1
0
Fork 0
forked from fte/fteqw
fteqw/engine/common/com_mesh.c
2013-11-11 05:14:45 +00:00

7061 lines
No EOL
193 KiB
C

#include "quakedef.h"
#include "com_mesh.h"
extern model_t *loadmodel;
extern char loadname[];
qboolean r_loadbumpmapping;
extern cvar_t dpcompat_psa_ungroup;
extern cvar_t r_noframegrouplerp;
//Common loader function.
void Mod_DoCRC(model_t *mod, char *buffer, int buffersize)
{
#ifndef SERVERONLY
//we've got to have this bit
if (loadmodel->engineflags & MDLF_DOCRC)
{
unsigned short crc;
qbyte *p;
int len;
char st[40];
QCRC_Init(&crc);
for (len = buffersize, p = buffer; len; len--, p++)
QCRC_ProcessByte(&crc, *p);
sprintf(st, "%d", (int) crc);
Info_SetValueForKey (cls.userinfo[0],
(loadmodel->engineflags & MDLF_PLAYER) ? pmodel_name : emodel_name,
st, sizeof(cls.userinfo[0]));
if (cls.state >= ca_connected)
{
CL_SendClientCommand(true, "setinfo %s %d",
(loadmodel->engineflags & MDLF_PLAYER) ? pmodel_name : emodel_name,
(int)crc);
}
if (!(loadmodel->engineflags & MDLF_PLAYER))
{ //eyes
loadmodel->tainted = (crc != 6967);
}
}
#endif
}
#if 1
#ifdef _WIN32
#include <malloc.h>
#else
#include <alloca.h>
#endif
extern cvar_t gl_part_flame, r_fullbrightSkins, r_fb_models;
extern cvar_t r_noaliasshadows;
extern cvar_t r_skin_overlays;
extern cvar_t mod_md3flags;
typedef struct
{
char *name;
float furthestallowedextremety; //this field is the combined max-min square, added together
//note that while this allows you to move models about a little, you cannot resize the visible part
} clampedmodel_t;
//these should be rounded up slightly.
//really this is only to catch spiked models. This doesn't prevent more visible models, just bigger ones.
clampedmodel_t clampedmodel[] = {
{"maps/b_bh100.bsp", 3440},
{"progs/player.mdl", 22497},
{"progs/eyes.mdl", 755},
{"progs/gib1.mdl", 374},
{"progs/gib2.mdl", 1779},
{"progs/gib3.mdl", 2066},
{"progs/bolt2.mdl", 1160},
{"progs/end1.mdl", 764},
{"progs/end2.mdl", 981},
{"progs/end3.mdl", 851},
{"progs/end4.mdl", 903},
{"progs/g_shot.mdl", 3444},
{"progs/g_nail.mdl", 2234},
{"progs/g_nail2.mdl", 3660},
{"progs/g_rock.mdl", 3441},
{"progs/g_rock2.mdl", 3660},
{"progs/g_light.mdl", 2698},
{"progs/invisibl.mdl", 196},
{"progs/quaddama.mdl", 2353},
{"progs/invulner.mdl", 2746},
{"progs/suit.mdl", 3057},
{"progs/missile.mdl", 416},
{"progs/grenade.mdl", 473},
{"progs/spike.mdl", 112},
{"progs/s_spike.mdl", 112},
{"progs/backpack.mdl", 1117},
{"progs/armor.mdl", 2919},
{"progs/s_bubble.spr", 100},
{"progs/s_explod.spr", 1000},
//and now TF models
#ifdef warningmsg
#pragma warningmsg("FIXME: these are placeholders")
#endif
{"progs/disp.mdl", 3000},
{"progs/tf_flag.mdl", 3000},
{"progs/tf_stan.mdl", 3000},
{"progs/turrbase.mdl", 3000},
{"progs/turrgun.mdl", 3000}
};
void Mod_AccumulateTextureVectors(vecV_t *vc, vec2_t *tc, vec3_t *nv, vec3_t *sv, vec3_t *tv, index_t *idx, int numidx)
{
int i;
float *v0, *v1, *v2;
float *tc0, *tc1, *tc2;
vec3_t d1, d2;
float td1, td2;
vec3_t norm, t, s;
vec3_t temp;
for (i = 0; i < numidx; i += 3)
{
//this is the stuff we're working from
v0 = vc[idx[i+0]];
v1 = vc[idx[i+1]];
v2 = vc[idx[i+2]];
tc0 = tc[idx[i+0]];
tc1 = tc[idx[i+1]];
tc2 = tc[idx[i+2]];
//calc perpendicular directions
VectorSubtract(v1, v0, d1);
VectorSubtract(v2, v0, d2);
//calculate s as the pependicular of the t dir
td1 = tc1[1] - tc0[1];
td2 = tc2[1] - tc0[1];
s[0] = td1 * d2[0] - td2 * d1[0];
s[1] = td1 * d2[1] - td2 * d1[1];
s[2] = td1 * d2[2] - td2 * d1[2];
//calculate t as the pependicular of the s dir
td1 = tc1[0] - tc0[0];
td2 = tc2[0] - tc0[0];
t[0] = td1 * d2[0] - td2 * d1[0];
t[1] = td1 * d2[1] - td2 * d1[1];
t[2] = td1 * d2[2] - td2 * d1[2];
//the surface might be a back face and thus textured backwards
//calc the normal twice and compare.
norm[0] = d2[1] * d1[2] - d2[2] * d1[1];
norm[1] = d2[2] * d1[0] - d2[0] * d1[2];
norm[2] = d2[0] * d1[1] - d2[1] * d1[0];
CrossProduct(t, s, temp);
if (DotProduct(temp, norm) < 0)
{
VectorNegate(s, s);
VectorNegate(t, t);
}
//and we're done, accumulate the result
VectorAdd(sv[idx[i+0]], s, sv[idx[i+0]]);
VectorAdd(sv[idx[i+1]], s, sv[idx[i+1]]);
VectorAdd(sv[idx[i+2]], s, sv[idx[i+2]]);
VectorAdd(tv[idx[i+0]], t, tv[idx[i+0]]);
VectorAdd(tv[idx[i+1]], t, tv[idx[i+1]]);
VectorAdd(tv[idx[i+2]], t, tv[idx[i+2]]);
}
}
void Mod_AccumulateMeshTextureVectors(mesh_t *m)
{
Mod_AccumulateTextureVectors(m->xyz_array, m->st_array, m->normals_array, m->snormals_array, m->tnormals_array, m->indexes, m->numindexes);
}
void Mod_NormaliseTextureVectors(vec3_t *n, vec3_t *s, vec3_t *t, int v)
{
int i;
float f;
vec3_t tmp;
for (i = 0; i < v; i++)
{
f = -DotProduct(s[i], n[i]);
VectorMA(s[i], f, n[i], tmp);
VectorNormalize2(tmp, s[i]);
f = -DotProduct(t[i], n[i]);
VectorMA(t[i], f, n[i], tmp);
VectorNormalize2(tmp, t[i]);
}
}
#ifdef SKELETALMODELS
static void GenMatrixPosQuat4Scale(vec3_t pos, vec4_t quat, vec3_t scale, float result[12])
{
float xx, xy, xz, xw, yy, yz, yw, zz, zw;
float x2, y2, z2;
float s;
x2 = quat[0] + quat[0];
y2 = quat[1] + quat[1];
z2 = quat[2] + quat[2];
xx = quat[0] * x2; xy = quat[0] * y2; xz = quat[0] * z2;
yy = quat[1] * y2; yz = quat[1] * z2; zz = quat[2] * z2;
xw = quat[3] * x2; yw = quat[3] * y2; zw = quat[3] * z2;
s = scale[0];
result[0*4+0] = s*(1.0f - (yy + zz));
result[1*4+0] = s*(xy + zw);
result[2*4+0] = s*(xz - yw);
s = scale[1];
result[0*4+1] = s*(xy - zw);
result[1*4+1] = s*(1.0f - (xx + zz));
result[2*4+1] = s*(yz + xw);
s = scale[2];
result[0*4+2] = s*(xz + yw);
result[1*4+2] = s*(yz - xw);
result[2*4+2] = s*(1.0f - (xx + yy));
result[0*4+3] = pos[0];
result[1*4+3] = pos[1];
result[2*4+3] = pos[2];
}
/*like above, but guess the quat.w*/
static void GenMatrixPosQuat3Scale(vec3_t pos, vec3_t quat3, vec3_t scale, float result[12])
{
vec4_t quat4;
float term = 1 - DotProduct(quat3, quat3);
if (term < 0)
quat4[3] = 0;
else
quat4[3] = - (float) sqrt(term);
VectorCopy(quat3, quat4);
GenMatrixPosQuat4Scale(pos, quat4, scale, result);
}
static void GenMatrix(float x, float y, float z, float qx, float qy, float qz, float result[12])
{
float qw;
{ //figure out qw
float term = 1 - (qx*qx) - (qy*qy) - (qz*qz);
if (term < 0)
qw = 0;
else
qw = - (float) sqrt(term);
}
{ //generate the matrix
/*
float xx = qx * qx;
float xy = qx * qy;
float xz = qx * qz;
float xw = qx * qw;
float yy = qy * qy;
float yz = qy * qz;
float yw = qy * qw;
float zz = qz * qz;
float zw = qz * qw;
result[0*4+0] = 1 - 2 * ( yy + zz );
result[0*4+1] = 2 * ( xy - zw );
result[0*4+2] = 2 * ( xz + yw );
result[0*4+3] = x;
result[1*4+0] = 2 * ( xy + zw );
result[1*4+1] = 1 - 2 * ( xx + zz );
result[1*4+2] = 2 * ( yz - xw );
result[1*4+3] = y;
result[2*4+0] = 2 * ( xz - yw );
result[2*4+1] = 2 * ( yz + xw );
result[2*4+2] = 1 - 2 * ( xx + yy );
result[2*4+3] = z;
*/
float xx, xy, xz, xw, yy, yz, yw, zz, zw;
float x2, y2, z2;
x2 = qx + qx;
y2 = qy + qy;
z2 = qz + qz;
xx = qx * x2; xy = qx * y2; xz = qx * z2;
yy = qy * y2; yz = qy * z2; zz = qz * z2;
xw = qw * x2; yw = qw * y2; zw = qw * z2;
result[0*4+0] = 1.0f - (yy + zz);
result[1*4+0] = xy + zw;
result[2*4+0] = xz - yw;
result[0*4+1] = xy - zw;
result[1*4+1] = 1.0f - (xx + zz);
result[2*4+1] = yz + xw;
result[0*4+2] = xz + yw;
result[1*4+2] = yz - xw;
result[2*4+2] = 1.0f - (xx + yy);
result[0*4+3] = x;
result[1*4+3] = y;
result[2*4+3] = z;
}
}
static void PSKGenMatrix(float x, float y, float z, float qx, float qy, float qz, float qw, float result[12])
{
float xx, xy, xz, xw, yy, yz, yw, zz, zw;
float x2, y2, z2;
x2 = qx + qx;
y2 = qy + qy;
z2 = qz + qz;
xx = qx * x2; xy = qx * y2; xz = qx * z2;
yy = qy * y2; yz = qy * z2; zz = qz * z2;
xw = qw * x2; yw = qw * y2; zw = qw * z2;
result[0*4+0] = 1.0f - (yy + zz);
result[1*4+0] = xy + zw;
result[2*4+0] = xz - yw;
result[0*4+1] = xy - zw;
result[1*4+1] = 1.0f - (xx + zz);
result[2*4+1] = yz + xw;
result[0*4+2] = xz + yw;
result[1*4+2] = yz - xw;
result[2*4+2] = 1.0f - (xx + yy);
result[0*4+3] = x;
result[1*4+3] = y;
result[2*4+3] = z;
}
#if 0
/*transforms some skeletal vecV_t values*/
static void Alias_TransformVerticies_V(float *bonepose, int vertcount, qbyte *bidx, float *weights, float *xyzin, float *fte_restrict xyzout)
{
int i;
float *matrix;
for (i = 0; i < vertcount; i++, xyzout+=sizeof(vecV_t)/sizeof(vec_t), xyzin+=sizeof(vecV_t)/sizeof(vec_t), bidx+=4, weights+=4)
{
matrix = &bonepose[12*bidx[0]];
xyzout[0] = weights[0] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] = weights[0] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] = weights[0] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[1] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[1]];
xyzout[0] += weights[1] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[1] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[1] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[2] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[2]];
xyzout[0] += weights[2] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[2] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[2] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[3] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[3]];
xyzout[0] += weights[3] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[3] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[3] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
}
}
}
}
}
#endif
/*transforms some skeletal vecV_t values*/
static void Alias_TransformVerticies_VN(float *bonepose, int vertcount, qbyte *bidx, float *weights,
float *xyzin, float *fte_restrict xyzout,
float *normin, float *fte_restrict normout)
{
int i, j;
float *matrix;
float mat[12];
for (i = 0; i < vertcount; i++,
xyzout+=sizeof(vecV_t)/sizeof(vec_t), xyzin+=sizeof(vecV_t)/sizeof(vec_t),
normout+=sizeof(vec3_t)/sizeof(vec_t), normin+=sizeof(vec3_t)/sizeof(vec_t),
bidx+=4, weights+=4)
{
matrix = &bonepose[12*bidx[0]];
for (j = 0; j < 12; j++)
mat[j] = weights[0] * matrix[j];
if (weights[1])
{
matrix = &bonepose[12*bidx[1]];
for (j = 0; j < 12; j++)
mat[j] += weights[1] * matrix[j];
if (weights[2])
{
matrix = &bonepose[12*bidx[2]];
for (j = 0; j < 12; j++)
mat[j] += weights[2] * matrix[j];
if (weights[3])
{
matrix = &bonepose[12*bidx[3]];
for (j = 0; j < 12; j++)
mat[j] += weights[3] * matrix[j];
}
}
}
matrix = mat;
xyzout[0] = (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + matrix[ 3]);
xyzout[1] = (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + matrix[ 7]);
xyzout[2] = (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + matrix[11]);
normout[0] = (normin[0] * matrix[0] + normin[1] * matrix[1] + normin[2] * matrix[ 2]);
normout[1] = (normin[0] * matrix[4] + normin[1] * matrix[5] + normin[2] * matrix[ 6]);
normout[2] = (normin[0] * matrix[8] + normin[1] * matrix[9] + normin[2] * matrix[10]);
}
}
#if 0
/*transforms some skeletal vec3_t values*/
static void Alias_TransformVerticies_3(float *fte_restrict bonepose, int vertcount, qbyte *bidx, float *weights, float *xyzin, float *fte_restrict xyzout)
{
int i;
float *matrix;
for (i = 0; i < vertcount; i++, xyzout+=sizeof(vec3_t)/sizeof(vec_t), xyzin+=sizeof(vec3_t)/sizeof(vec_t), bidx+=4, weights+=4)
{
matrix = &bonepose[12*bidx[0]];
xyzout[0] = weights[0] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] = weights[0] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] = weights[0] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[1] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[1]];
xyzout[0] += weights[1] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[1] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[1] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[2] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[2]];
xyzout[0] += weights[2] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[2] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[2] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
if (bidx[3] != ~(qbyte)0)
{
matrix = &bonepose[12*bidx[3]];
xyzout[0] += weights[3] * (xyzin[0] * matrix[0] + xyzin[1] * matrix[1] + xyzin[2] * matrix[ 2] + xyzin[3] * matrix[ 3]);
xyzout[1] += weights[3] * (xyzin[0] * matrix[4] + xyzin[1] * matrix[5] + xyzin[2] * matrix[ 6] + xyzin[3] * matrix[ 7]);
xyzout[2] += weights[3] * (xyzin[0] * matrix[8] + xyzin[1] * matrix[9] + xyzin[2] * matrix[10] + xyzin[3] * matrix[11]);
}
}
}
}
}
#endif
static void Alias_TransformVerticies_SW(float *bonepose, galisskeletaltransforms_t *weights, int numweights, vecV_t *xyzout, vec3_t *normout)
{
int i;
float *out, *matrix;
galisskeletaltransforms_t *v = weights;
#ifndef SERVERONLY
float *normo;
if (normout)
{
for (i = 0;i < numweights;i++, v++)
{
out = xyzout[v->vertexindex];
normo = normout[v->vertexindex];
matrix = bonepose+v->boneindex*12;
// FIXME: this can very easily be optimized with SSE or 3DNow
out[0] += v->org[0] * matrix[0] + v->org[1] * matrix[1] + v->org[2] * matrix[ 2] + v->org[3] * matrix[ 3];
out[1] += v->org[0] * matrix[4] + v->org[1] * matrix[5] + v->org[2] * matrix[ 6] + v->org[3] * matrix[ 7];
out[2] += v->org[0] * matrix[8] + v->org[1] * matrix[9] + v->org[2] * matrix[10] + v->org[3] * matrix[11];
normo[0] += v->normal[0] * matrix[0] + v->normal[1] * matrix[1] + v->normal[2] * matrix[ 2];
normo[1] += v->normal[0] * matrix[4] + v->normal[1] * matrix[5] + v->normal[2] * matrix[ 6];
normo[2] += v->normal[0] * matrix[8] + v->normal[1] * matrix[9] + v->normal[2] * matrix[10];
}
}
else
#elif defined(_DEBUG)
if (normout)
Sys_Error("norms error");
#endif
{
for (i = 0;i < numweights;i++, v++)
{
out = xyzout[v->vertexindex];
matrix = bonepose+v->boneindex*12;
// FIXME: this can very easily be optimized with SSE or 3DNow
out[0] += v->org[0] * matrix[0] + v->org[1] * matrix[1] + v->org[2] * matrix[ 2] + v->org[3] * matrix[ 3];
out[1] += v->org[0] * matrix[4] + v->org[1] * matrix[5] + v->org[2] * matrix[ 6] + v->org[3] * matrix[ 7];
out[2] += v->org[0] * matrix[8] + v->org[1] * matrix[9] + v->org[2] * matrix[10] + v->org[3] * matrix[11];
}
}
}
static float Alias_CalculateSkeletalNormals(galiasinfo_t *model)
{
#ifndef SERVERONLY
//servers don't need normals. except maybe for tracing... but hey. The normal is calculated on a per-triangle basis.
