for (int i = 0; i < n; i++)
|
|
idxs[i] = fidx[i];
|
|
|
|
retp.index = sg_make_buffer(&(sg_buffer_desc){
|
|
.data.ptr = idxs,
|
|
.data.size = sizeof(*idxs) * n,
|
|
.type = SG_BUFFERTYPE_INDEXBUFFER,
|
|
.label = "mesh index buffer",
|
|
});
|
|
|
|
retp.idx_count = n;
|
|
free(idxs);
|
|
} else {
|
|
retp.idx_count = cgltf_accessor_unpack_floats(prim->attributes[0].data, NULL, 0);
|
|
primitive_gen_indices(&retp);
|
|
}
|
|
|
|
printf("adding material\n");
|
|
mesh_add_material(&retp, prim->material);
|
|
|
|
for (int k = 0; k < prim->attributes_count; k++) {
|
|
cgltf_attribute attribute = prim->attributes[k];
|
|
|
|
int n = cgltf_accessor_unpack_floats(attribute.data, NULL, 0); /* floats per vertex x num elements. In other words, total floats pulled */
|
|
int comp = cgltf_num_components(attribute.data->type);
|
|
int verts = n/comp;
|
|
float vs[n];
|
|
cgltf_accessor_unpack_floats(attribute.data, vs, n);
|
|
|
|
switch (attribute.type) {
|
|
case cgltf_attribute_type_position:
|
|
retp.pos = sg_make_buffer(&(sg_buffer_desc){
|
|
.data.ptr = vs,
|
|
.data.size = sizeof(float) * n,
|
|
.label = "mesh vert buffer"
|
|
});
|
|
break;
|
|
|
|
case cgltf_attribute_type_normal:
|
|
retp.norm = normal_floats(vs, n);
|
|
break;
|
|
|
|
case cgltf_attribute_type_tangent:
|
|
break;
|
|
|
|
case cgltf_attribute_type_color:
|
|
retp.color = ubyten_buffer(vs,n);
|
|
break;
|
|
|
|
case cgltf_attribute_type_weights:
|
|
retp.weight = ubyten_buffer(vs, n);
|
|
break;
|
|
|
|
case cgltf_attribute_type_joints:
|
|
retp.bone = ubyte_buffer(vs, n);
|
|
break;
|
|
|
|
case cgltf_attribute_type_texcoord:
|
|
retp.uv = texcoord_floats(vs, n);
|
|
break;
|
|
case cgltf_attribute_type_invalid:
|
|
YughWarn("Invalid type.");
|
|
break;
|
|
|
|
case cgltf_attribute_type_custom:
|
|
break;
|
|
|
|
case cgltf_attribute_type_max_enum:
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (!retp.bone.id) {
|
|
char joints[retp.idx_count*4];
|
|
memset(joints, 0, retp.idx_count*4);
|
|
retp.bone = sg_make_buffer(&(sg_buffer_desc){ .data = SG_RANGE(joints)});
|
|
}
|
|
|
|
if (!retp.weight.id) {
|
|
char weights[retp.idx_count*4];
|
|
memset(weights,0,retp.idx_count*4);
|
|
retp.weight = sg_make_buffer(&(sg_buffer_desc){ .data = SG_RANGE(weights)});
|
|
}
|
|
|
|
if (!retp.color.id) {
|
|
char colors[retp.idx_count*4];
|
|
memset(colors,0,retp.idx_count*4);
|
|
retp.color = sg_make_buffer(&(sg_buffer_desc) { .data = SG_RANGE(colors) });
|
|
}
|
|
|
|
if (!retp.norm.id) {
|
|
YughInfo("Making normals.");
|
|
cgltf_attribute *pa = get_attr_type(prim, cgltf_attribute_type_position);
|
|
int n = cgltf_accessor_unpack_floats(pa->data, NULL,0);
|
|
int comp = 3;
|
|
int verts = n/comp;
|
|
uint32_t face_norms[verts];
|
|
float ps[n];
|
|
cgltf_accessor_unpack_floats(pa->data,ps,n);
|
|
|
|
for (int i = 0; i < verts; i+=3) {
|
|