#define TriangleNormal(a,b,c,n) ( \
(n)[0] = ((a)[1] - (b)[1]) * ((c)[2] - (b)[2]) - ((a)[2] - (b)[2]) * ((c)[1] - (b)[1]), \
(n)[1] = ((a)[2] - (b)[2]) * ((c)[0] - (b)[0]) - ((a)[0] - (b)[0]) * ((c)[2] - (b)[2]), \
(n)[2] = ((a)[0] - (b)[0]) * ((c)[1] - (b)[1]) - ((a)[1] - (b)[1]) * ((c)[0] - (b)[0]) \
)
int i, j;
vecV_t *xyz;
vec3_t *normals;
int *mvert;
float *inversepose;
galiasinfo_t *next;
vec3_t tn;
vec3_t d1, d2;
index_t *idx;
float *bonepose = NULL;
float angle;
float maxvdist = 0, d, maxbdist = 0;
float absmatrix[MAX_BONES*12];
float bonedist[MAX_BONES];
int modnum = 0;
int bcmodnum = -1;
int vcmodnum = -1;
while (model)
{
int numbones = model->numbones;
galisskeletaltransforms_t *v = model->ofsswtransforms;
int numweights = model->numswtransforms;
int numverts = model->numverts;
next = model->nextsurf;
xyz = Z_Malloc(numverts*sizeof(vecV_t));
normals = Z_Malloc(numverts*sizeof(vec3_t));
inversepose = Z_Malloc(numbones*sizeof(float)*9);
mvert = Z_Malloc(numverts*sizeof(*mvert));
if (bcmodnum != model->shares_bones)
{
galiasgroup_t *g;
galiasbone_t *bones = model->ofsbones;
bcmodnum = model->shares_bones;
if (model->baseframeofs)
bonepose = model->baseframeofs;
else
{
//figure out the pose from frame0pose0
if (!model->groups)
return 0;
g = model->groupofs;
if (g->numposes < 1)
return 0;
bonepose = g->boneofs;
if (g->isheirachical)
{
/*needs to be an absolute skeleton*/
for (i = 0; i < model->numbones; i++)
{
if (bones[i].parent >= 0)
R_ConcatTransforms((void*)(absmatrix + bones[i].parent*12), (void*)(bonepose+i*12), (void*)(absmatrix+i*12));
else
for (j = 0;j < 12;j++) //parentless
absmatrix[i*12+j] = (bonepose)[i*12+j];
}
bonepose = absmatrix;
}
}
/*calculate the bone sizes (assuming the bones are strung up and hanging or such)*/
for (i = 0; i < model->numbones; i++)
{
vec3_t d;
float *b;
b = bonepose + i*12;
d[0] = b[3];
d[1] = b[7];
d[2] = b[11];
if (bones[i].parent >= 0)
{
b = bonepose + bones[i].parent*12;
d[0] -= b[3];
d[1] -= b[7];
d[2] -= b[11];
}
bonedist[i] = Length(d);
if (bones[i].parent >= 0)
bonedist[i] += bonedist[bones[i].parent];
if (maxbdist < bonedist[i])
maxbdist = bonedist[i];
}
for (i = 0; i < numbones; i++)
Matrix3x4_InvertTo3x3(bonepose+i*12, inversepose+i*9);
}
for (i = 0; i < numweights; i++)
{
d = Length(v[i].org);
if (maxvdist < d)
maxvdist = d;
}
//build the actual base pose positions
Alias_TransformVerticies_SW(bonepose, v, numweights, xyz, NULL);
//work out which verticies are identical
//this is needed as two verts can have same origin but different tex coords
//without this, we end up with a seam that splits the normals each side on arms, etc
for (i = 0; i < numverts; i++)
{
mvert[i] = i;
for (j = 0; j < i; j++)
{
if ( xyz[i][0] == xyz[j][0]
&& xyz[i][1] == xyz[j][1]
&& xyz[i][2] == xyz[j][2])
{
mvert[i] = j;
break;
}
}
}
//use that base pose to calculate the normals
memset(normals, 0, numverts*sizeof(vec3_t));
vcmodnum = modnum;
idx = model->ofs_indexes;
//calculate the triangle normal and accumulate them
for (i = 0; i < model->numindexes; i+=3, idx+=3)
{
TriangleNormal(xyz[idx[0]], xyz[idx[1]], xyz[idx[2]], tn);
//note that tn is relative to the size of the triangle
//Imagine a cube, each side made of two triangles
VectorSubtract(xyz[idx[1]], xyz[idx[0]], d1);
VectorSubtract(xyz[idx[2]], xyz[idx[0]], d2);
angle = acos(DotProduct(d1, d2)/(Length(d1)*Length(d2)));
VectorMA(normals[mvert[idx[0]]], angle, tn, normals[mvert[idx[0]]]);
VectorSubtract(xyz[idx[0]], xyz[idx[1]], d1);
VectorSubtract(xyz[idx[2]], xyz[idx[1]], d2);
angle = acos(DotProduct(d1, d2)/(Length(d1)*Length(d2)));
VectorMA(normals[mvert[idx[1]]], angle, tn, normals[mvert[idx[1]]]);
VectorSubtract(xyz[idx[0]], xyz[idx[2]], d1);
VectorSubtract(xyz[idx[1]], xyz[idx[2]], d2);
angle = acos(DotProduct(d1, d2)/(Length(d1)*Length(d2)));
VectorMA(normals[mvert[idx[2]]], angle, tn, normals[mvert[idx[2]]]);
}
/*skip over each additional surface that shares the same verts*/
for(;;)
{
if (next && next->shares_verts == vcmodnum)
{
modnum++;
model = next;
next = model->nextsurf;
}
else
break;
}
//the normals are not normalized yet.
for (i = 0; i < numverts; i++)
{
VectorNormalize(normals[i]);
}
for (i = 0; i < numweights; i++, v++)
{
v->normal[0] = DotProduct(normals[mvert[v->vertexindex]], inversepose+9*v->boneindex+0) * v->org[3];
v->normal[1] = DotProduct(normals[mvert[v->vertexindex]], inversepose+9*v->boneindex+3) * v->org[3];
v->normal[2] = DotProduct(normals[mvert[v->vertexindex]], inversepose+9*v->boneindex+6) * v->org[3];
}
if (model->ofs_skel_norm)
memcpy(model->ofs_skel_norm, normals, numverts*sizeof(vec3_t));
//FIXME: save off the xyz+normals for this base pose as an optimisation for world objects.
Z_Free(inversepose);
Z_Free(normals);
Z_Free(xyz);
Z_Free(mvert);
model = next;
modnum++;
}
return maxvdist+maxbdist;
#else
return 0;
#endif
}
static int Alias_BuildLerps(float plerp[4], float *pose[4], int numbones, galiasgroup_t *g1, galiasgroup_t *g2, float lerpfrac, float fg1time, float fg2time)
{
int frame1;
int frame2;
float mlerp; //minor lerp, poses within a group.
int l = 0;
if (g1 == g2)
lerpfrac = 0;
if (fg1time < 0)
fg1time = 0;
mlerp = (fg1time)*g1->rate;
frame1=mlerp;
frame2=frame1+1;
mlerp-=frame1;
if (g1->loop)
{
frame1=frame1%g1->numposes;
frame2=frame2%g1->numposes;
}
else
{
frame1=(frame1>g1->numposes-1)?g1->numposes-1:frame1;
frame2=(frame2>g1->numposes-1)?g1->numposes-1:frame2;
}
if (frame1 == frame2 || r_noframegrouplerp.ival)
mlerp = 0;
plerp[l] = (1-mlerp)*(1-lerpfrac);
if (plerp[l]>0)
pose[l++] = g1->boneofs + numbones*12*frame1;
plerp[l] = (mlerp)*(1-lerpfrac);
if (plerp[l]>0)
pose[l++] = g1->boneofs + numbones*12*frame2;
if (lerpfrac)
{
if (fg2time < 0)
fg2time = 0;
mlerp = (fg2time)*g2->rate;
frame1=mlerp;
frame2=frame1+1;
mlerp-=frame1;
if (g2->loop)
{
frame1=frame1%g2->numposes;
frame2=frame2%g2->numposes;
}
else
{
frame1=(frame1>g2->numposes-1)?g2->numposes-1:frame1;
frame2=(frame2>g2->numposes-1)?g2->numposes-1:frame2;
}
if (frame1 == frame2 || r_noframegrouplerp.ival)
mlerp = 0;
plerp[l] = (1-mlerp)*(lerpfrac);
if (plerp[l]>0)
pose[l++] = g2->boneofs + numbones*12*frame1;
plerp[l] = (mlerp)*(lerpfrac);
if (plerp[l]>0)
pose[l++] = g2->boneofs + numbones*12*frame2;
}
return l;
}
//ignores any skeletal objects
int Alias_GetBoneRelations(galiasinfo_t *inf, framestate_t *fstate, float *result, int firstbone, int lastbones)
{
#ifdef SKELETALMODELS
if (inf->numbones)
{
galiasgroup_t *g1, *g2;
float *matrix; //the matrix for a single bone in a single pose.
int b, k; //counters
float *pose[4]; //the per-bone matricies (one for each pose)
float plerp[4]; //the ammount of that pose to use (must combine to 1)
int numposes = 0;
int frame1, frame2;
float f1time, f2time;
float f2ness;
int bonegroup;
int cbone = 0;
int endbone;
if (lastbones > inf->numbones)
lastbones = inf->numbones;
if (!lastbones)
return 0;
for (bonegroup = 0; bonegroup < FS_COUNT; bonegroup++)
{
endbone = fstate->g[bonegroup].endbone;
if (bonegroup == FS_COUNT-1 || endbone > lastbones)
endbone = lastbones;
if (endbone == cbone)
continue;
frame1 = fstate->g[bonegroup].frame[0];
frame2 = fstate->g[bonegroup].frame[1];
f1time = fstate->g[bonegroup].frametime[0];
f2time = fstate->g[bonegroup].frametime[1];
f2ness = fstate->g[bonegroup].lerpfrac;
//FIXME: fixup these framestates earlier, because this just isn't nice
if (frame1 < 0 || frame1 >= inf->groups)
{
if (frame2 < 0 || frame2 >= inf->groups || f2ness == 0)
{
if (bonegroup != FS_COUNT-1)
continue; //just ignore this group
//there's no escaping it, both are bad. use the base pose
f2ness = 0;
frame1 = frame2 = 0;
}
else
{
//kill it, just use frame2
f2ness = 1;
frame1 = frame2;
}
}
else
{
if (frame2 < 0 || frame2 >= inf->groups)
{
//kill this anim
f2ness = 0;
frame2 = frame1;
}
}
//the higher level merges old/new anims, but we still need to blend between automated frame-groups.
g1 = &inf->groupofs[frame1];
g2 = &inf->groupofs[frame2];
if (!g1->isheirachical)
return 0;
if (!g2->isheirachical)
g2 = g1;
numposes = Alias_BuildLerps(plerp, pose, inf->numbones, g1, g2, f2ness, f1time, f2time);
if (numposes == 1)
{
memcpy(result, pose[0]+cbone*12, (lastbones-cbone)*12*sizeof(float));
result += (lastbones-cbone)*12;
cbone = lastbones;
}
else
{
//set up the identity matrix
for (; cbone < lastbones; cbone++)
{
//set up the per-bone transform matrix
for (k = 0;k < 12;k++)
result[k] = 0;
for (b = 0;b < numposes;b++)
{
matrix = pose[b] + cbone*12;
for (k = 0;k < 12;k++)
result[k] += matrix[k] * plerp[b];
}
result += 12;
}
}
}
return cbone;
}
#endif
return 0;
}
//_may_ write into bonepose, return value is the real result. obtains absolute values
float *Alias_GetBonePositions(galiasinfo_t *inf, framestate_t *fstate, float *buffer, int buffersize, qboolean renderable)
{
#ifdef SKELETALMODELS
float relationsbuf[MAX_BONES][12];
float *relations = NULL;
galiasbone_t *bones = inf->ofsbones;
int numbones;
if (buffersize < inf->numbones)
numbones = 0;
else if (fstate->bonestate && fstate->bonecount >= inf->numbones)
{
relations = fstate->bonestate;
numbones = inf->numbones;
if (fstate->boneabs)
{
/*we may need to invert by the inverse of the base pose to get the bones into the proper positions*/
if (!inf->numswtransforms && renderable)
{
int i;
for (i = 0; i < inf->numbones; i++)
{
R_ConcatTransforms((void*)(relations + i*12), (void*)(bones[i].inverse), (void*)(buffer + i*12));
}
return buffer;
}
return relations;
}
}
else
{
numbones = Alias_GetBoneRelations(inf, fstate, (float*)relationsbuf, 0, inf->numbones);
if (numbones == inf->numbones)
relations = (float*)relationsbuf;
}
if (relations)
{
int i, k;
if (!inf->numswtransforms && renderable)
{
float absbuf[MAX_BONES][12];
for (i = 0; i < numbones; i++)
{
if (bones[i].parent >= 0)
R_ConcatTransforms((void*)(absbuf[bones[i].parent]), (void*)((float*)relations+i*12), (void*)absbuf[i]);
else
for (k = 0;k < 12;k++) //parentless
absbuf[i][k] = ((float*)relations)[i*12+k];
R_ConcatTransforms((void*)absbuf[i], (void*)bones[i].inverse, (void*)(buffer+i*12));
}
}
else
{
for (i = 0; i < numbones; i++)
{
if (bones[i].parent >= 0)
R_ConcatTransforms((void*)(buffer + bones[i].parent*12), (void*)((float*)relations+i*12), (void*)(buffer+i*12));
else
for (k = 0;k < 12;k++) //parentless
buffer[i*12+k] = ((float*)relations)[i*12+k];
}
}
return buffer;
}
else
{
int i, k;
int l=0;
float plerp[4];
float *pose[4];
int numposes;
int f;
float lerpfrac = fstate->g[FS_REG].lerpfrac;
galiasgroup_t *g1, *g2;
//galiasbone_t *bones = (galiasbone_t *)((char*)inf+inf->ofsbones); //unsed variable
if (buffersize < inf->numbones)
return NULL;
f = fstate->g[FS_REG].frame[0];
if (f < 0 || f >= inf->groups)
f = 0;
g1 = &inf->groupofs[bound(0, f, inf->groups-1)];
f = fstate->g[FS_REG].frame[1];
if (f < 0 || f >= inf->groups)
g2 = g1;
else
g2 = &inf->groupofs[bound(0, f, inf->groups-1)];
if (g2->isheirachical)
g2 = g1;
numposes = Alias_BuildLerps(plerp, pose, inf->numbones, g1, g2, lerpfrac, fstate->g[FS_REG].frametime[0], fstate->g[FS_REG].frametime[1]);
{
//this is not hierachal, using base frames is not a good idea.
//just blend the poses here
if (numposes == 1)
return pose[0];
else if (numposes == 2)
{
for (i = 0; i < inf->numbones*12; i++)
{
((float*)buffer)[i] = pose[0][i]*plerp[0] + pose[1][i]*plerp[1];
}
}
else
{
for (i = 0; i < inf->numbones; i++)
{
for (l = 0; l < 12; l++)
buffer[i*12+l] = 0;
for (k = 0; k < numposes; k++)
{
for (l = 0; l < 12; l++)
buffer[i*12+l] += pose[k][i*12+l] * plerp[k];
}
}
}
}
return buffer;
}
#endif
return NULL;
}
static void R_LerpBones(float *plerp, float **pose, int poses, galiasbone_t *bones, int bonecount, float bonepose[MAX_BONES][12])
{
int i, k, b;
float *matrix, m[12];
if (poses == 1)
{
// vertex weighted skeletal
// interpolate matrices and concatenate them to their parents
for (i = 0;i < bonecount;i++)
{
matrix = pose[0] + i*12;
if (bones[i].parent >= 0)
R_ConcatTransforms((void*)bonepose[bones[i].parent], (void*)matrix, (void*)bonepose[i]);
else
for (k = 0;k < 12;k++) //parentless
bonepose[i][k] = matrix[k];
}
}
else
{
// vertex weighted skeletal
// interpolate matrices and concatenate them to their parents
for (i = 0;i < bonecount;i++)
{
for (k = 0;k < 12;k++)
m[k] = 0;
for (b = 0;b < poses;b++)
{
matrix = pose[b] + i*12;
for (k = 0;k < 12;k++)
m[k] += matrix[k] * plerp[b];
}
if (bones[i].parent >= 0)
R_ConcatTransforms((void*)bonepose[bones[i].parent], (void*)m, (void*)bonepose[i]);
else
for (k = 0;k < 12;k++) //parentless
bonepose[i][k] = m[k];
}
}
}
#endif
#if defined(D3DQUAKE) || defined(GLQUAKE)
struct
{
int numcolours;
avec4_t *colours;
int numcoords;
vecV_t *coords;
int numnorm;
vec3_t *norm;
int surfnum;
entity_t *ent;
#ifdef SKELETALMODELS
float bonepose[MAX_BONES*12];
float *usebonepose;
int bonecount;
#endif
vecV_t *acoords1;
vecV_t *acoords2;
vec3_t *anorm;
vec3_t *anorms;
vec3_t *anormt;
vbo_t vbo;
vbo_t *vbop;
} meshcache;
//#define SSE_INTRINSICS
#ifdef SSE_INTRINSICS
#include <xmmintrin.h>
#endif
void R_LightArraysByte_BGR(const entity_t *entity, vecV_t *coords, byte_vec4_t *colours, int vertcount, vec3_t *normals)
{
int i;
int c;
float l;
byte_vec4_t ambientlightb;
byte_vec4_t shadelightb;
const float *lightdir = entity->light_dir;
for (i = 0; i < 3; i++)
{
l = entity->light_avg[2-i]*255;
ambientlightb[i] = bound(0, l, 255);
l = entity->light_range[2-i]*255;
shadelightb[i] = bound(0, l, 255);
}
if (ambientlightb[0] == shadelightb[0] && ambientlightb[1] == shadelightb[1] && ambientlightb[2] == shadelightb[2])
{
for (i = vertcount-1; i >= 0; i--)
{
*(int*)colours[i] = *(int*)ambientlightb;
// colours[i][0] = ambientlightb[0];
// colours[i][1] = ambientlightb[1];
// colours[i][2] = ambientlightb[2];
}
}
else
{
for (i = vertcount-1; i >= 0; i--)
{
l = DotProduct(normals[i], lightdir);
c = l*shadelightb[0];
c += ambientlightb[0];
colours[i][0] = bound(0, c, 255);
c = l*shadelightb[1];
c += ambientlightb[1];
colours[i][1] = bound(0, c, 255);
c = l*shadelightb[2];
c += ambientlightb[2];
colours[i][2] = bound(0, c, 255);
}
}
}
void R_LightArrays(const entity_t *entity, vecV_t *coords, avec4_t *colours, int vertcount, vec3_t *normals, float scale)
{
extern cvar_t r_vertexdlights;
int i;
float l;
//float *lightdir = currententity->light_dir; //unused variable
if (!entity->light_range[0] && !entity->light_range[1] && !entity->light_range[2])
{
for (i = vertcount-1; i >= 0; i--)
{
colours[i][0] = entity->light_avg[0];
colours[i][1] = entity->light_avg[1];
colours[i][2] = entity->light_avg[2];
}
}
else
{
vec3_t la, lr;
VectorScale(entity->light_avg, scale, la);
VectorScale(entity->light_range, scale, lr);
#ifdef SSE_INTRINSICS
__m128 va, vs, vl, vr;
va = _mm_load_ps(ambientlight);
vs = _mm_load_ps(shadelight);
va.m128_f32[3] = 0;
vs.m128_f32[3] = 1;
#endif
/*dotproduct will return a value between 1 and -1, so increase the ambient to be correct for normals facing away from the light*/
for (i = vertcount-1; i >= 0; i--)
{
l = DotProduct(normals[i], entity->light_dir);
#ifdef SSE_INTRINSICS
vl = _mm_load1_ps(&l);
vr = _mm_mul_ss(va,vl);
vr = _mm_add_ss(vr,vs);
_mm_storeu_ps(colours[i], vr);
//stomp on colour[i][3] (will be set to 1)
#else
colours[i][0] = l*lr[0]+la[0];
colours[i][1] = l*lr[1]+la[1];
colours[i][2] = l*lr[2]+la[2];
#endif
}
}
if (r_vertexdlights.ival && r_dynamic.ival)
{
unsigned int lno, v;
vec3_t dir, rel;
float dot, d, a;
//don't include world lights
for (lno = rtlights_first; lno < RTL_FIRST; lno++)
{
if (cl_dlights[lno].radius)
{
VectorSubtract (cl_dlights[lno].origin,
entity->origin,
dir);
if (Length(dir)>cl_dlights[lno].radius+256) //far out man!