HMM_Vec3 a = index_to_vert(i,ps);
|
|
HMM_Vec3 b = index_to_vert(i+1,ps);
|
|
HMM_Vec3 c = index_to_vert(i+2,ps);
|
|
HMM_Vec3 norm = HMM_NormV3(HMM_Cross(HMM_SubV3(b,a), HMM_SubV3(c,a)));
|
|
uint32_t packed_norm = pack_int10_n2(norm.Elements);
|
|
face_norms[i] = face_norms[i+1] = face_norms[i+2] = packed_norm;
|
|
}
|
|
|
|
retp.norm = sg_make_buffer(&(sg_buffer_desc){
|
|
.data.ptr = face_norms,
|
|
.data.size = sizeof(uint32_t) * verts});
|
|
}
|
|
|
|
return retp;
|
|
}
|
|
|
|
|
|
|
|
void model_add_cgltf_mesh(mesh *m, cgltf_mesh *gltf_mesh)
|
|
{
|
|
for (int i = 0; i < gltf_mesh->primitives_count; i++)
|
|
arrput(m->primitives, mesh_add_primitive(gltf_mesh->primitives+i));
|
|
}
|
|
|
|
void packFloats(float *src, float *dest, int srcLength) {
|
|
int i, j;
|
|
for (i = 0, j = 0; i < srcLength; i += 3, j += 4) {
|
|
dest[j] = src[i];
|
|
dest[j + 1] = src[i + 1];
|
|
dest[j + 2] = src[i + 2];
|
|
dest[j + 3] = 0.0f;
|
|
}
|
|
}
|
|
|
|
void model_add_cgltf_anim(model *model, cgltf_animation *anim)
|
|
{
|
|
YughInfo("FOUND ANIM, using %d channels and %d samplers", anim->channels_count, anim->samplers_count);
|
|
|
|
struct animation an = (struct animation){0};
|
|
arrsetlen(an.samplers, anim->samplers_count);
|
|
|
|
for (int i = 0; i < anim->samplers_count; i++) {
|
|
cgltf_animation_sampler s = anim->samplers[i];
|
|
sampler samp = (sampler){0};
|
|
int n = cgltf_accessor_unpack_floats(s.input, NULL, 0);
|
|
arrsetlen(samp.times, n);
|
|
cgltf_accessor_unpack_floats(s.input, samp.times, n);
|
|
|
|
n = cgltf_accessor_unpack_floats(s.output, NULL, 0);
|
|
int comp = cgltf_num_components(s.output->type);
|
|
arrsetlen(samp.data, n/comp);
|
|
if (comp == 4)
|
|
cgltf_accessor_unpack_floats(s.output, samp.data, n);
|
|
else {
|
|
float *out = malloc(sizeof(*out)*n);
|
|
cgltf_accessor_unpack_floats(s.output, out, n);
|
|
packFloats(out, samp.data, n);
|
|
free(out);
|
|
}
|
|
|
|
samp.type = s.interpolation;
|
|
|
|
if (samp.type == LINEAR && comp == 4)
|
|
samp.type = SLERP;
|
|
|
|
an.samplers[i] = samp;
|
|
}
|
|
|
|
for (int i = 0; i < anim->channels_count; i++) {
|
|
cgltf_animation_channel ch = anim->channels[i];
|
|
struct anim_channel ach = (struct anim_channel){0};
|
|
md5joint *md = model->nodes+(ch.target_node-cdata->nodes);
|
|
switch(ch.target_path) {
|
|
case cgltf_animation_path_type_translation:
|
|
ach.target = &md->pos;
|
|
break;
|
|
case cgltf_animation_path_type_rotation:
|
|
ach.target = &md->rot;
|
|
break;
|
|
case cgltf_animation_path_type_scale:
|
|
ach.target = &md->scale;
|
|
break;
|
|
default: break;
|
|
}
|
|
ach.sampler = an.samplers+(ch.sampler-anim->samplers);
|
|
|
|
arrput(an.channels, ach);
|
|
}
|
|
|
|
model->anim = an;
|
|
model->anim.time = apptime();
|
|
}
|
|
|
|
void model_add_cgltf_skin(model *model, cgltf_skin *skin)
|
|
{
|
|
int n = cgltf_accessor_unpack_floats(skin->inverse_bind_matrices, NULL, 0);
|
|
struct skin sk = (struct skin){0};
|
|
arrsetlen(sk.invbind, n/16);
|
|
cgltf_accessor_unpack_floats(skin->inverse_bind_matrices, sk.invbind, n);
|
|
YughInfo("FOUND SKIN, of %d bones, and %d vert comps", skin->joints_count, n);
|
|
|
|
cgltf_node *root = skin->skeleton;
|
|
|
|
arrsetlen(sk.joints, skin->joints_count);
|
|
sk.root = model->nodes+(skin->skeleton-cdata->nodes);
|
|
|
|
for (int i = 0; i < 50; i++)
|
|
sk.binds[i] = MAT1;
|
|
|
|
for (int i = 0; i < skin->joints_count; i++) {
|
|
int offset = skin->joints[i]-cdata->nodes;
|
|
sk.joints[i] = model->nodes+offset;
|
|
md5joint *j = sk.joints[i];
|
|
cgltf_node *n = skin->joints[i];
|
|
for (int i = 0; i < 3; i++) {
|
|
j->pos.e[i] = n->translation[i];
|
|
j->scale.e[i] = n->scale[i];
|
|
}
|
|
for (int i = 0; i < 4; i++)
|
|
j->rot.e[i] = n->rotation[i];
|
|
}
|
|
|
|
model->skin = sk;
|
|
}
|
|
|
|
void model_process_node(model *model, cgltf_node *node)
|
|
{
|
|
int n = node-cdata->nodes;
|
|
cgltf_node_transform_world(node, model->nodes[n].t.e);
|
|
model->nodes[n].parent = model->nodes+(node->parent-cdata->nodes);
|
|
|
|
if (node->mesh) {
|
|
int meshn = node->mesh-cdata->meshes;
|
|
arrsetlen(model->meshes, meshn+1);
|
|
model->meshes[meshn].m = &model->nodes[n].t;
|
|
model_add_cgltf_mesh(model->meshes+meshn, node->mesh);
|
|
}
|
|
}
|
|
|
|
struct model *model_make(const char *path)
|
|
{
|
|
YughInfo("Making the model from %s.", path);
|
|
cpath = path;
|
|
cgltf_options options = {0};
|
|
cgltf_data *data = NULL;
|
|
cgltf_result result = cgltf_parse_file(&options, path, &data);
|
|
struct model *model = NULL;
|
|
|
|
if (result) {
|
|
YughError("CGLTF could not parse file %s, err %d.", path, result);
|
|
goto CLEAN;
|
|
}
|
|
|
|
result = cgltf_load_buffers(&options, data, path);
|
|
|
|
if (result) {
|
|
YughError("CGLTF could not load buffers for file %s, err %d.", path, result);
|
|
goto CLEAN;
|
|
}
|
|
|
|
cdata = data;
|
|
|
|
model = calloc(1, sizeof(*model));
|
|
|
|
arrsetlen(model->nodes, data->nodes_count);
|
|
for (int i = 0; i < data->nodes_count; i++)
|
|
model_process_node(model, data->nodes+i);
|
|
|
|
for (int i = 0; i < data->animations_count; i++)
|
|
model_add_cgltf_anim(model, data->animations+i);
|
|
|
|
for (int i = 0; i < data->skins_count; i++)
|
|
model_add_cgltf_skin(model, data->skins+i);
|
|
|
|
CLEAN:
|
|
cgltf_free(data);
|
|
return model;
|
|
}
|
|
|
|
void model_free(model *m)
|
|
{
|
|
|
|
}
|
|
|
|
sg_bindings primitive_bind(primitive *p)
|
|
{
|
|
sg_bindings b = {0};
|
|
b.vertex_buffers[MAT_POS] = p->pos;
|
|
b.vertex_buffers[MAT_UV] = p->uv;
|
|
b.vertex_buffers[MAT_NORM] = p->norm;
|
|
b.vertex_buffers[MAT_BONE] = p->bone;
|
|
b.vertex_buffers[MAT_WEIGHT] = p->weight;
|
|
b.vertex_buffers[MAT_COLOR] = p->color;
|
|
b.index_buffer = p->index;
|
|