continue;
rel[0] = -DotProduct(dir, entity->axis[0]);
rel[1] = -DotProduct(dir, entity->axis[1]);
rel[2] = -DotProduct(dir, entity->axis[2]);
for (v = 0; v < vertcount; v++)
{
VectorSubtract(coords[v], rel, dir);
dot = DotProduct(dir, normals[v]);
if (dot>0)
{
d = DotProduct(dir, dir);
a = 1/d;
if (a>0)
{
a *= 10000000*dot/sqrt(d);
colours[v][0] += a*cl_dlights[lno].color[0];
colours[v][1] += a*cl_dlights[lno].color[1];
colours[v][2] += a*cl_dlights[lno].color[2];
}
}
}
}
}
}
}
static void R_LerpFrames(mesh_t *mesh, galiaspose_t *p1, galiaspose_t *p2, float lerp, float expand)
{
extern cvar_t r_nolerp; // r_nolightdir is unused
float blerp = 1-lerp;
int i;
vecV_t *p1v, *p2v;
vec3_t *p1n, *p2n;
vec3_t *p1s, *p2s;
vec3_t *p1t, *p2t;
p1v = p1->ofsverts;
p2v = p2->ofsverts;
p1n = p1->ofsnormals;
p2n = p2->ofsnormals;
p1s = p1->ofssvector;
p2s = p2->ofssvector;
p1t = p1->ofstvector;
p2t = p2->ofstvector;
mesh->snormals_array = blerp>0.5?p2s:p1s; //never lerp
mesh->tnormals_array = blerp>0.5?p2t:p1t; //never lerp
mesh->colors4f_array[0] = NULL; //not generated
if (p1v == p2v || r_nolerp.value || !blerp)
{
mesh->normals_array = p1n;
mesh->snormals_array = p1s;
mesh->tnormals_array = p1t;
if (expand)
{
for (i = 0; i < mesh->numvertexes; i++)
{
mesh->xyz_array[i][0] = p1v[i][0] + p1n[i][0]*expand;
mesh->xyz_array[i][1] = p1v[i][1] + p1n[i][1]*expand;
mesh->xyz_array[i][2] = p1v[i][2] + p1n[i][2]*expand;
}
return;
}
else
mesh->xyz_array = p1v;
}
else
{
for (i = 0; i < mesh->numvertexes; i++)
{
mesh->normals_array[i][0] = p1n[i][0]*lerp + p2n[i][0]*blerp;
mesh->normals_array[i][1] = p1n[i][1]*lerp + p2n[i][1]*blerp;
mesh->normals_array[i][2] = p1n[i][2]*lerp + p2n[i][2]*blerp;
mesh->xyz_array[i][0] = p1v[i][0]*lerp + p2v[i][0]*blerp;
mesh->xyz_array[i][1] = p1v[i][1]*lerp + p2v[i][1]*blerp;
mesh->xyz_array[i][2] = p1v[i][2]*lerp + p2v[i][2]*blerp;
}
if (expand)
{
for (i = 0; i < mesh->numvertexes; i++)
{
mesh->xyz_array[i][0] += mesh->normals_array[i][0]*expand;
mesh->xyz_array[i][1] += mesh->normals_array[i][1]*expand;
mesh->xyz_array[i][2] += mesh->normals_array[i][2]*expand;
}
}
}
}
#ifdef SKELETALMODELS
#ifndef SERVERONLY
static void Alias_BuildSkeletalMesh(mesh_t *mesh, float *bonepose, galiasinfo_t *inf)
{
galisskeletaltransforms_t *weights = inf->ofsswtransforms;
int numweights = inf->numswtransforms;
if (inf->ofs_skel_idx)
{
float *fte_restrict xyzout = mesh->xyz_array[0];
float *fte_restrict normout = mesh->normals_array[0];
qbyte *fte_restrict bidx = inf->ofs_skel_idx[0];
float *fte_restrict xyzin = inf->ofs_skel_xyz[0];
float *fte_restrict normin = inf->ofs_skel_norm[0];
// float *fte_restrict svect = inf->ofs_skel_svect[0];
// float *fte_restrict tvect = inf->ofs_skel_tvect[0];
float *fte_restrict weight = inf->ofs_skel_weight[0];
Alias_TransformVerticies_VN(bonepose, inf->numverts, bidx, weight, xyzin, xyzout, normin, normout);
// Alias_TransformVerticies_3(bonepose, inf->numverts, bidx, weight, svect, mesh->snormals_array[0]);
// Alias_TransformVerticies_3(bonepose, inf->numverts, bidx, weight, tvect, mesh->tnormals_array[0]);
}
else
{
memset(mesh->xyz_array, 0, mesh->numvertexes*sizeof(vecV_t));
memset(mesh->normals_array, 0, mesh->numvertexes*sizeof(vec3_t));
Alias_TransformVerticies_SW(bonepose, weights, numweights, mesh->xyz_array, mesh->normals_array);
}
}
#ifdef GLQUAKE
#include "glquake.h"
static void Alias_GLDrawSkeletalBones(galiasbone_t *bones, float *bonepose, int bonecount)
{
PPL_RevertToKnownState();
BE_SelectEntity(currententity);
qglColor3f(1, 0, 0);
{
int i;
int p;
vec3_t org, dest;
qglBegin(GL_LINES);
for (i = 0; i < bonecount; i++)
{
p = bones[i].parent;
if (p < 0)
p = 0;
qglVertex3f(bonepose[i*12+3], bonepose[i*12+7], bonepose[i*12+11]);
qglVertex3f(bonepose[p*12+3], bonepose[p*12+7], bonepose[p*12+11]);
}
qglEnd();
qglColor3f(1, 1, 1);
qglBegin(GL_LINES);
for (i = 0; i < bonecount; i++)
{
p = bones[i].parent;
if (p < 0)
p = 0;
org[0] = bonepose[i*12+3]; org[1] = bonepose[i*12+7]; org[2] = bonepose[i*12+11];
qglVertex3fv(org);
qglVertex3f(bonepose[p*12+3], bonepose[p*12+7], bonepose[p*12+11]);
dest[0] = org[0]+bonepose[i*12+0];dest[1] = org[1]+bonepose[i*12+1];dest[2] = org[2]+bonepose[i*12+2];
qglVertex3fv(org);
qglVertex3fv(dest);
qglVertex3fv(dest);
qglVertex3f(bonepose[p*12+3], bonepose[p*12+7], bonepose[p*12+11]);
dest[0] = org[0]+bonepose[i*12+4];dest[1] = org[1]+bonepose[i*12+5];dest[2] = org[2]+bonepose[i*12+6];
qglVertex3fv(org);
qglVertex3fv(dest);
qglVertex3fv(dest);
qglVertex3f(bonepose[p*12+3], bonepose[p*12+7], bonepose[p*12+11]);
dest[0] = org[0]+bonepose[i*12+8];dest[1] = org[1]+bonepose[i*12+9];dest[2] = org[2]+bonepose[i*12+10];
qglVertex3fv(org);
qglVertex3fv(dest);
qglVertex3fv(dest);
qglVertex3f(bonepose[p*12+3], bonepose[p*12+7], bonepose[p*12+11]);
}
qglEnd();
// mesh->numindexes = 0; //don't draw this mesh, as that would obscure the bones. :(
}
}
#endif //GLQUAKE
#endif //!SERVERONLY
#endif //SKELETALMODELS
void Alias_FlushCache(void)
{
meshcache.ent = NULL;
}
void Alias_Shutdown(void)
{
if (meshcache.colours)
BZ_Free(meshcache.colours);
meshcache.colours = NULL;
meshcache.numcolours = 0;
if (meshcache.norm)
BZ_Free(meshcache.norm);
meshcache.norm = NULL;
meshcache.numnorm = 0;
if (meshcache.coords)
BZ_Free(meshcache.coords);
meshcache.coords = NULL;
meshcache.numcoords = 0;
}
qboolean Alias_GAliasBuildMesh(mesh_t *mesh, vbo_t **vbop, galiasinfo_t *inf, int surfnum, entity_t *e, qboolean usebones)
{
extern cvar_t r_nolerp;
galiasgroup_t *g1, *g2;
int frame1;
int frame2;
float lerp;
float fg1time;
// float fg2time;
if (!inf->groups)
{
#ifdef SKELETALMODELS
if (inf->ofs_skel_xyz && !inf->ofs_skel_weight)
{}
else
#endif
{
Con_DPrintf("Model with no frames (%s)\n", e->model->name);
return false;
}
}
if (meshcache.numcolours < inf->numverts)
{
if (meshcache.colours)
BZ_Free(meshcache.colours);
meshcache.colours = BZ_Malloc(sizeof(*meshcache.colours)*inf->numverts);
meshcache.numcolours = inf->numverts;
}
if (meshcache.numnorm < inf->numverts)
{
if (meshcache.norm)
BZ_Free(meshcache.norm);
meshcache.norm = BZ_Malloc(sizeof(*meshcache.norm)*inf->numverts*3);
meshcache.numnorm = inf->numverts;
}
if (meshcache.numcoords < inf->numverts)
{
if (meshcache.coords)
BZ_Free(meshcache.coords);
meshcache.coords = BZ_Malloc(sizeof(*meshcache.coords)*inf->numverts);
meshcache.numcoords = inf->numverts;
}
mesh->numvertexes = inf->numverts;
mesh->indexes = inf->ofs_indexes;
mesh->numindexes = inf->numindexes;
mesh->st_array = inf->ofs_st_array;
mesh->trneighbors = inf->ofs_trineighbours;
mesh->colors4f_array[0] = meshcache.colours;
if (meshcache.surfnum == inf->shares_verts && meshcache.ent == e)
{
mesh->xyz_array = meshcache.acoords1;
mesh->xyz2_array = meshcache.acoords2;
mesh->normals_array = meshcache.anorm;
mesh->snormals_array = meshcache.anorms;
mesh->tnormals_array = meshcache.anormt;
if (vbop)
*vbop = meshcache.vbop;
#ifdef SKELETALMODELS
if (meshcache.usebonepose)
{
mesh->bonenums = inf->ofs_skel_idx;
mesh->boneweights = inf->ofs_skel_weight;
mesh->bones = meshcache.usebonepose;
mesh->numbones = inf->numbones;
}
#endif
return false; //don't generate the new vertex positions. We still have them all.
}
meshcache.surfnum = inf->shares_verts;
meshcache.ent = e;
#ifndef SERVERONLY
mesh->st_array = inf->ofs_st_array;
mesh->trneighbors = inf->ofs_trineighbours;
mesh->normals_array = meshcache.norm;
mesh->snormals_array = meshcache.norm+meshcache.numnorm;
mesh->tnormals_array = meshcache.norm+meshcache.numnorm*2;
#endif
mesh->xyz_array = meshcache.coords;
//we don't support meshes with one pose skeletal and annother not.
//we don't support meshes with one group skeletal and annother not.
#ifdef SKELETALMODELS
meshcache.usebonepose = NULL;
meshcache.vbop = NULL;
if (vbop)
*vbop = NULL;
if (inf->ofs_skel_xyz && !inf->ofs_skel_weight)
{
//if we have skeletal xyz info, but no skeletal weights, then its a partial model that cannot possibly be animated.
meshcache.usebonepose = NULL;
mesh->xyz_array = inf->ofs_skel_xyz;
mesh->xyz2_array = NULL;
mesh->normals_array = inf->ofs_skel_norm;
mesh->snormals_array = inf->ofs_skel_svect;
mesh->tnormals_array = inf->ofs_skel_tvect;
if (vbop)
{
meshcache.vbo.indicies = inf->vboindicies;
meshcache.vbo.indexcount = inf->numindexes;
meshcache.vbo.vertcount = inf->numverts;
meshcache.vbo.texcoord = inf->vbotexcoords;
meshcache.vbo.coord = inf->vbo_skel_verts;
memset(&meshcache.vbo.coord2, 0, sizeof(meshcache.vbo.coord2));
meshcache.vbo.normals = inf->vbo_skel_normals;
meshcache.vbo.svector = inf->vbo_skel_svector;
meshcache.vbo.tvector = inf->vbo_skel_tvector;
meshcache.vbo.bonenums = inf->vbo_skel_bonenum;
meshcache.vbo.boneweights = inf->vbo_skel_bweight;
if (meshcache.vbo.indicies.dummy)
*vbop = meshcache.vbop = &meshcache.vbo;
}
}
else if (inf->numbones)
{
mesh->xyz2_array = NULL;
meshcache.usebonepose = Alias_GetBonePositions(inf, &e->framestate, meshcache.bonepose, MAX_BONES, true);
if (e->fatness || !inf->ofs_skel_idx || !usebones || inf->numswtransforms)
{
//software bone animation
//there are two ways to animate a skeleton, one is to transform
Alias_BuildSkeletalMesh(mesh, meshcache.usebonepose, inf);
#ifdef PEXT_FATNESS
if (e->fatness)
{
int i;
for (i = 0; i < mesh->numvertexes; i++)
{
VectorMA(mesh->xyz_array[i], e->fatness, mesh->normals_array[i], meshcache.coords[i]);
}
mesh->xyz_array = meshcache.coords;
}
#endif
#ifdef GLQUAKE
if (!inf->numswtransforms && qrenderer == QR_OPENGL)
{
Alias_GLDrawSkeletalBones(inf->ofsbones, (float *)meshcache.usebonepose, inf->numbones);
}
#endif
meshcache.usebonepose = NULL;
}
else
{
//hardware bone animation
mesh->xyz_array = inf->ofs_skel_xyz;
mesh->normals_array = inf->ofs_skel_norm;
mesh->snormals_array = inf->ofs_skel_svect;
mesh->tnormals_array = inf->ofs_skel_tvect;
}
}
else
#endif
{
frame1 = e->framestate.g[FS_REG].frame[0];
frame2 = e->framestate.g[FS_REG].frame[1];
lerp = e->framestate.g[FS_REG].lerpfrac;
fg1time = e->framestate.g[FS_REG].frametime[0];
//fg2time = e->framestate.g[FS_REG].frametime[1];
if (frame1 < 0)
{
Con_DPrintf("Negative frame (%s)\n", e->model->name);
frame1 = 0;
}
if (frame2 < 0)
{
Con_DPrintf("Negative frame (%s)\n", e->model->name);
frame2 = frame1;
}
if (frame1 >= inf->groups)
{
Con_DPrintf("Too high frame %i (%s)\n", frame1, e->model->name);
frame1 %= inf->groups;
}
if (frame2 >= inf->groups)
{
Con_DPrintf("Too high frame %i (%s)\n", frame2, e->model->name);
frame2 = frame1;
}
if (lerp <= 0)
frame2 = frame1;
else if (lerp >= 1)
frame1 = frame2;
g1 = &inf->groupofs[frame1];
g2 = &inf->groupofs[frame2];
if (g1 == g2) //lerping within group is only done if not changing group
{
lerp = fg1time*g1->rate;
if (lerp < 0) lerp = 0; //hrm
frame1=lerp;
frame2=frame1+1;
lerp-=frame1;
if (r_noframegrouplerp.ival)
lerp = 0;
if (g1->loop)
{
frame1=frame1%g1->numposes;
frame2=frame2%g1->numposes;
}
else
{
frame1=(frame1>g1->numposes-1)?g1->numposes-1:frame1;
frame2=(frame2>g1->numposes-1)?g1->numposes-1:frame2;
}
}
else //don't bother with a four way lerp. Yeah, this will produce jerkyness with models with just framegroups.
{
frame1=0;
frame2=0;
}
if (Sh_StencilShadowsActive() || qrenderer != QR_OPENGL)
{
mesh->xyz2_array = NULL;
mesh->xyz_blendw[0] = 1;
mesh->xyz_blendw[1] = 0;
R_LerpFrames(mesh, &g1->poseofs[frame1], &g2->poseofs[frame2], 1-lerp, e->fatness);
}
else
{
galiaspose_t *p1 = &g1->poseofs[frame1];
galiaspose_t *p2 = &g2->poseofs[frame2];
mesh->normals_array = p1->ofsnormals;
mesh->snormals_array = p1->ofssvector;
mesh->tnormals_array = p1->ofstvector;
meshcache.vbo.indicies = inf->vboindicies;
meshcache.vbo.indexcount = inf->numindexes;
meshcache.vbo.vertcount = inf->numverts;
meshcache.vbo.texcoord = inf->vbotexcoords;
meshcache.vbo.normals = p1->vbonormals;
meshcache.vbo.svector = p1->vbosvector;
meshcache.vbo.tvector = p1->vbotvector;
if (p1 == p2 || r_nolerp.ival)
{
meshcache.vbo.coord = p1->vboverts;
memset(&meshcache.vbo.coord2, 0, sizeof(meshcache.vbo.coord2));
mesh->xyz_array = p1->ofsverts;
mesh->xyz2_array = NULL;
}
else
{
meshcache.vbo.coord = p1->vboverts;
meshcache.vbo.coord2 = p2->vboverts;
mesh->xyz_blendw[0] = 1-lerp;
mesh->xyz_blendw[1] = lerp;
mesh->xyz_array = p1->ofsverts;
mesh->xyz2_array = p2->ofsverts;
}
if (vbop && meshcache.vbo.indicies.dummy)
*vbop = meshcache.vbop = &meshcache.vbo;
}
}
meshcache.acoords1 = mesh->xyz_array;
meshcache.acoords2 = mesh->xyz2_array;
meshcache.anorm = mesh->normals_array;
meshcache.anorms = mesh->snormals_array;
meshcache.anormt = mesh->tnormals_array;
if (vbop)
meshcache.vbop = *vbop;
#ifdef SKELETALMODELS
if (meshcache.usebonepose)
{
mesh->bonenums = inf->ofs_skel_idx;
mesh->boneweights = inf->ofs_skel_weight;
mesh->bones = meshcache.usebonepose;
mesh->numbones = inf->numbones;
}
#endif
return true; //to allow the mesh to be dlighted.
}
#endif
static float PlaneNearest(vec3_t normal, vec3_t mins, vec3_t maxs)
{
float result;
return 128;
#if 1
result = fabs(normal[0] * maxs[0]);
result += fabs(normal[1] * maxs[1]);
result += fabs(normal[2] * maxs[2]);
#else
result = normal[0] * ((normal[0] < 0)?mins[0]:maxs[0]);
result += normal[1] * ((normal[1] < 0)?mins[1]:maxs[1]);
result += normal[2] * ((normal[2] < 0)?mins[2]:maxs[2]);
#endif
return result;
}
static qboolean Mod_Trace_Trisoup(vecV_t *posedata, index_t *indexes, int numindexes, vec3_t start, vec3_t end, vec3_t mins, vec3_t maxs, trace_t *trace)
{
qboolean impacted = false;
int i;
float *p1, *p2, *p3;
vec3_t edge1, edge2, edge3;
vec3_t normal;
vec3_t edgenormal;
float planedist;
float diststart, distend;
float expand;
float frac;
// float temp;
vec3_t impactpoint;
for (i = 0; i < numindexes; i+=3)
{
p1 = posedata[indexes[i+0]];
p2 = posedata[indexes[i+1]];
p3 = posedata[indexes[i+2]];
VectorSubtract(p1, p2, edge1);
VectorSubtract(p3, p2, edge2);
CrossProduct(edge1, edge2, normal);
expand = PlaneNearest(normal, mins, maxs);
planedist = DotProduct(p1, normal);
diststart = DotProduct(start, normal);
if (diststart <= planedist)
continue; //start on back side.
distend = DotProduct(end, normal);
if (distend >= planedist)
continue; //end on front side (as must start - doesn't cross).
frac = (diststart - planedist - 1) / (diststart-distend);
if (frac < 0)
frac = 0;
if (frac >= trace->fraction) //already found one closer.
continue;
impactpoint[0] = start[0] + frac*(end[0] - start[0]);
impactpoint[1] = start[1] + frac*(end[1] - start[1]);
impactpoint[2] = start[2] + frac*(end[2] - start[2]);
// temp = DotProduct(impactpoint, normal)-planedist;
CrossProduct(edge1, normal, edgenormal);
// temp = DotProduct(impactpoint, edgenormal)-DotProduct(p2, edgenormal);
if (DotProduct(impactpoint, edgenormal) > DotProduct(p2, edgenormal))
continue;
CrossProduct(normal, edge2, edgenormal);
if (DotProduct(impactpoint, edgenormal) > DotProduct(p3, edgenormal))
continue;
VectorSubtract(p1, p3, edge3);
CrossProduct(normal, edge3, edgenormal);
if (DotProduct(impactpoint, edgenormal) > DotProduct(p1, edgenormal))
continue;
trace->fraction = frac;
VectorCopy(impactpoint, trace->endpos);
VectorCopy(normal, trace->plane.normal);
impacted = true;
}
return impacted;
}
//The whole reason why model loading is supported in the server.
qboolean Mod_Trace(model_t *model, int forcehullnum, int frame, vec3_t axis[3], vec3_t start, vec3_t end, vec3_t mins, vec3_t maxs, unsigned int contentsmask, trace_t *trace)
{
galiasinfo_t *mod = Mod_Extradata(model);
galiasgroup_t *group;
galiaspose_t *pose;
float frac;
// float temp;
vecV_t *posedata;
index_t *indexes;
int surfnum = 0;
int cursurfnum = -1;
vec3_t start_l, end_l;
if (axis)
{
start_l[0] = DotProduct(start, axis[0]);
start_l[1] = DotProduct(start, axis[1]);
start_l[2] = DotProduct(start, axis[2]);
end_l[0] = DotProduct(end, axis[0]);
end_l[1] = DotProduct(end, axis[1]);
end_l[2] = DotProduct(end, axis[2]);
}
else
{
VectorCopy(start, start_l);
VectorCopy(end, end_l);
}
while(mod)
{
indexes = mod->ofs_indexes;
#ifdef SKELETALMODELS
if (!mod->groups)
{
//certain models have no possibility of animation.
posedata = mod->ofs_skel_xyz;
}
else
#endif
{
group = mod->groupofs;
pose = group[0].poseofs;
posedata = pose->ofsverts;
#ifdef SKELETALMODELS
if (mod->numbones && mod->shares_verts != cursurfnum)
{
float bonepose[MAX_BONES][12];
posedata = alloca(mod->numverts*sizeof(vecV_t));
frac = 1;
if (group->isheirachical)
{
if (mod->shares_bones != cursurfnum)
R_LerpBones(&frac, (float**)posedata, 1, mod->ofsbones, mod->numbones, bonepose);
Alias_TransformVerticies_SW((float*)bonepose, mod->ofsswtransforms, mod->numswtransforms, posedata, NULL);
}
else
Alias_TransformVerticies_SW((float*)posedata, mod->ofsswtransforms, mod->numswtransforms, posedata, NULL);
cursurfnum = mod->shares_verts;
}
#endif
}
if (Mod_Trace_Trisoup(posedata, indexes, mod->numindexes, start_l, end_l, mins, maxs, trace) && axis)
{
if (axis)
{
vec3_t iaxis[3];
vec3_t norm;
Matrix3x3_RM_Invert_Simple((void *)axis, iaxis);
VectorCopy(trace->plane.normal, norm);
trace->plane.normal[0] = DotProduct(norm, iaxis[0]);
trace->plane.normal[1] = DotProduct(norm, iaxis[1]);
trace->plane.normal[2] = DotProduct(norm, iaxis[2]);
}
// frac = traceinfo.truefraction;
/*
diststart = DotProduct(traceinfo.start, trace->plane.normal);
distend = DotProduct(traceinfo.end, trace->plane.normal);
if (diststart == distend)
frac = 0;
else
{
frac = (diststart - trace->plane.dist) / (diststart-distend);
if (frac < 0)
frac = 0;
else if (frac > 1)
frac = 1;
}*/
/*okay, this is where it hits this plane*/
// trace->endpos[0] = traceinfo.start[0] + frac*(traceinfo.end[0] - traceinfo.start[0]);
// trace->endpos[1] = traceinfo.start[1] + frac*(traceinfo.end[1] - traceinfo.start[1]);
// trace->endpos[2] = traceinfo.start[2] + frac*(traceinfo.end[2] - traceinfo.start[2]);
}
mod = mod->nextsurf;
surfnum++;
}
trace->allsolid = false;
return trace->fraction != 1;
}
static void Mod_ClampModelSize(model_t *mod)
{
#ifndef SERVERONLY
int i;
float rad=0, axis;
axis = (mod->maxs[0] - mod->mins[0]);
rad += axis*axis;
axis = (mod->maxs[1] - mod->mins[1]);
rad += axis*axis;
axis = (mod->maxs[2] - mod->mins[2]);
rad += axis*axis;
mod->tainted = false;
if (mod->engineflags & MDLF_DOCRC)
{
if (!strcmp(mod->name, "progs/eyes.mdl"))
{ //this is checked elsewhere to make sure the crc matches (this is to make sure the crc check was actually called)
if (mod->type != mod_alias || mod->fromgame != fg_quake || mod->flags)
mod->tainted = true;
}
}
mod->clampscale = 1;
for (i = 0; i < sizeof(clampedmodel)/sizeof(clampedmodel[0]); i++)
{
if (!strcmp(mod->name, clampedmodel[i].name))
{
if (rad > clampedmodel[i].furthestallowedextremety)
{
axis = clampedmodel[i].furthestallowedextremety;
mod->clampscale = axis/rad;
Con_DPrintf("\"%s\" will be clamped.\n", mod->name);
}
return;
}
}
Con_DPrintf("Don't know what size to clamp \"%s\" to (size:%f).\n", mod->name, rad);
#endif
}
static int R_FindTriangleWithEdge (index_t *indexes, int numtris, int start, int end, int ignore)
{
int i;
int match, count;
count = 0;
match = -1;
for (i = 0; i < numtris; i++, indexes += 3)
{
if ( (indexes[0] == start && indexes[1] == end)
|| (indexes[1] == start && indexes[2] == end)
|| (indexes[2] == start && indexes[0] == end) ) {
if (i != ignore)
match = i;
count++;
} else if ( (indexes[1] == start && indexes[0] == end)
|| (indexes[2] == start && indexes[1] == end)
|| (indexes[0] == start && indexes[2] == end) ) {
count++;
}
}
// detect edges shared by three triangles and make them seams
if (count > 2)
match = -1;
return match;
}
static void Mod_BuildTriangleNeighbours ( int *neighbours, index_t *indexes, int numtris )
{
int i, *n;
index_t *index;
for (i = 0, index = indexes, n = neighbours; i < numtris; i++, index += 3, n += 3)
{
n[0] = R_FindTriangleWithEdge (indexes, numtris, index[1], index[0], i);
n[1] = R_FindTriangleWithEdge (indexes, numtris, index[2], index[1], i);
n[2] = R_FindTriangleWithEdge (indexes, numtris, index[0], index[2], i);
}
}
void Mod_CompileTriangleNeighbours(galiasinfo_t *galias)
{
if (Sh_StencilShadowsActive())
{
int *neighbours;
neighbours = ZG_Malloc(&loadmodel->memgroup, sizeof(int)*galias->numindexes/3*3);
galias->ofs_trineighbours = neighbours;
Mod_BuildTriangleNeighbours(neighbours, galias->ofs_indexes, galias->numindexes/3);
}
}
typedef struct
{
int firstpose;
int posecount;
float fps;
qboolean loop;
char name[MAX_QPATH];
} frameinfo_t;
static frameinfo_t *ParseFrameInfo(char *modelname, int *numgroups)
{
int count = 0;
int maxcount = 0;
char *line, *file;
frameinfo_t *frames = NULL;
line = file = FS_LoadMallocFile(va("%s.framegroups", modelname));
if (!file)
return NULL;
while(line && *line)
{
line = Cmd_TokenizeString(line, false, false);
if (Cmd_Argc())
{
if (count == maxcount)
{
maxcount += 32;
frames = realloc(frames, sizeof(*frames)*maxcount);
}
frames[count].firstpose = atoi(Cmd_Argv(0));
frames[count].posecount = atoi(Cmd_Argv(1));
frames[count].fps = atof(Cmd_Argv(2));
frames[count].loop = !!atoi(Cmd_Argv(3));
Q_strncpyz(frames[count].name, Cmd_Argv(4), sizeof(frames[count].name));
count++;
}
}
BZ_Free(file);
*numgroups = count;
return frames;
}
//called for non-skeletal model formats.
void Mod_BuildTextureVectors(galiasinfo_t *galias)
//vec3_t *vc, vec2_t *tc, vec3_t *nv, vec3_t *sv, vec3_t *tv, index_t *idx, int numidx, int numverts)
{
#ifndef SERVERONLY
int i, p;
galiasgroup_t *group;
galiaspose_t *pose;
vecV_t *vc;
vec3_t *nv, *sv, *tv;
vec2_t *tc;
index_t *idx;
int vbospace = 0;
vbobctx_t vboctx;
//don't fail on dedicated servers
if (!BE_VBO_Begin)
return;
idx = galias->ofs_indexes;
tc = galias->ofs_st_array;
group = galias->groupofs;
//determine the amount of space we need for our vbos.
vbospace += sizeof(*tc) * galias->numverts;
for (i = 0; i < galias->groups; i++)
{
vbospace += group[i].numposes * galias->numverts * (sizeof(vecV_t)+sizeof(vec3_t)*3);
}
BE_VBO_Begin(&vboctx, vbospace);
BE_VBO_Data(&vboctx, tc, sizeof(*tc) * galias->numverts, &galias->vbotexcoords);
for (i = 0; i < galias->groups; i++, group++)
{
pose = group->poseofs;
for (p = 0; p < group->numposes; p++, pose++)
{
vc = pose->ofsverts;
nv = pose->ofsnormals;
if (pose->ofssvector != 0 && pose->ofstvector != 0)
{
sv = pose->ofssvector;
tv = pose->ofstvector;
Mod_AccumulateTextureVectors(vc, tc, nv, sv, tv, idx, galias->numindexes);
Mod_NormaliseTextureVectors(nv, sv, tv, galias->numverts);
}
else
{ //shouldn't really happen... make error?
sv = NULL;
tv = NULL;
}
BE_VBO_Data(&vboctx, vc, sizeof(*vc) * galias->numverts, &pose->vboverts);
BE_VBO_Data(&vboctx, nv, sizeof(*nv) * galias->numverts, &pose->vbonormals);
BE_VBO_Data(&vboctx, sv, sizeof(*sv) * galias->numverts, &pose->vbosvector);
BE_VBO_Data(&vboctx, tv, sizeof(*tv) * galias->numverts, &pose->vbotvector);
}
}
BE_VBO_Finish(&vboctx, idx, sizeof(*idx) * galias->numindexes, &galias->vboindicies);
#endif
}
#if defined(D3DQUAKE) || defined(GLQUAKE)
/*
=================
Mod_FloodFillSkin
Fill background pixels so mipmapping doesn't have haloes - Ed
=================
*/
typedef struct
{
short x, y;
} floodfill_t;
// must be a power of 2
#define FLOODFILL_FIFO_SIZE 0x1000
#define FLOODFILL_FIFO_MASK (FLOODFILL_FIFO_SIZE - 1)
#define FLOODFILL_STEP( off, dx, dy ) \
{ \
if (pos[off] == fillcolor) \
{ \
pos[off] = 255; \
fifo[inpt].x = x + (dx), fifo[inpt].y = y + (dy); \
inpt = (inpt + 1) & FLOODFILL_FIFO_MASK; \
} \
else if (pos[off] != 255) fdc = pos[off]; \
}
static void Mod_FloodFillSkin( qbyte *skin, int skinwidth, int skinheight )
{
qbyte fillcolor = *skin; // assume this is the pixel to fill
floodfill_t fifo[FLOODFILL_FIFO_SIZE];
int inpt = 0, outpt = 0;
int filledcolor = -1;
int i;
if (filledcolor == -1)
{
filledcolor = 0;
// attempt to find opaque black
for (i = 0; i < 256; ++i)
if (d_8to24rgbtable[i] == (255 << 0)) // alpha 1.0
{
filledcolor = i;
break;
}
}
// can't fill to filled color or to transparent color (used as visited marker)
if ((fillcolor == filledcolor) || (fillcolor == 255))
{
//printf( "not filling skin from %d to %d\n", fillcolor, filledcolor );
return;
}
fifo[inpt].x = 0, fifo[inpt].y = 0;
inpt = (inpt + 1) & FLOODFILL_FIFO_MASK;
while (outpt != inpt)
{
int x = fifo[outpt].x, y = fifo[outpt].y;
int fdc = filledcolor;
qbyte *pos = &skin[x + skinwidth * y];
outpt = (outpt + 1) & FLOODFILL_FIFO_MASK;
if (x > 0) FLOODFILL_STEP( -1, -1, 0 );
if (x < skinwidth - 1) FLOODFILL_STEP( 1, 1, 0 );
if (y > 0) FLOODFILL_STEP( -skinwidth, 0, -1 );
if (y < skinheight - 1) FLOODFILL_STEP( skinwidth, 0, 1 );
skin[x + skinwidth * y] = fdc;
}
}
#endif
//additional skin loading
char ** skinfilelist;
int skinfilecount;
static qboolean VARGS Mod_TryAddSkin(const char *skinname, ...)
{
va_list argptr;
char string[MAX_QPATH];
//make sure we don't add it twice
int i;
va_start (argptr, skinname);
vsnprintf (string,sizeof(string)-1, skinname,argptr);
va_end (argptr);
string[MAX_QPATH-1] = '\0';
for (i = 0; i < skinfilecount; i++)
{
if (!strcmp(skinfilelist[i], string))
return true; //already added
}
if (!COM_FCheckExists(string))
return false;
skinfilelist = BZ_Realloc(skinfilelist, sizeof(*skinfilelist)*(skinfilecount+1));
skinfilelist[skinfilecount] = Z_Malloc(strlen(string)+1);
strcpy(skinfilelist[skinfilecount], string);
skinfilecount++;
return true;
}
int QDECL Mod_EnumerateSkins(const char *name, int size, void *param, searchpathfuncs_t *spath)
{
Mod_TryAddSkin(name);
return true;
}
int Mod_BuildSkinFileList(char *modelname)
{
int i;
char skinfilename[MAX_QPATH];
//flush the old list
for (i = 0; i < skinfilecount; i++)
{
Z_Free(skinfilelist[i]);
skinfilelist[i] = NULL;
}
skinfilecount=0;
COM_StripExtension(modelname, skinfilename, sizeof(skinfilename));
//try and add numbered skins, and then try fixed names.
for (i = 0; ; i++)
{
if (!Mod_TryAddSkin("%s_%i.skin", modelname, i))
{
if (i == 0)
{
if (!Mod_TryAddSkin("%s_default.skin", skinfilename, i))
break;
}
else if (i == 1)
{
if (!Mod_TryAddSkin("%s_blue.skin", skinfilename, i))
break;
}
else if (i == 2)
{
if (!Mod_TryAddSkin("%s_red.skin", skinfilename, i))
break;
}
else if (i == 3)
{
if (!Mod_TryAddSkin("%s_green.skin", skinfilename, i))
break;
}
else if (i == 4)
{
if (!Mod_TryAddSkin("%s_yellow.skin", skinfilename, i))
break;
}
else
break;
}
}
// if (strstr(modelname, "lower") || strstr(modelname, "upper") || strstr(modelname, "head"))
// {
COM_EnumerateFiles(va("%s_*.skin", modelname), Mod_EnumerateSkins, NULL);
COM_EnumerateFiles(va("%s_*.skin", skinfilename), Mod_EnumerateSkins, NULL);
// }
// else
// COM_EnumerateFiles("*.skin", Mod_EnumerateSkins, NULL);
return skinfilecount;
}
//This is a hack. It uses an assuption about q3 player models.