b.fs.images[0] = p->mat->diffuse->id;
|
|
b.fs.samplers[0] = tex_sampler;
|
|
return b;
|
|
}
|
|
|
|
void model_draw_go(model *model, transform *go)
|
|
{
|
|
HMM_Mat4 gom = transform2mat(*go);
|
|
|
|
animation_run(&model->anim, apptime());
|
|
|
|
skin *sk = &model->skin;
|
|
for (int i = 0; i < arrlen(sk->joints); i++) {
|
|
md5joint *md = sk->joints[i];
|
|
HMM_Mat4 local = HMM_M4TRS(md->pos.xyz, md->rot, md->scale.xyz);
|
|
if (md->parent)
|
|
local = HMM_MulM4(md->parent->t, local);
|
|
md->t = local;
|
|
sk->binds[i] = HMM_MulM4(md->t, sk->invbind[i]);
|
|
}
|
|
/*sg_apply_uniforms(SG_SHADERSTAGE_VS, SLOT_skinv, &(sg_range){
|
|
.ptr = sk->binds,
|
|
.size = sizeof(*sk->binds)*50
|
|
});
|
|
*/
|
|
|
|
for (int i = 0; i < arrlen(model->meshes); i++) {
|
|
HMM_Mat4 mod = *model->meshes[i].m;
|
|
mod = HMM_MulM4(mod, gom);
|
|
mesh msh = model->meshes[i];
|
|
for (int j = 0; j < arrlen(msh.primitives); j++) {
|
|
sg_bindings b = primitive_bind(msh.primitives+j);
|
|
sg_apply_bindings(&b);
|
|
sg_draw(0, msh.primitives[j].idx_count, 1);
|
|
}
|
|
}
|
|
}
|
|
|
|
int mat2type(int mat)
|
|
{
|
|
switch(mat) {
|
|
case MAT_POS:
|
|
case MAT_NORM:
|
|
return SG_VERTEXFORMAT_FLOAT3;
|
|
case MAT_PPOS:
|
|
case MAT_WH:
|
|
case MAT_ST:
|
|
return SG_VERTEXFORMAT_FLOAT2;
|
|
case MAT_UV:
|
|
case MAT_TAN:
|
|
return SG_VERTEXFORMAT_USHORT2N;
|
|
return SG_VERTEXFORMAT_UINT10_N2;
|
|
case MAT_BONE:
|
|
return SG_VERTEXFORMAT_UBYTE4;
|
|
case MAT_WEIGHT:
|
|
case MAT_COLOR:
|
|
return SG_VERTEXFORMAT_UBYTE4N;
|
|
case MAT_ANGLE:
|
|
case MAT_SCALE:
|
|
return SG_VERTEXFORMAT_FLOAT;
|
|
};
|
|
return 0;
|
|
}
|
|
|
|
int mat2step(int mat)
|
|
{
|
|
switch(mat) {
|
|
case MAT_POS:
|
|
case MAT_UV:
|
|
case MAT_TAN:
|
|
case MAT_NORM:
|
|
case MAT_BONE:
|
|
case MAT_WEIGHT:
|
|
return SG_VERTEXSTEP_PER_VERTEX;
|
|
};
|
|
return SG_VERTEXSTEP_PER_INSTANCE;
|
|
}
|
|
|
|
sg_buffer mat2buffer(int mat, primitive *p)
|
|
{
|
|
switch(mat) {
|
|
case MAT_POS: return p->pos;
|
|
case MAT_NORM: return p->norm;
|
|
case MAT_UV: return p->uv;
|
|
case MAT_BONE: return p->bone;
|
|
case MAT_WEIGHT: return p->weight;
|
|
case MAT_COLOR: return p->color;
|
|
};
|
|
|
|
return p->pos;
|
|
}
|
|
|
|
sg_bindings primitive_bindings(primitive *p, JSValue v)
|
|
{
|
|
sg_bindings b = {0};
|
|
JSValue inputs = js_getpropstr(js_getpropstr(v, "vs"), "inputs");
|
|
for (int i = 0; i < js_arrlen(inputs); i++) {
|
|
JSValue attr = js_getpropidx(inputs, i);
|
|
int mat = js2number(js_getpropstr(attr, "mat"));
|
|
int slot = js2number(js_getpropstr(attr, "slot"));
|
|
sg_buffer buf = mat2buffer(mat,p);
|
|
if (!buf.id) {
|
|
// ERROR
|
|
}
|
|
b.vertex_buffers[slot] = buf;
|
|
}
|
|
|
|
b.index_buffer = p->index;
|
|
|
|
return b;
|
|
}
|
|
|
|
void primitive_free(primitive *prim)
|
|
{
|
|
|
|
}
|
|
|
|
void material_free(material *mat)
|
|
{
|
|
|
|
}
|