void Mod_ParseQ3SkinFile(char *out, char *surfname, char *modelname, int skinnum, char *skinfilename)
{
const char *f = NULL, *p;
int len;
if (skinnum >= skinfilecount)
return;
if (skinfilename)
strcpy(skinfilename, skinfilelist[skinnum]);
f = COM_LoadTempMoreFile(skinfilelist[skinnum]);
while(f)
{
f = COM_ParseToken(f,NULL);
if (!f)
return;
if (!strcmp(com_token, "replace"))
{
f = COM_ParseToken(f, NULL);
len = strlen(com_token);
//copy surfname -> out, until we meet the part we need to replace
while(*surfname)
{
if (!strncmp(com_token, surfname, len))
//found it
{
surfname+=len;
f = COM_ParseToken(f, NULL);
p = com_token;
while(*p) //copy the replacement
*out++ = *p++;
while(*surfname) //copy the remaining
*out++ = *surfname++;
*out++ = '\0'; //we didn't find it.
return;
}
*out++ = *surfname++;
}
*out++ = '\0'; //we didn't find it.
return;
}
else
{
while(*f == ' ' || *f == '\t')
f++;
if (*f == ',')
{
if (!strcmp(com_token, surfname))
{
f++;
COM_ParseToken(f, NULL);
strcpy(out, com_token);
return;
}
}
}
p = strchr(f, '\n');
if (!p)
f = f+strlen(f);
else
f = p+1;
if (!*f)
break;
}
}
#if defined(D3DQUAKE) || defined(GLQUAKE)
shader_t *Mod_LoadSkinFile(char *surfacename, int skinnumber, unsigned char *rawdata, int width, int height, unsigned char *palette)
{
shader_t *shader;
char shadername[MAX_QPATH];
Q_strncpyz(shadername, surfacename, sizeof(shadername));
Mod_ParseQ3SkinFile(shadername, surfacename, loadmodel->name, skinnumber, NULL);
shader = R_RegisterSkin(shadername, loadmodel->name);
R_BuildDefaultTexnums(&shader->defaulttextures, shader);
if (shader->flags & SHADER_NOIMAGE)
Con_Printf("Unable to load texture for shader \"%s\" for model \"%s\"\n", shader->name, loadmodel->name);
return shader;
}
#endif
//Q1 model loading
#if 1
static galiasinfo_t *galias;
static dmdl_t *pq1inmodel;
#define NUMVERTEXNORMALS 162
extern float r_avertexnormals[NUMVERTEXNORMALS][3];
// mdltype 0 = q1, 1 = qtest, 2 = rapo/h2
static void Alias_LoadPose(vecV_t *verts, vec3_t *normals, vec3_t *svec, vec3_t *tvec, dtrivertx_t *pinframe, int *seamremaps, int mdltype)
{
int j;
if (mdltype == 2)
{
for (j = 0; j < galias->numverts; j++)
{
verts[j][0] = pinframe[seamremaps[j]].v[0]*pq1inmodel->scale[0]+pq1inmodel->scale_origin[0];
verts[j][1] = pinframe[seamremaps[j]].v[1]*pq1inmodel->scale[1]+pq1inmodel->scale_origin[1];
verts[j][2] = pinframe[seamremaps[j]].v[2]*pq1inmodel->scale[2]+pq1inmodel->scale_origin[2];
#ifndef SERVERONLY
VectorCopy(r_avertexnormals[pinframe[seamremaps[j]].lightnormalindex], normals[j]);
#endif
}
}
else
{
for (j = 0; j < pq1inmodel->numverts; j++)
{
verts[j][0] = pinframe[j].v[0]*pq1inmodel->scale[0]+pq1inmodel->scale_origin[0];
verts[j][1] = pinframe[j].v[1]*pq1inmodel->scale[1]+pq1inmodel->scale_origin[1];
verts[j][2] = pinframe[j].v[2]*pq1inmodel->scale[2]+pq1inmodel->scale_origin[2];
#ifndef SERVERONLY
VectorCopy(r_avertexnormals[pinframe[j].lightnormalindex], normals[j]);
#endif
if (seamremaps[j] != j)
{
VectorCopy(verts[j], verts[seamremaps[j]]);
#ifndef SERVERONLY
VectorCopy(normals[j], normals[seamremaps[j]]);
#endif
}
}
}
}
static void *Alias_LoadFrameGroup (daliasframetype_t *pframetype, int *seamremaps, int mdltype)
{
galiaspose_t *pose;
galiasgroup_t *frame = galias->groupofs;
dtrivertx_t *pinframe;
daliasframe_t *frameinfo;
int i, k;
daliasgroup_t *ingroup;
daliasinterval_t *intervals;
float sinter;
vec3_t *normals, *svec, *tvec;
vecV_t *verts;
int aliasframesize = (mdltype == 1) ? sizeof(daliasframe_t)-16 : sizeof(daliasframe_t);
#ifdef SERVERONLY
normals = NULL;
svec = NULL;
tvec = NULL;
#endif
for (i = 0; i < pq1inmodel->numframes; i++)
{
switch(LittleLong(pframetype->type))
{
case ALIAS_SINGLE:
frameinfo = (daliasframe_t*)((char *)(pframetype+1)); // qtest aliasframe is a subset
pinframe = (dtrivertx_t*)((char*)frameinfo+aliasframesize);
#ifndef SERVERONLY
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, sizeof(galiaspose_t) + (sizeof(vecV_t)+sizeof(vec3_t)*3)*galias->numverts);
#else
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, sizeof(galiaspose_t) + (sizeof(vecV_t))*galias->numverts);
#endif
frame->poseofs = pose;
frame->numposes = 1;
galias->groups++;
if (mdltype == 1)
frame->name[0] = '\0';
else
Q_strncpyz(frame->name, frameinfo->name, sizeof(frame->name));
verts = (vecV_t *)(pose+1);
pose->ofsverts = verts;
#ifndef SERVERONLY
normals = (vec3_t*)&verts[galias->numverts];
svec = &normals[galias->numverts];
tvec = &svec[galias->numverts];
pose->ofsnormals = normals;
pose->ofssvector = svec;
pose->ofstvector = tvec;
#endif
Alias_LoadPose(verts, normals, svec, tvec, pinframe, seamremaps, mdltype);
// GL_GenerateNormals((float*)verts, (float*)normals, (int *)((char *)galias + galias->ofs_indexes), galias->numindexes/3, galias->numverts);
pframetype = (daliasframetype_t *)&pinframe[pq1inmodel->numverts];
break;
case ALIAS_GROUP:
case ALIAS_GROUP_SWAPPED: // prerelease
ingroup = (daliasgroup_t *)(pframetype+1);
frame->numposes = LittleLong(ingroup->numframes);
#ifdef SERVERONLY
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, frame->numposes*(sizeof(galiaspose_t) + sizeof(vecV_t)*galias->numverts));
verts = (vecV_t *)(pose+frame->numposes);
#else
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, frame->numposes*(sizeof(galiaspose_t) + (sizeof(vecV_t)+sizeof(vec3_t)*3)*galias->numverts));
verts = (vecV_t *)(pose+frame->numposes);
normals = (vec3_t*)&verts[galias->numverts];
svec = &normals[galias->numverts];
tvec = &svec[galias->numverts];
#endif
frame->poseofs = pose;
frame->loop = true;
galias->groups++;
intervals = (daliasinterval_t *)(ingroup+1);
sinter = LittleFloat(intervals->interval);
if (sinter <= 0)
sinter = 0.1;
frame->rate = 1/sinter;
pinframe = (dtrivertx_t *)(intervals+frame->numposes);
for (k = 0; k < frame->numposes; k++)
{
pose->ofsverts = verts;
#ifndef SERVERONLY
pose->ofsnormals = normals;
pose->ofssvector = svec;
pose->ofstvector = tvec;
#endif
frameinfo = (daliasframe_t*)pinframe;
pinframe = (dtrivertx_t *)((char *)frameinfo + aliasframesize);
if (k == 0)
{
if (mdltype == 1)
frame->name[0] = '\0';
else
Q_strncpyz(frame->name, frameinfo->name, sizeof(frame->name));
}
Alias_LoadPose(verts, normals, svec, tvec, pinframe, seamremaps, mdltype);
#ifndef SERVERONLY
verts = (vecV_t*)&tvec[galias->numverts];
normals = (vec3_t*)&verts[galias->numverts];
svec = &normals[galias->numverts];
tvec = &svec[galias->numverts];
#else
verts = &verts[galias->numverts];
#endif
pose++;
pinframe += pq1inmodel->numverts;
}
// GL_GenerateNormals((float*)verts, (float*)normals, (int *)((char *)galias + galias->ofs_indexes), galias->numindexes/3, galias->numverts);
pframetype = (daliasframetype_t *)pinframe;
break;
default:
Con_Printf(CON_ERROR "Bad frame type in %s\n", loadmodel->name);
return NULL;
}
frame++;
}
return pframetype;
}
//greatly reduced version of Q1_LoadSkins
//just skips over the data
static void *Q1_LoadSkins_SV (daliasskintype_t *pskintype, qboolean alpha)
{
int i;
int s;
int *count;
float *intervals;
qbyte *data;
s = pq1inmodel->skinwidth*pq1inmodel->skinheight;
for (i = 0; i < pq1inmodel->numskins; i++)
{
switch(LittleLong(pskintype->type))
{
case ALIAS_SKIN_SINGLE:
pskintype = (daliasskintype_t *)((char *)(pskintype+1)+s);
break;
default:
count = (int *)(pskintype+1);
intervals = (float *)(count+1);
data = (qbyte *)(intervals + LittleLong(*count));
data += s*LittleLong(*count);
pskintype = (daliasskintype_t *)data;
break;
}
}
galias->numskins=pq1inmodel->numskins;
return pskintype;
}
#if defined(GLQUAKE) || defined(D3DQUAKE)
static void *Q1_LoadSkins_GL (daliasskintype_t *pskintype, unsigned int skintranstype)
{
shader_t **shaders;
qbyte **ofstexels;
char skinname[MAX_QPATH];
int i;
int s, t;
float sinter;
daliasskingroup_t *count;
daliasskininterval_t *intervals;
qbyte *data, *saved;
galiasskin_t *outskin = galias->ofsskins;
texid_t texture;
texid_t fbtexture;
texid_t bumptexture;
s = pq1inmodel->skinwidth*pq1inmodel->skinheight;
for (i = 0; i < pq1inmodel->numskins; i++)
{
switch(LittleLong(pskintype->type))
{
case ALIAS_SKIN_SINGLE:
outskin->skinwidth = pq1inmodel->skinwidth;
outskin->skinheight = pq1inmodel->skinheight;
//LH's naming scheme ("models" is likly to be ignored)
fbtexture = r_nulltex;
bumptexture = r_nulltex;
snprintf(skinname, sizeof(skinname), "%s_%i", loadmodel->name, i);
texture = R_LoadReplacementTexture(skinname, "models", IF_NOALPHA);
if (TEXVALID(texture))
{
if (TEXVALID(texture) && r_fb_models.ival)
{
snprintf(skinname, sizeof(skinname), "%s_%i_luma", loadmodel->name, i);
fbtexture = R_LoadReplacementTexture(skinname, "models", 0);
}
if (r_loadbumpmapping)
{
snprintf(skinname, sizeof(skinname), "%s_%i_bump", loadmodel->name, i);
bumptexture = R_LoadBumpmapTexture(skinname, "models");
}
}
else
{
snprintf(skinname, sizeof(skinname), "%s_%i", loadname, i);
texture = R_LoadReplacementTexture(skinname, "models", IF_NOALPHA);
if (TEXVALID(texture) && r_fb_models.ival)
{
snprintf(skinname, sizeof(skinname), "%s_%i_luma", loadname, i);
fbtexture = R_LoadReplacementTexture(skinname, "models", 0);
}
if (TEXVALID(texture) && r_loadbumpmapping)
{
snprintf(skinname, sizeof(skinname), "%s_%i_bump", loadname, i);
bumptexture = R_LoadBumpmapTexture(skinname, "models");
}
}
//but only preload it if we have no replacement.
if (!TEXVALID(texture) || (loadmodel->engineflags & MDLF_NOTREPLACEMENTS))
{
//we're not using 24bits
shaders = ZG_Malloc(&loadmodel->memgroup, sizeof(*shaders)+sizeof(*ofstexels)+s);
ofstexels = (qbyte**)(shaders+1);
saved = (qbyte*)(ofstexels+1);
outskin->ofstexels = ofstexels;
ofstexels[0] = saved;
memcpy(saved, pskintype+1, s);
Mod_FloodFillSkin(saved, outskin->skinwidth, outskin->skinheight);
//the extra underscore is to stop replacement matches
if (!TEXVALID(texture))
{
snprintf(skinname, sizeof(skinname), "%s__%i", loadname, i);
switch (skintranstype)
{
default:
texture = R_LoadTexture(skinname,outskin->skinwidth,outskin->skinheight, TF_SOLID8, saved, IF_NOALPHA|IF_NOGAMMA);
if (r_fb_models.ival)
{
snprintf(skinname, sizeof(skinname), "%s__%i_luma", loadname, i);
fbtexture = R_LoadTextureFB(skinname, outskin->skinwidth, outskin->skinheight, saved, IF_NOGAMMA);
}
if (r_loadbumpmapping)
{
snprintf(skinname, sizeof(skinname), "%s__%i_bump", loadname, i);
bumptexture = R_LoadTexture8BumpPal(skinname, outskin->skinwidth, outskin->skinheight, saved, IF_NOGAMMA);
}
break;
case 2:
texture = R_LoadTexture(skinname,outskin->skinwidth,outskin->skinheight, TF_H2_T7G1, saved, IF_NOGAMMA);
break;
case 3:
texture = R_LoadTexture(skinname,outskin->skinwidth,outskin->skinheight, TF_H2_TRANS8_0, saved, IF_NOGAMMA);
break;
case 4:
texture = R_LoadTexture(skinname,outskin->skinwidth,outskin->skinheight, TF_H2_T4A4, saved, IF_NOGAMMA);
break;
}
}
}
else
shaders = ZG_Malloc(&loadmodel->memgroup, sizeof(*shaders));
outskin->numshaders=1;
outskin->ofsshaders = shaders;
Q_snprintfz(skinname, sizeof(skinname), "%s_%i", loadname, i);
if (skintranstype == 4)
shaders[0] = R_RegisterShader(skinname, SUF_NONE,
"{\n"
"{\n"
"map $diffuse\n"
"blendfunc gl_one_minus_src_alpha gl_src_alpha\n"
"rgbgen lightingDiffuse\n"
"cull disable\n"
"depthwrite\n"
"}\n"
"}\n");
else if (skintranstype == 3)
shaders[0] = R_RegisterShader(skinname, SUF_NONE,
"{\n"
"{\n"
"map $diffuse\n"
"alphafunc ge128\n"
"rgbgen lightingDiffuse\n"
"depthwrite\n"
"}\n"
"}\n");
else if (skintranstype)
shaders[0] = R_RegisterShader(skinname, SUF_NONE,
"{\n"
"{\n"
"map $diffuse\n"
"blendfunc gl_src_alpha gl_one_minus_src_alpha\n"
"rgbgen lightingDiffuse\n"
"depthwrite\n"
"}\n"
"}\n");
else
shaders[0] = R_RegisterSkin(skinname, loadmodel->name);
shaders[0]->defaulttextures.base = texture;
shaders[0]->defaulttextures.fullbright = fbtexture;
shaders[0]->defaulttextures.bump = bumptexture;
//13/4/08 IMPLEMENTME
if (r_skin_overlays.ival)
{
snprintf(skinname, sizeof(skinname), "%s_%i_pants", loadname, i);
shaders[0]->defaulttextures.loweroverlay = R_LoadReplacementTexture(skinname, "models", 0);
snprintf(skinname, sizeof(skinname), "%s_%i_shirt", loadname, i);
shaders[0]->defaulttextures.upperoverlay = R_LoadReplacementTexture(skinname, "models", 0);
}
R_BuildDefaultTexnums(&shaders[0]->defaulttextures, shaders[0]);
pskintype = (daliasskintype_t *)((char *)(pskintype+1)+s);
break;
default:
outskin->skinwidth = pq1inmodel->skinwidth;
outskin->skinheight = pq1inmodel->skinheight;
count = (daliasskingroup_t*)(pskintype+1);
intervals = (daliasskininterval_t *)(count+1);
outskin->numshaders = LittleLong(count->numskins);
data = (qbyte *)(intervals + outskin->numshaders);
shaders = ZG_Malloc(&loadmodel->memgroup, sizeof(*shaders)*outskin->numshaders + sizeof(*ofstexels)*outskin->numshaders);
ofstexels = (qbyte**)(shaders+outskin->numshaders);
outskin->ofsshaders = shaders;
outskin->ofstexels = ofstexels;
sinter = LittleFloat(intervals[0].interval);
if (sinter <= 0)
sinter = 0.1;
outskin->skinspeed = 1/sinter;
for (t = 0; t < outskin->numshaders; t++,data+=s)
{
texture = r_nulltex;
fbtexture = r_nulltex;
//LH naming scheme
if (!TEXVALID(texture))
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i", loadmodel->name, i, t);
texture = R_LoadReplacementTexture(skinname, "models", IF_NOALPHA);
}
if (!TEXVALID(fbtexture) && r_fb_models.ival)
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i_luma", loadmodel->name, i, t);
fbtexture = R_LoadReplacementTexture(skinname, "models", 0);
}
//Fuhquake naming scheme
if (!TEXVALID(texture))
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i", loadname, i, t);
texture = R_LoadReplacementTexture(skinname, "models", IF_NOALPHA);
}
if (!TEXVALID(fbtexture) && r_fb_models.ival)
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i_luma", loadname, i, t);
fbtexture = R_LoadReplacementTexture(skinname, "models", 0);
}
if (!TEXVALID(texture) || (!TEXVALID(fbtexture) && r_fb_models.ival))
{
saved = ZG_Malloc(&loadmodel->memgroup, s);
ofstexels[t] = saved;
memcpy(saved, data, s);
Mod_FloodFillSkin(saved, outskin->skinwidth, outskin->skinheight);
if (!TEXVALID(texture))
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i", loadname, i, t);
texture = R_LoadTexture8(skinname, outskin->skinwidth, outskin->skinheight, saved, (skintranstype?0:IF_NOALPHA)|IF_NOGAMMA, skintranstype);
}
if (!TEXVALID(fbtexture) && r_fb_models.value)
{
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i_luma", loadname, i, t);
fbtexture = R_LoadTextureFB(skinname, outskin->skinwidth, outskin->skinheight, saved, IF_NOGAMMA);
}
}
Q_snprintfz(skinname, sizeof(skinname), "%s_%i_%i", loadname, i, t);
shaders[t] = R_RegisterSkin(skinname, loadmodel->name);
TEXASSIGN(shaders[t]->defaulttextures.base, texture);
TEXASSIGN(shaders[t]->defaulttextures.fullbright, fbtexture);
TEXASSIGN(shaders[t]->defaulttextures.loweroverlay, r_nulltex);
TEXASSIGN(shaders[t]->defaulttextures.upperoverlay, r_nulltex);
R_BuildDefaultTexnums(&shaders[t]->defaulttextures, shaders[t]);
}
pskintype = (daliasskintype_t *)data;
break;
}
outskin++;
}
galias->numskins=pq1inmodel->numskins;
return pskintype;
}
#endif
qboolean QDECL Mod_LoadQ1Model (model_t *mod, void *buffer)
{
#ifndef SERVERONLY
vec2_t *st_array;
int j;
#endif
int version;
int i, onseams;
dstvert_t *pinstverts;
dtriangle_t *pinq1triangles;
dh2triangle_t *pinh2triangles;
int *seamremap;
index_t *indexes;
daliasskintype_t *skinstart;
int skintranstype;
int size;
unsigned int hdrsize;
void *end;
qboolean qtest = false;
qboolean rapo = false;
loadmodel=mod;
pq1inmodel = (dmdl_t *)buffer;
hdrsize = sizeof(dmdl_t) - sizeof(int);
loadmodel->engineflags |= MDLF_NEEDOVERBRIGHT;
version = LittleLong(pq1inmodel->version);
if (version == QTESTALIAS_VERSION)
{
hdrsize = (size_t)&((dmdl_t*)NULL)->flags;
qtest = true;
}
else if (version == 50)
{
hdrsize = sizeof(dmdl_t);
rapo = true;
}
else if (version != ALIAS_VERSION)
{
Con_Printf (CON_ERROR "%s has wrong version number (%i should be %i)\n",
mod->name, version, ALIAS_VERSION);
return false;
}
seamremap = (int*)pq1inmodel; //I like overloading locals.
i = hdrsize/4 - 1;
for (; i >= 0; i--)
seamremap[i] = LittleLong(seamremap[i]);
if (pq1inmodel->numframes < 1 ||
pq1inmodel->numskins < 1 ||
pq1inmodel->numtris < 1 ||
pq1inmodel->numverts < 3 ||
pq1inmodel->skinheight < 1 ||
pq1inmodel->skinwidth < 1)
{
Con_Printf(CON_ERROR "Model %s has an invalid quantity\n", mod->name);
return false;
}
if (qtest)
mod->flags = 0; // Qtest has no flags in header
else
mod->flags = pq1inmodel->flags;
size = sizeof(galiasinfo_t)
#ifndef SERVERONLY
+ pq1inmodel->numskins*sizeof(galiasskin_t)
#endif
+ pq1inmodel->numframes*sizeof(galiasgroup_t);
galias = ZG_Malloc(&loadmodel->memgroup, size);
galias->groupofs = (galiasgroup_t*)(galias+1);
#ifndef SERVERONLY
galias->ofsskins = (galiasskin_t*)(galias->groupofs+pq1inmodel->numframes);
#endif
galias->nextsurf = 0;
loadmodel->numframes = pq1inmodel->numframes;
//skins
skinstart = (daliasskintype_t *)((char*)pq1inmodel+hdrsize);
if( mod->flags & MFH2_HOLEY )
skintranstype = 3; //hexen2
else if( mod->flags & MFH2_TRANSPARENT )
skintranstype = 2; //hexen2
else if( mod->flags & MFH2_SPECIAL_TRANS )
skintranstype = 4; //hexen2
else
skintranstype = 0;
switch(qrenderer)
{
default:
#ifndef SERVERONLY
pinstverts = (dstvert_t *)Q1_LoadSkins_GL(skinstart, skintranstype);
break;
#endif
case QR_NONE:
pinstverts = (dstvert_t *)Q1_LoadSkins_SV(skinstart, skintranstype);
break;
}
if (rapo)
{
/*each triangle can use one coord and one st, for each vert, that's a lot of combinations*/
#ifdef SERVERONLY
/*separate st + vert lists*/
pinh2triangles = (dh2triangle_t *)&pinstverts[pq1inmodel->num_st];
seamremap = BZ_Malloc(sizeof(*seamremap)*pq1inmodel->numtris*3);
galias->numverts = pq1inmodel->numverts;
galias->numindexes = pq1inmodel->numtris*3;
indexes = ZG_Malloc(&loadmodel->memgroup, galias->numindexes*sizeof(*indexes));
galias->ofs_indexes = indexes;
for (i = 0; i < pq1inmodel->numverts; i++)
seamremap[i] = i;
for (i = 0; i < pq1inmodel->numtris; i++)
{
indexes[i*3+0] = LittleShort(pinh2triangles[i].vertindex[0]);
indexes[i*3+1] = LittleShort(pinh2triangles[i].vertindex[1]);
indexes[i*3+2] = LittleShort(pinh2triangles[i].vertindex[2]);
}
#else
int t, v, k;
int *stremap;
/*separate st + vert lists*/
pinh2triangles = (dh2triangle_t *)&pinstverts[pq1inmodel->num_st];
seamremap = BZ_Malloc(sizeof(int)*pq1inmodel->numtris*6);
stremap = seamremap + pq1inmodel->numtris*3;
/*output the indicies as we figure out which verts we want*/
galias->numindexes = pq1inmodel->numtris*3;
indexes = ZG_Malloc(&loadmodel->memgroup, galias->numindexes*sizeof(*indexes));
galias->ofs_indexes = indexes;
for (i = 0; i < pq1inmodel->numtris; i++)
{
for (j = 0; j < 3; j++)
{
v = LittleShort(pinh2triangles[i].vertindex[j]);
t = LittleShort(pinh2triangles[i].stindex[j]);
if (pinstverts[t].onseam && !pinh2triangles[i].facesfront)
t += pq1inmodel->num_st;
for (k = 0; k < galias->numverts; k++) /*big fatoff slow loop*/
{
if (stremap[k] == t && seamremap[k] == v)
break;
}
if (k == galias->numverts)
{
galias->numverts++;
stremap[k] = t;
seamremap[k] = v;
}
indexes[i*3+j] = k;
}
}
st_array = ZG_Malloc(&loadmodel->memgroup, sizeof(*st_array)*(galias->numverts));
galias->ofs_st_array = st_array;
/*generate our st_array now we know which vertexes we want*/
for (k = 0; k < galias->numverts; k++)
{
if (stremap[k] > pq1inmodel->num_st)
{ /*onseam verts? shrink the index, and add half a texture width to the s coord*/
st_array[k][0] = 0.5+(LittleLong(pinstverts[stremap[k]-pq1inmodel->num_st].s)+0.5)/(float)pq1inmodel->skinwidth;
st_array[k][1] = (LittleLong(pinstverts[stremap[k]-pq1inmodel->num_st].t)+0.5)/(float)pq1inmodel->skinheight;
}
else
{
st_array[k][0] = (LittleLong(pinstverts[stremap[k]].s)+0.5)/(float)pq1inmodel->skinwidth;
st_array[k][1] = (LittleLong(pinstverts[stremap[k]].t)+0.5)/(float)pq1inmodel->skinheight;
}
}
#endif
end = &pinh2triangles[pq1inmodel->numtris];
if (Alias_LoadFrameGroup((daliasframetype_t *)end, seamremap, 2) == NULL)
{
BZ_Free(seamremap);
ZG_FreeGroup(&loadmodel->memgroup);
return false;
}
BZ_Free(seamremap);
}
else
{
/*onseam means +=skinwidth/2
verticies that are marked as onseam potentially generate two output verticies.
the triangle chooses which side based upon its 'onseam' field.
*/
//count number of verts that are onseam.
for (onseams=0,i = 0; i < pq1inmodel->numverts; i++)
{
if (pinstverts[i].onseam)
onseams++;
}
seamremap = BZ_Malloc(sizeof(*seamremap)*pq1inmodel->numverts);
galias->numverts = pq1inmodel->numverts+onseams;
//st
#ifndef SERVERONLY
st_array = ZG_Malloc(&loadmodel->memgroup, sizeof(*st_array)*(pq1inmodel->numverts+onseams));
galias->ofs_st_array = st_array;
for (j=pq1inmodel->numverts,i = 0; i < pq1inmodel->numverts; i++)
{
st_array[i][0] = (LittleLong(pinstverts[i].s)+0.5)/(float)pq1inmodel->skinwidth;
st_array[i][1] = (LittleLong(pinstverts[i].t)+0.5)/(float)pq1inmodel->skinheight;
if (pinstverts[i].onseam)
{
st_array[j][0] = st_array[i][0]+0.5;
st_array[j][1] = st_array[i][1];
seamremap[i] = j;
j++;
}
else
seamremap[i] = i;
}
#else
for (i = 0; i < pq1inmodel->numverts; i++)
{
seamremap[i] = i;
}
#endif
//trianglelists;
pinq1triangles = (dtriangle_t *)&pinstverts[pq1inmodel->numverts];
galias->numindexes = pq1inmodel->numtris*3;
indexes = ZG_Malloc(&loadmodel->memgroup, galias->numindexes*sizeof(*indexes));
galias->ofs_indexes = indexes;
for (i=0 ; i<pq1inmodel->numtris ; i++)
{
if (!pinq1triangles[i].facesfront)
{
indexes[i*3+0] = seamremap[LittleLong(pinq1triangles[i].vertindex[0])];
indexes[i*3+1] = seamremap[LittleLong(pinq1triangles[i].vertindex[1])];
indexes[i*3+2] = seamremap[LittleLong(pinq1triangles[i].vertindex[2])];
}
else
{
indexes[i*3+0] = LittleLong(pinq1triangles[i].vertindex[0]);
indexes[i*3+1] = LittleLong(pinq1triangles[i].vertindex[1]);
indexes[i*3+2] = LittleLong(pinq1triangles[i].vertindex[2]);
}
}
end = &pinq1triangles[pq1inmodel->numtris];
//frames
if (Alias_LoadFrameGroup((daliasframetype_t *)end, seamremap, qtest ? 1 : 0) == NULL)
{
BZ_Free(seamremap);
ZG_FreeGroup(&loadmodel->memgroup);
return false;
}
BZ_Free(seamremap);
}
Mod_CompileTriangleNeighbours(galias);
Mod_BuildTextureVectors(galias);
VectorCopy (pq1inmodel->scale_origin, mod->mins);
VectorMA (mod->mins, 255, pq1inmodel->scale, mod->maxs);
mod->type = mod_alias;
Mod_ClampModelSize(mod);
mod->meshinfo = galias;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif
int Mod_ReadFlagsFromMD1(char *name, int md3version)
{
dmdl_t *pinmodel;
char fname[MAX_QPATH];
COM_StripExtension(name, fname, sizeof(fname));
COM_DefaultExtension(fname, ".mdl", sizeof(fname));
if (strcmp(name, fname)) //md3 renamed as mdl
{
COM_StripExtension(name, fname, sizeof(fname)); //seeing as the md3 is named over the mdl,
COM_DefaultExtension(fname, ".md1", sizeof(fname));//read from a file with md1 (one, not an ell)
return 0;
}
pinmodel = (dmdl_t *)COM_LoadTempFile(fname);
if (!pinmodel) //not found
return 0;
if (LittleLong(pinmodel->ident) != IDPOLYHEADER)
return 0;
if (LittleLong(pinmodel->version) != ALIAS_VERSION)
return 0;
return LittleLong(pinmodel->flags);
}
#ifdef MD2MODELS
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
//Q2 model loading
typedef struct
{
float scale[3]; // multiply qbyte verts by this
float translate[3]; // then add this
char name[16]; // frame name from grabbing
dtrivertx_t verts[1]; // variable sized
} dmd2aliasframe_t;
//static galiasinfo_t *galias;
//static md2_t *pq2inmodel;
#define Q2NUMVERTEXNORMALS 162
extern vec3_t bytedirs[Q2NUMVERTEXNORMALS];
static void Q2_LoadSkins(md2_t *pq2inmodel, char *skins)
{
#ifndef SERVERONLY
int i;
shader_t **shaders;
galiasskin_t *outskin = galias->ofsskins;
for (i = 0; i < LittleLong(pq2inmodel->num_skins); i++, outskin++)
{
shaders = ZG_Malloc(&loadmodel->memgroup, sizeof(*shaders));
outskin->ofsshaders = shaders;
outskin->numshaders=1;
COM_CleanUpPath(skins); //blooming tanks.
shaders[0] = R_RegisterSkin(skins, loadmodel->name);
TEXASSIGN(shaders[0]->defaulttextures.base, R_LoadReplacementTexture(skins, "models", IF_NOALPHA));
R_BuildDefaultTexnums(NULL, shaders[0]);
outskin->skinwidth = 0;
outskin->skinheight = 0;
outskin->skinspeed = 0;
skins += MD2MAX_SKINNAME;
}
#endif
galias->numskins = LittleLong(pq2inmodel->num_skins);
/*
#ifndef SERVERONLY
outskin = (galiasskin_t *)((char *)galias + galias->ofsskins);
outskin += galias->numskins - 1;
if (galias->numskins)
{
if (*(shader_t**)((char *)outskin + outskin->ofstexnums))
return;
galias->numskins--;
}
#endif
*/
}
#define MD2_MAX_TRIANGLES 4096
qboolean QDECL Mod_LoadQ2Model (model_t *mod, void *buffer)
{
#ifndef SERVERONLY
dmd2stvert_t *pinstverts;
vec2_t *st_array;
vec3_t *normals;
#endif
md2_t *pq2inmodel;
int version;
int i, j;
dmd2triangle_t *pintri;
index_t *indexes;
int numindexes;
vec3_t min;
vec3_t max;
galiaspose_t *pose;
galiasgroup_t *poutframe;
dmd2aliasframe_t *pinframe;
int framesize;
vecV_t *verts;
int indremap[MD2_MAX_TRIANGLES*3];
unsigned short ptempindex[MD2_MAX_TRIANGLES*3], ptempstindex[MD2_MAX_TRIANGLES*3];
int numverts;
int size;
loadmodel=mod;
loadmodel->engineflags |= MDLF_NEEDOVERBRIGHT;
pq2inmodel = (md2_t *)buffer;
version = LittleLong (pq2inmodel->version);
if (version != MD2ALIAS_VERSION)
{
Con_Printf (CON_ERROR "%s has wrong version number (%i should be %i)\n",
mod->name, version, MD2ALIAS_VERSION);
return false;
}
if (LittleLong(pq2inmodel->num_frames) < 1 ||
LittleLong(pq2inmodel->num_skins) < 0 ||
LittleLong(pq2inmodel->num_tris) < 1 ||
LittleLong(pq2inmodel->num_xyz) < 3 ||
LittleLong(pq2inmodel->num_st) < 3 ||
LittleLong(pq2inmodel->skinheight) < 1 ||
LittleLong(pq2inmodel->skinwidth) < 1)
{
Con_Printf(CON_ERROR "Model %s has an invalid quantity\n", mod->name);
return false;
}
mod->flags = 0;
loadmodel->numframes = LittleLong(pq2inmodel->num_frames);
size = sizeof(galiasinfo_t)
#ifndef SERVERONLY
+ LittleLong(pq2inmodel->num_skins)*sizeof(galiasskin_t)
#endif
+ LittleLong(pq2inmodel->num_frames)*sizeof(galiasgroup_t);
galias = ZG_Malloc(&loadmodel->memgroup, size);
galias->groupofs = (galiasgroup_t*)(galias+1);
#ifndef SERVERONLY
galias->ofsskins = (galiasskin_t*)(galias->groupofs + LittleLong(pq2inmodel->num_frames));
#endif
galias->nextsurf = 0;
//skins
Q2_LoadSkins(pq2inmodel, ((char *)pq2inmodel+LittleLong(pq2inmodel->ofs_skins)));
//trianglelists;
pintri = (dmd2triangle_t *)((char *)pq2inmodel + LittleLong(pq2inmodel->ofs_tris));
for (i=0 ; i<LittleLong(pq2inmodel->num_tris) ; i++, pintri++)
{
for (j=0 ; j<3 ; j++)
{
ptempindex[i*3+j] = ( unsigned short )LittleShort ( pintri->xyz_index[j] );
ptempstindex[i*3+j] = ( unsigned short )LittleShort ( pintri->st_index[j] );
}
}
numindexes = galias->numindexes = LittleLong(pq2inmodel->num_tris)*3;
indexes = ZG_Malloc(&loadmodel->memgroup, galias->numindexes*sizeof(*indexes));
galias->ofs_indexes = indexes;
memset ( indremap, -1, sizeof(indremap) );
numverts=0;
for ( i = 0; i < numindexes; i++ )
{
if ( indremap[i] != -1 ) {
continue;
}
for ( j = 0; j < numindexes; j++ )
{
if ( j == i ) {
continue;
}
if ( (ptempindex[i] == ptempindex[j]) && (ptempstindex[i] == ptempstindex[j]) ) {
indremap[j] = i;
}
}
}
// count unique vertexes
for ( i = 0; i < numindexes; i++ )
{
if ( indremap[i] != -1 ) {
continue;
}
indexes[i] = numverts++;
indremap[i] = i;
}
Con_DPrintf ( "%s: remapped %i verts to %i\n", mod->name, LittleLong(pq2inmodel->num_xyz), numverts );
galias->numverts = numverts;
// remap remaining indexes
for ( i = 0; i < numindexes; i++ )
{
if ( indremap[i] != i ) {
indexes[i] = indexes[indremap[i]];
}
}
// s and t vertices
#ifndef SERVERONLY
pinstverts = ( dmd2stvert_t * ) ( ( qbyte * )pq2inmodel + LittleLong (pq2inmodel->ofs_st) );
st_array = ZG_Malloc(&loadmodel->memgroup, sizeof(*st_array)*(numverts));
galias->ofs_st_array = st_array;
for (j=0 ; j<numindexes; j++)
{
st_array[indexes[j]][0] = (float)(((double)LittleShort (pinstverts[ptempstindex[indremap[j]]].s) + 0.5f) /LittleLong(pq2inmodel->skinwidth));
st_array[indexes[j]][1] = (float)(((double)LittleShort (pinstverts[ptempstindex[indremap[j]]].t) + 0.5f) /LittleLong(pq2inmodel->skinheight));
}
#endif
//frames
ClearBounds ( mod->mins, mod->maxs );
poutframe = galias->groupofs;
framesize = LittleLong (pq2inmodel->framesize);
for (i=0 ; i<LittleLong(pq2inmodel->num_frames) ; i++)
{
size = sizeof(galiaspose_t) + sizeof(vecV_t)*numverts;
#ifndef SERVERONLY
size += 3*sizeof(vec3_t)*numverts;
#endif
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, size);
poutframe->poseofs = pose;
poutframe->numposes = 1;
galias->groups++;
verts = (vecV_t *)(pose+1);
pose->ofsverts = verts;
#ifndef SERVERONLY
normals = (vec3_t*)&verts[galias->numverts];
pose->ofsnormals = normals;
pose->ofssvector = &normals[galias->numverts];
pose->ofstvector = &normals[galias->numverts*2];
#endif
pinframe = ( dmd2aliasframe_t * )( ( qbyte * )pq2inmodel + LittleLong (pq2inmodel->ofs_frames) + i * framesize );
Q_strncpyz(poutframe->name, pinframe->name, sizeof(poutframe->name));
for (j=0 ; j<3 ; j++)
{
pose->scale[j] = LittleFloat (pinframe->scale[j]);
pose->scale_origin[j] = LittleFloat (pinframe->translate[j]);
}
for (j=0 ; j<numindexes; j++)
{
// verts are all 8 bit, so no swapping needed
verts[indexes[j]][0] = pose->scale_origin[0]+pose->scale[0]*pinframe->verts[ptempindex[indremap[j]]].v[0];
verts[indexes[j]][1] = pose->scale_origin[1]+pose->scale[1]*pinframe->verts[ptempindex[indremap[j]]].v[1];
verts[indexes[j]][2] = pose->scale_origin[2]+pose->scale[2]*pinframe->verts[ptempindex[indremap[j]]].v[2];
#ifndef SERVERONLY
VectorCopy(bytedirs[pinframe->verts[ptempindex[indremap[j]]].lightnormalindex], normals[indexes[j]]);
#endif
}
// Mod_AliasCalculateVertexNormals ( numindexes, poutindex, numverts, poutvertex, qfalse );
VectorCopy ( pose->scale_origin, min );
VectorMA ( pose->scale_origin, 255, pose->scale, max );
// poutframe->radius = RadiusFromBounds ( min, max );
// mod->radius = max ( mod->radius, poutframe->radius );
AddPointToBounds ( min, mod->mins, mod->maxs );
AddPointToBounds ( max, mod->mins, mod->maxs );
// GL_GenerateNormals((float*)verts, (float*)normals, indexes, numindexes/3, numverts);
poutframe++;
}
Mod_CompileTriangleNeighbours(galias);
Mod_BuildTextureVectors(galias);
/*
VectorCopy (pq2inmodel->scale_origin, mod->mins);
VectorMA (mod->mins, 255, pq2inmodel->scale, mod->maxs);
*/
Mod_ClampModelSize(mod);
mod->meshinfo = galias;
mod->type = mod_alias;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif
int Mod_GetNumBones(model_t *model, qboolean allowtags)
{
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
#ifdef SKELETALMODELS
if (inf->numbones)
return inf->numbones;
else
#endif
if (allowtags)
return inf->numtags;
else
return 0;
}
int Mod_GetBoneRelations(model_t *model, int firstbone, int lastbone, framestate_t *fstate, float *result)
{
#ifdef SKELETALMODELS
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return false;
inf = Mod_Extradata(model);
return Alias_GetBoneRelations(inf, fstate, result, firstbone, lastbone);
#endif
return 0;
}
galiasbone_t *Mod_GetBoneInfo(model_t *model, int *numbones)
{
#ifdef SKELETALMODELS
galiasbone_t *bone;
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return NULL;
inf = Mod_Extradata(model);
bone = inf->ofsbones;
*numbones = inf->numbones;
return bone;
#else
*numbones = 0;
return NULL;
#endif
}
int Mod_GetBoneParent(model_t *model, int bonenum)
{
#ifdef SKELETALMODELS
galiasbone_t *bone;
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
bonenum--;
if ((unsigned int)bonenum >= inf->numbones)
return 0; //no parent
bone = inf->ofsbones;
return bone[bonenum].parent+1;
#endif
return 0;
}
char *Mod_GetBoneName(model_t *model, int bonenum)
{
#ifdef SKELETALMODELS
galiasbone_t *bone;
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
bonenum--;
if ((unsigned int)bonenum >= inf->numbones)
return 0; //no parent
bone = inf->ofsbones;
return bone[bonenum].name;
#endif
return 0;
}
qboolean Mod_GetTag(model_t *model, int tagnum, framestate_t *fstate, float *result)
{
galiasinfo_t *inf;
if (!model || model->type != mod_alias)
return false;
inf = Mod_Extradata(model);
#ifdef SKELETALMODELS
if (inf->numbones)
{
galiasbone_t *bone;
galiasgroup_t *g1, *g2;
float tempmatrix[12]; //flipped between this and bonematrix
float *matrix; //the matrix for a single bone in a single pose.
float m[12]; //combined interpolated version of 'matrix'.
int b, k; //counters
float *pose[4]; //the per-bone matricies (one for each pose)
float plerp[4]; //the ammount of that pose to use (must combine to 1)
int numposes = 0;
int frame1, frame2;
float f1time, f2time;
float f2ness;
#ifdef warningmsg
#pragma warningmsg("fixme: no baseframe info")
#endif
if (tagnum <= 0 || tagnum > inf->numbones)
return false;
tagnum--; //tagnum 0 is 'use my angles/org'
bone = inf->ofsbones;
if (fstate->bonestate)
{
if (tagnum >= fstate->bonecount)
return false;
if (fstate->boneabs)
{
memcpy(result, fstate->bonestate + 12 * tagnum, 12*sizeof(*result));
return true;
}
pose[0] = fstate->bonestate;
plerp[0] = 1;
numposes = 1;
}
else
{
frame1 = fstate->g[FS_REG].frame[0];
frame2 = fstate->g[FS_REG].frame[1];
f1time = fstate->g[FS_REG].frametime[0];
f2time = fstate->g[FS_REG].frametime[1];
f2ness = fstate->g[FS_REG].lerpfrac;
if (frame1 < 0 || frame1 >= inf->groups)
return false;
if (frame2 < 0 || frame2 >= inf->groups)
{
f2ness = 0;
frame2 = frame1;
}
//the higher level merges old/new anims, but we still need to blend between automated frame-groups.
g1 = &inf->groupofs[frame1];
g2 = &inf->groupofs[frame2];
if (f2ness != 1)
{
f1time *= g1->rate;
frame1 = (int)f1time%g1->numposes;
frame2 = ((int)f1time+1)%g1->numposes;
f1time = f1time - (int)f1time;
pose[numposes] = g1->boneofs + inf->numbones*12*frame1;
plerp[numposes] = (1-f1time) * (1-f2ness);
numposes++;
if (frame1 != frame2)
{
pose[numposes] = g1->boneofs + inf->numbones*12*frame2;
plerp[numposes] = f1time * (1-f2ness);
numposes++;
}
}
if (f2ness)
{
f2time *= g2->rate;
frame1 = (int)f2time%g2->numposes;
frame2 = ((int)f2time+1)%g2->numposes;
f2time = f2time - (int)f2time;
pose[numposes] = g2->boneofs + inf->numbones*12*frame1;
plerp[numposes] = (1-f2time) * f2ness;
numposes++;
if (frame1 != frame2)
{
pose[numposes] = g2->boneofs + inf->numbones*12*frame2;
plerp[numposes] = f2time * f2ness;
numposes++;
}
}
}
//set up the identity matrix
for (k = 0;k < 12;k++)
result[k] = 0;
result[0] = 1;
result[5] = 1;
result[10] = 1;
while(tagnum >= 0)
{
//set up the per-bone transform matrix
for (k = 0;k < 12;k++)
m[k] = 0;
for (b = 0;b < numposes;b++)
{
matrix = pose[b] + tagnum*12;
for (k = 0;k < 12;k++)
m[k] += matrix[k] * plerp[b];
}
memcpy(tempmatrix, result, sizeof(tempmatrix));
R_ConcatTransforms((void*)m, (void*)tempmatrix, (void*)result);
tagnum = bone[tagnum].parent;
}
return true;
}
#endif
if (inf->numtags)
{
md3tag_t *t1, *t2;
int frame1, frame2;
//float f1time, f2time; //tags/md3s don't support framegroups.
float f2ness;
frame1 = fstate->g[FS_REG].frame[0];
frame2 = fstate->g[FS_REG].frame[1];
//f1time = fstate->g[FS_REG].frametime[0];
//f2time = fstate->g[FS_REG].frametime[1];
f2ness = fstate->g[FS_REG].lerpfrac;
if (tagnum <= 0 || tagnum > inf->numtags)
return false;
if (frame1 < 0)
return false;
if (frame1 >= inf->numtagframes)
frame1 = inf->numtagframes - 1;
if (frame2 < 0 || frame2 >= inf->numtagframes)
frame2 = frame1;
tagnum--; //tagnum 0 is 'use my angles/org'
t1 = inf->ofstags;
t1 += tagnum;
t1 += inf->numtags*frame1;
t2 = inf->ofstags;
t2 += tagnum;
t2 += inf->numtags*frame2;
if (t1 == t2)
{
result[0] = t1->ang[0][0];
result[1] = t1->ang[0][1];
result[2] = t1->ang[0][2];
result[3] = t1->org[0];
result[4] = t1->ang[1][0];
result[5] = t1->ang[1][1];
result[6] = t1->ang[1][2];
result[7] = t1->org[1];
result[8] = t1->ang[2][0];
result[9] = t1->ang[2][1];
result[10] = t1->ang[2][2];
result[11] = t1->org[2];
}
else
{
float f1ness = 1-f2ness;
result[0] = t1->ang[0][0]*f1ness + t2->ang[0][0]*f2ness;
result[1] = t1->ang[0][1]*f1ness + t2->ang[0][1]*f2ness;
result[2] = t1->ang[0][2]*f1ness + t2->ang[0][2]*f2ness;
result[3] = t1->org[0]*f1ness + t2->org[0]*f2ness;
result[4] = t1->ang[1][0]*f1ness + t2->ang[1][0]*f2ness;
result[5] = t1->ang[1][1]*f1ness + t2->ang[1][1]*f2ness;
result[6] = t1->ang[1][2]*f1ness + t2->ang[1][2]*f2ness;
result[7] = t1->org[1]*f1ness + t2->org[1]*f2ness;
result[8] = t1->ang[2][0]*f1ness + t2->ang[2][0]*f2ness;
result[9] = t1->ang[2][1]*f1ness + t2->ang[2][1]*f2ness;
result[10] = t1->ang[2][2]*f1ness + t2->ang[2][2]*f2ness;
result[11] = t1->org[2]*f1ness + t2->org[2]*f2ness;
}
VectorNormalize(result);
VectorNormalize(result+4);
VectorNormalize(result+8);
return true;
}
return false;
}
int Mod_TagNumForName(model_t *model, char *name)
{
int i;
galiasinfo_t *inf;
md3tag_t *t;
if (!model)
return 0;
#ifdef HALFLIFEMODELS
if (model->type == mod_halflife)
return HLMod_BoneForName(model, name);
#endif
if (model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
#ifdef SKELETALMODELS
if (inf->numbones)
{
galiasbone_t *b;
b = inf->ofsbones;
for (i = 0; i < inf->numbones; i++)
{
if (!strcmp(b[i].name, name))
return i+1;
}
}
#endif
t = inf->ofstags;
for (i = 0; i < inf->numtags; i++)
{
if (!strcmp(t[i].name, name))
return i+1;
}
return 0;
}
int Mod_FrameNumForName(model_t *model, char *name)
{
galiasgroup_t *group;
galiasinfo_t *inf;
int i;
if (!model)
return -1;
#ifdef HALFLIFEMODELS
if (model->type == mod_halflife)
return HLMod_FrameForName(model, name);
#endif
if (model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
group = inf->groupofs;
for (i = 0; i < inf->groups; i++, group++)
{
if (!strcmp(group->name, name))
return i;
}
return -1;
}
#ifndef SERVERONLY
int Mod_SkinNumForName(model_t *model, char *name)
{
int i;
galiasinfo_t *inf;
galiasskin_t *skin;
if (!model || model->type != mod_alias)
return -1;
inf = Mod_Extradata(model);
skin = inf->ofsskins;
for (i = 0; i < inf->numskins; i++, skin++)
{
if (!strcmp(skin->name, name))
return i;
}
return -1;
}
#endif
const char *Mod_FrameNameForNum(model_t *model, int num)
{
galiasgroup_t *group;
galiasinfo_t *inf;
if (!model)
return NULL;
if (model->type != mod_alias)
return NULL;
inf = Mod_Extradata(model);
if (num >= inf->groups)
return NULL;
group = inf->groupofs;
return group[num].name;
}
qboolean Mod_FrameInfoForNum(model_t *model, int num, char **name, int *numframes, float *duration, qboolean *loop)
{
galiasgroup_t *group;
galiasinfo_t *inf;
if (!model)
return false;
if (model->type != mod_alias)
return false;
inf = Mod_Extradata(model);
if (num >= inf->groups)
return false;
group = inf->groupofs;
*name = group[num].name;
*numframes = group[num].numposes;
*loop = group[num].loop;
*duration = group->numposes/group->rate;
return true;
}
const char *Mod_SkinNameForNum(model_t *model, int num)
{
#ifdef SERVERONLY
return NULL;
#else
galiasinfo_t *inf;
galiasskin_t *skin;
if (!model || model->type != mod_alias)
return NULL;
inf = Mod_Extradata(model);
if (num >= inf->numskins)
return NULL;
skin = inf->ofsskins;
return skin[num].name;
#endif
}
float Mod_GetFrameDuration(model_t *model, int frameno)
{
galiasinfo_t *inf;
galiasgroup_t *group;
if (!model || model->type != mod_alias)
return 0;
inf = Mod_Extradata(model);
group = inf->groupofs;
if (frameno < 0 || frameno >= inf->groups)
return 0;
group += frameno;
return group->numposes/group->rate;
}
#ifdef MD3MODELS
//structures from Tenebrae
typedef struct {
int ident;
int version;
char name[64];
int flags; //Does anyone know what these are?
int numFrames;
int numTags;
int numSurfaces;
int numSkins;
int ofsFrames;
int ofsTags;
int ofsSurfaces;
int ofsEnd;
} md3Header_t;
//then has header->numFrames of these at header->ofs_Frames
typedef struct md3Frame_s {
vec3_t bounds[2];
vec3_t localOrigin;
float radius;
char name[16];
} md3Frame_t;
//there are header->numSurfaces of these at header->ofsSurfaces, following from ofsEnd
typedef struct {
int ident; //
char name[64]; // polyset name
int flags;
int numFrames; // all surfaces in a model should have the same
int numShaders; // all surfaces in a model should have the same
int numVerts;
int numTriangles;
int ofsTriangles;
int ofsShaders; // offset from start of md3Surface_t
int ofsSt; // texture coords are common for all frames
int ofsXyzNormals; // numVerts * numFrames
int ofsEnd; // next surface follows
} md3Surface_t;
//at surf+surf->ofsXyzNormals
typedef struct {
short xyz[3];
qbyte latlong[2];
} md3XyzNormal_t;
//surf->numTriangles at surf+surf->ofsTriangles
typedef struct {
int indexes[3];
} md3Triangle_t;
//surf->numVerts at surf+surf->ofsSt
typedef struct {
float s;
float t;
} md3St_t;
typedef struct {
char name[64];
int shaderIndex;
} md3Shader_t;
//End of Tenebrae 'assistance'
qboolean QDECL Mod_LoadQ3Model(model_t *mod, void *buffer)
{
#ifndef SERVERONLY
galiasskin_t *skin;
shader_t **shaders;
float lat, lng;
md3St_t *inst;
vec3_t *normals;
vec3_t *svector;
vec3_t *tvector;
vec2_t *st_array;
md3Shader_t *inshader;
#endif
// int version;
int s, i, j, d;
index_t *indexes;
vec3_t min;
vec3_t max;
galiaspose_t *pose;
galiasinfo_t *parent, *root;
galiasgroup_t *group;
vecV_t *verts;
md3Triangle_t *intris;
md3XyzNormal_t *invert;
int size;
int externalskins;
md3Header_t *header;
md3Surface_t *surf;
loadmodel=mod;
header = buffer;
// if (header->version != sdfs)
// Sys_Error("GL_LoadQ3Model: Bad version\n");
parent = NULL;
root = NULL;
#ifndef SERVERONLY
externalskins = Mod_BuildSkinFileList(mod->name);
#else
externalskins = 0;
#endif
min[0] = min[1] = min[2] = 0;
max[0] = max[1] = max[2] = 0;
surf = (md3Surface_t *)((qbyte *)header + LittleLong(header->ofsSurfaces));
for (s = 0; s < LittleLong(header->numSurfaces); s++)
{
if (LittleLong(surf->ident) != MD3_IDENT)
Con_Printf(CON_WARNING "Warning: md3 sub-surface doesn't match ident\n");
size = sizeof(galiasinfo_t) + sizeof(galiasgroup_t)*LittleLong(header->numFrames);
galias = ZG_Malloc(&loadmodel->memgroup, size);
galias->groupofs = (galiasgroup_t*)(galias+1); //frame groups
galias->groups = LittleLong(header->numFrames);
galias->numverts = LittleLong(surf->numVerts);
galias->numindexes = LittleLong(surf->numTriangles)*3;
galias->shares_verts = s;
if (parent)
parent->nextsurf = galias;
else
root = galias;
parent = galias;
#ifndef SERVERONLY
st_array = ZG_Malloc(&loadmodel->memgroup, sizeof(vec2_t)*galias->numindexes);
galias->ofs_st_array = st_array;
inst = (md3St_t*)((qbyte*)surf + LittleLong(surf->ofsSt));
for (i = 0; i < galias->numverts; i++)
{
st_array[i][0] = LittleFloat(inst[i].s);
st_array[i][1] = LittleFloat(inst[i].t);
}
#endif
indexes = ZG_Malloc(&loadmodel->memgroup, sizeof(*indexes)*galias->numindexes);
galias->ofs_indexes = indexes;
intris = (md3Triangle_t *)((qbyte*)surf + LittleLong(surf->ofsTriangles));
for (i = 0; i < LittleLong(surf->numTriangles); i++)
{
indexes[i*3+0] = LittleLong(intris[i].indexes[0]);
indexes[i*3+1] = LittleLong(intris[i].indexes[1]);
indexes[i*3+2] = LittleLong(intris[i].indexes[2]);
}
group = (galiasgroup_t *)(galias+1);
invert = (md3XyzNormal_t *)((qbyte*)surf + LittleLong(surf->ofsXyzNormals));
for (i = 0; i < LittleLong(surf->numFrames); i++)
{
int size = sizeof(galiaspose_t) + sizeof(vecV_t)*LittleLong(surf->numVerts);
#ifndef SERVERONLY
size += 3*sizeof(vec3_t)*LittleLong(surf->numVerts);
#endif
pose = (galiaspose_t *)ZG_Malloc(&loadmodel->memgroup, size);
verts = (vecV_t*)(pose+1);
pose->ofsverts = verts;
#ifndef SERVERONLY
normals = (vec3_t*)(verts + LittleLong(surf->numVerts));
pose->ofsnormals = normals;
svector = normals + LittleLong(surf->numVerts);
pose->ofssvector = svector;
tvector = svector + LittleLong(surf->numVerts);
pose->ofstvector = tvector;
#endif
for (j = 0; j < LittleLong(surf->numVerts); j++)
{
#ifndef SERVERONLY
lat = (float)invert[j].latlong[0] * (2 * M_PI)*(1.0 / 255.0);
lng = (float)invert[j].latlong[1] * (2 * M_PI)*(1.0 / 255.0);
normals[j][0] = cos ( lng ) * sin ( lat );
normals[j][1] = sin ( lng ) * sin ( lat );
normals[j][2] = cos ( lat );
#endif
for (d = 0; d < 3; d++)
{
verts[j][d] = LittleShort(invert[j].xyz[d])/64.0f;
if (verts[j][d]<min[d])
min[d] = verts[j][d];
if (verts[j][d]>max[d])
max[d] = verts[j][d];
}
}
pose->scale[0] = 1;
pose->scale[1] = 1;
pose->scale[2] = 1;
pose->scale_origin[0] = 0;
pose->scale_origin[1] = 0;
pose->scale_origin[2] = 0;
snprintf(group->name, sizeof(group->name)-1, "frame%i", i);
group->numposes = 1;
group->rate = 1;
group->poseofs = pose;
group++;
invert += LittleLong(surf->numVerts);
}
#ifndef SERVERONLY
if (externalskins<LittleLong(surf->numShaders))
externalskins = LittleLong(surf->numShaders);
if (externalskins)
{
char shadname[1024];
skin = ZG_Malloc(&loadmodel->memgroup, (LittleLong(surf->numShaders)+externalskins)*((sizeof(galiasskin_t)+sizeof(shader_t*))));
galias->ofsskins = skin;
shaders = (shader_t **)(skin + LittleLong(surf->numShaders)+externalskins);
inshader = (md3Shader_t *)((qbyte *)surf + LittleLong(surf->ofsShaders));
for (i = 0; i < externalskins; i++)
{
skin->numshaders = 1;
skin->ofsshaders = &shaders[i];
skin->ofstexels = 0;
skin->skinwidth = 0;
skin->skinheight = 0;
skin->skinspeed = 0;
shadname[0] = '\0';
Mod_ParseQ3SkinFile(shadname, surf->name, loadmodel->name, i, skin->name);
if (!*shadname)
{
if (i >= LittleLong(surf->numShaders) || !*inshader->name)
strcpy(shadname, "missingskin"); //this shouldn't be possible
else
strcpy(shadname, inshader->name);
Q_strncpyz(skin->name, shadname, sizeof(skin->name));
}
if (qrenderer != QR_NONE)
{
shaders[i] = R_RegisterSkin(shadname, mod->name);
R_BuildDefaultTexnums(NULL, shaders[i]);
if (shaders[i]->flags & SHADER_NOIMAGE)
Con_Printf("Unable to load texture for shader \"%s\" for model \"%s\"\n", shaders[i]->name, loadmodel->name);
}
inshader++;
skin++;
}
galias->numskins = i;
}
#endif
VectorCopy(min, loadmodel->mins);
VectorCopy(max, loadmodel->maxs);
Mod_CompileTriangleNeighbours (galias);
Mod_BuildTextureVectors(galias);
surf = (md3Surface_t *)((qbyte *)surf + LittleLong(surf->ofsEnd));
}
if (!root)
root = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasinfo_t));
root->numtagframes = LittleLong(header->numFrames);
root->numtags = LittleLong(header->numTags);
root->ofstags = ZG_Malloc(&loadmodel->memgroup, LittleLong(header->numTags)*sizeof(md3tag_t)*LittleLong(header->numFrames));
{
md3tag_t *src;
md3tag_t *dst;
src = (md3tag_t *)((char*)header+LittleLong(header->ofsTags));
dst = root->ofstags;
for(i=0;i<LittleLong(header->numTags)*LittleLong(header->numFrames);i++)
{
memcpy(dst->name, src->name, sizeof(dst->name));
for(j=0;j<3;j++)
{
dst->org[j] = LittleFloat(src->org[j]);
}
for(j=0;j<3;j++)
{
for(s=0;s<3;s++)
{
dst->ang[j][s] = LittleFloat(src->ang[j][s]);
}
}
src++;
dst++;
}
}
#ifndef SERVERONLY
if (mod_md3flags.value)
mod->flags = LittleLong(header->flags);
else
#endif
mod->flags = 0;
if (!mod->flags)
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0);
Mod_ClampModelSize(mod);
mod->type = mod_alias;
mod->meshinfo = root;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif
#ifdef ZYMOTICMODELS
typedef struct zymlump_s
{
int start;
int length;
} zymlump_t;
typedef struct zymtype1header_s
{
char id[12]; // "ZYMOTICMODEL", length 12, no termination
int type; // 0 (vertex morph) 1 (skeletal pose) or 2 (skeletal scripted)
int filesize; // size of entire model file
float mins[3], maxs[3], radius; // for clipping uses
int numverts;
int numtris;
int numsurfaces;
int numbones; // this may be zero in the vertex morph format (undecided)
int numscenes; // 0 in skeletal scripted models
// skeletal pose header
// lump offsets are relative to the file
zymlump_t lump_scenes; // zymscene_t scene[numscenes]; // name and other information for each scene (see zymscene struct)
zymlump_t lump_poses; // float pose[numposes][numbones][6]; // animation data
zymlump_t lump_bones; // zymbone_t bone[numbones];
zymlump_t lump_vertbonecounts; // int vertbonecounts[numvertices]; // how many bones influence each vertex (separate mainly to make this compress better)
zymlump_t lump_verts; // zymvertex_t vert[numvertices]; // see vertex struct
zymlump_t lump_texcoords; // float texcoords[numvertices][2];
zymlump_t lump_render; // int renderlist[rendersize]; // sorted by shader with run lengths (int count), shaders are sequentially used, each run can be used with glDrawElements (each triangle is 3 int indices)
zymlump_t lump_surfnames; // char shadername[numsurfaces][32]; // shaders used on this model
zymlump_t lump_trizone; // byte trizone[numtris]; // see trizone explanation
} zymtype1header_t;
typedef struct zymbone_s
{
char name[32];
int flags;
int parent; // parent bone number
} zymbone_t;
typedef struct zymscene_s
{
char name[32];
float mins[3], maxs[3], radius; // for clipping
float framerate; // the scene will animate at this framerate (in frames per second)
int flags;
int start, length; // range of poses
} zymscene_t;
#define ZYMSCENEFLAG_NOLOOP 1
typedef struct zymvertex_s
{
int bonenum;
float origin[3];
} zymvertex_t;
//this can generate multiple meshes (one for each shader).
//but only one set of transforms are ever generated.
qboolean QDECL Mod_LoadZymoticModel(model_t *mod, void *buffer)
{
#ifndef SERVERONLY
galiasskin_t *skin;
shader_t **shaders;
int skinfiles;
int j;
#endif
int i;
zymtype1header_t *header;
galiasinfo_t *root;
galisskeletaltransforms_t *transforms;
zymvertex_t *intrans;
galiasbone_t *bone;
zymbone_t *inbone;
int v;
float multiplier;
float *matrix, *inmatrix;
vec2_t *stcoords;
vec2_t *inst;
int *vertbonecounts;
galiasgroup_t *grp;
zymscene_t *inscene;
int *renderlist, count;
index_t *indexes;
char *surfname;
loadmodel=mod;
header = buffer;
if (memcmp(header->id, "ZYMOTICMODEL", 12))
{
Con_Printf("Mod_LoadZymoticModel: %s, doesn't appear to BE a zymotic!\n", mod->name);
return false;
}
if (BigLong(header->type) != 1)
{
Con_Printf("Mod_LoadZymoticModel: %s, only type 1 is supported\n", mod->name);
return false;
}
for (i = 0; i < sizeof(zymtype1header_t)/4; i++)
((int*)header)[i] = BigLong(((int*)header)[i]);
if (!header->numverts)
{
Con_Printf("Mod_LoadZymoticModel: %s, no vertexes\n", mod->name);
return false;
}
if (!header->numsurfaces)
{
Con_Printf("Mod_LoadZymoticModel: %s, no surfaces\n", mod->name);
return false;
}
VectorCopy(header->mins, mod->mins);
VectorCopy(header->maxs, mod->maxs);
root = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasinfo_t)*header->numsurfaces);
root->numswtransforms = header->lump_verts.length/sizeof(zymvertex_t);
transforms = ZG_Malloc(&loadmodel->memgroup, root->numswtransforms*sizeof(*transforms));
root->ofsswtransforms = transforms;
vertbonecounts = (int *)((char*)header + header->lump_vertbonecounts.start);
intrans = (zymvertex_t *)((char*)header + header->lump_verts.start);
vertbonecounts[0] = BigLong(vertbonecounts[0]);
multiplier = 1.0f / vertbonecounts[0];
for (i = 0, v=0; i < root->numswtransforms; i++)
{
while(!vertbonecounts[v])
{
v++;
if (v == header->numverts)
{
Con_Printf("Mod_LoadZymoticModel: %s, too many transformations\n", mod->name);
return false;
}
vertbonecounts[v] = BigLong(vertbonecounts[v]);
multiplier = 1.0f / vertbonecounts[v];
}
transforms[i].vertexindex = v;
transforms[i].boneindex = BigLong(intrans[i].bonenum);
transforms[i].org[0] = multiplier*BigFloat(intrans[i].origin[0]);
transforms[i].org[1] = multiplier*BigFloat(intrans[i].origin[1]);
transforms[i].org[2] = multiplier*BigFloat(intrans[i].origin[2]);
transforms[i].org[3] = multiplier*1;
vertbonecounts[v]--;
}
if (intrans != (zymvertex_t *)((char*)header + header->lump_verts.start))
{
Con_Printf(CON_ERROR "%s, Vertex transforms list appears corrupt.\n", mod->name);
return false;
}
if (vertbonecounts != (int *)((char*)header + header->lump_vertbonecounts.start))
{
Con_Printf(CON_ERROR "%s, Vertex bone counts list appears corrupt.\n", mod->name);
return false;
}
root->numverts = v+1;
root->numbones = header->numbones;
bone = ZG_Malloc(&loadmodel->memgroup, root->numswtransforms*sizeof(*transforms));
inbone = (zymbone_t*)((char*)header + header->lump_bones.start);
for (i = 0; i < root->numbones; i++)
{
Q_strncpyz(bone[i].name, inbone[i].name, sizeof(bone[i].name));
bone[i].parent = BigLong(inbone[i].parent);
}
root->ofsbones = bone;
renderlist = (int*)((char*)header + header->lump_render.start);
for (i = 0;i < header->numsurfaces; i++)
{
count = BigLong(*renderlist++);
count *= 3;
indexes = ZG_Malloc(&loadmodel->memgroup, count*sizeof(*indexes));
root[i].ofs_indexes = indexes;
root[i].numindexes = count;
while(count)
{ //invert
indexes[count-1] = BigLong(renderlist[count-3]);
indexes[count-2] = BigLong(renderlist[count-2]);
indexes[count-3] = BigLong(renderlist[count-1]);
count-=3;
}
renderlist += root[i].numindexes;
}
if (renderlist != (int*)((char*)header + header->lump_render.start + header->lump_render.length))
{
Con_Printf(CON_ERROR "%s, render list appears corrupt.\n", mod->name);
return false;
}
grp = ZG_Malloc(&loadmodel->memgroup, sizeof(*grp)*header->numscenes*header->numsurfaces);
matrix = ZG_Malloc(&loadmodel->memgroup, header->lump_poses.length);
inmatrix = (float*)((char*)header + header->lump_poses.start);
for (i = 0; i < header->lump_poses.length/4; i++)
matrix[i] = BigFloat(inmatrix[i]);
inscene = (zymscene_t*)((char*)header + header->lump_scenes.start);
surfname = ((char*)header + header->lump_surfnames.start);
stcoords = ZG_Malloc(&loadmodel->memgroup, root[0].numverts*sizeof(vec2_t));
inst = (vec2_t *)((char *)header + header->lump_texcoords.start);
for (i = 0; i < header->lump_texcoords.length/8; i++)
{
stcoords[i][0] = BigFloat(inst[i][0]);
stcoords[i][1] = 1-BigFloat(inst[i][1]); //hmm. upside down skin coords?
}
#ifndef SERVERONLY
skinfiles = Mod_BuildSkinFileList(loadmodel->name);
if (skinfiles < 1)
skinfiles = 1;
#endif
for (i = 0; i < header->numsurfaces; i++, surfname+=32)
{
root[i].groups = header->numscenes;
root[i].groupofs = grp;
#ifdef SERVERONLY
root[i].numskins = 1;
#else
root[i].ofs_st_array = stcoords;
root[i].numskins = skinfiles;
skin = ZG_Malloc(&loadmodel->memgroup, (sizeof(galiasskin_t)+sizeof(shader_t*))*skinfiles);
shaders = (shader_t**)(skin+skinfiles);
for (j = 0; j < skinfiles; j++, shaders++)
{
skin[j].numshaders = 1; //non-sequenced skins.
skin[j].ofsshaders = shaders;
shaders[0] = Mod_LoadSkinFile(surfname, j, NULL, 0, 0, NULL);
}
root[i].ofsskins = skin;
#endif
}
for (i = 0; i < header->numscenes; i++, grp++, inscene++)
{
Q_strncpyz(grp->name, inscene->name, sizeof(grp->name));
grp->isheirachical = 1;
grp->rate = BigFloat(inscene->framerate);
grp->loop = !(BigLong(inscene->flags) & ZYMSCENEFLAG_NOLOOP);
grp->numposes = BigLong(inscene->length);
grp->boneofs = matrix + BigLong(inscene->start)*12*root->numbones;
}
if (inscene != (zymscene_t*)((char*)header + header->lump_scenes.start+header->lump_scenes.length))
{
Con_Printf(CON_ERROR "%s, scene list appears corrupt.\n", mod->name);
return false;
}
for (i = 0; i < header->numsurfaces-1; i++)
root[i].nextsurf = &root[i+1];
for (i = 1; i < header->numsurfaces; i++)
{
root[i].shares_verts = 0;
root[i].numbones = root[0].numbones;
root[i].numverts = root[0].numverts;
root[i].ofsbones = root[0].ofsbones;
}
Alias_CalculateSkeletalNormals(root);
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0); //file replacement - inherit flags from any defunc mdl files.
Mod_ClampModelSize(mod);
mod->meshinfo = root;
mod->type = mod_alias;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif //ZYMOTICMODELS
///////////////////////////////////////////////////////////////
//psk
#ifdef PSKMODELS
/*Typedefs copied from DarkPlaces*/
typedef struct pskchunk_s
{
// id is one of the following:
// .psk:
// ACTRHEAD (recordsize = 0, numrecords = 0)
// PNTS0000 (recordsize = 12, pskpnts_t)
// VTXW0000 (recordsize = 16, pskvtxw_t)
// FACE0000 (recordsize = 12, pskface_t)
// MATT0000 (recordsize = 88, pskmatt_t)
// REFSKELT (recordsize = 120, pskboneinfo_t)
// RAWWEIGHTS (recordsize = 12, pskrawweights_t)
// .psa:
// ANIMHEAD (recordsize = 0, numrecords = 0)
// BONENAMES (recordsize = 120, pskboneinfo_t)
// ANIMINFO (recordsize = 168, pskaniminfo_t)
// ANIMKEYS (recordsize = 32, pskanimkeys_t)
char id[20];
// in .psk always 0x1e83b9
// in .psa always 0x2e
int version;
int recordsize;
int numrecords;
} pskchunk_t;
typedef struct pskpnts_s
{
float origin[3];
} pskpnts_t;
typedef struct pskvtxw_s
{
unsigned short pntsindex; // index into PNTS0000 chunk
unsigned char unknown1[2]; // seems to be garbage
float texcoord[2];
unsigned char mattindex; // index into MATT0000 chunk
unsigned char unknown2; // always 0?
unsigned char unknown3[2]; // seems to be garbage
} pskvtxw_t;
typedef struct pskface_s
{
unsigned short vtxwindex[3]; // triangle
unsigned char mattindex; // index into MATT0000 chunk
unsigned char unknown; // seems to be garbage
unsigned int group; // faces seem to be grouped, possibly for smoothing?
} pskface_t;
typedef struct pskmatt_s
{
char name[64];
int unknown[6]; // observed 0 0 0 0 5 0
} pskmatt_t;
typedef struct pskpose_s
{
float quat[4];
float origin[3];
float unknown; // probably a float, always seems to be 0
float size[3];
} pskpose_t;
typedef struct pskboneinfo_s
{
char name[64];
int unknown1;
int numchildren;
int parent; // root bones have 0 here
pskpose_t basepose;
} pskboneinfo_t;
typedef struct pskrawweights_s
{
float weight;
int pntsindex;
int boneindex;
} pskrawweights_t;
typedef struct pskaniminfo_s
{
char name[64];
char group[64];
int numbones;
int unknown1;
int unknown2;
int unknown3;
float unknown4;
float playtime; // not really needed
float fps; // frames per second
int unknown5;
int firstframe;
int numframes;
// firstanimkeys = (firstframe + frameindex) * numbones
} pskaniminfo_t;
typedef struct pskanimkeys_s
{
float origin[3];
float quat[4];
float frametime;
} pskanimkeys_t;
qboolean QDECL Mod_LoadPSKModel(model_t *mod, void *buffer)
{
pskchunk_t *chunk;
unsigned int pos = 0;
unsigned int i, j;
qboolean fail = false;
char basename[MAX_QPATH];
galiasinfo_t *gmdl;
#ifndef SERVERONLY
vec2_t *stcoord;
galiasskin_t *skin;
shader_t **gshaders;
#endif
galiasbone_t *bones;
galiasgroup_t *group;
float *animmatrix, *basematrix;
index_t *indexes;
float vrad;
int bonemap[MAX_BONES];
char *e;
pskpnts_t *pnts = NULL;
pskvtxw_t *vtxw = NULL;
pskface_t *face = NULL;
pskmatt_t *matt = NULL;
pskboneinfo_t *boneinfo = NULL;
pskrawweights_t *rawweights = NULL;
unsigned int num_pnts, num_vtxw=0, num_face=0, num_matt = 0, num_boneinfo=0, num_rawweights=0;
pskaniminfo_t *animinfo = NULL;
pskanimkeys_t *animkeys = NULL;
unsigned int num_animinfo=0, num_animkeys=0;
//#define PSK_GPU
#ifndef PSK_GPU
unsigned int num_trans;
galisskeletaltransforms_t *trans;
#else
vecV_t *skel_xyz;
vec3_t *skel_norm;
byte_vec4_t *skel_idx;
vec4_t *skel_weights;
#endif
/*load the psk*/
while (pos < com_filesize && !fail)
{
chunk = (pskchunk_t*)((char*)buffer + pos);
chunk->version = LittleLong(chunk->version);
chunk->recordsize = LittleLong(chunk->recordsize);
chunk->numrecords = LittleLong(chunk->numrecords);
pos += sizeof(*chunk);
if (!strcmp("ACTRHEAD", chunk->id) && chunk->recordsize == 0 && chunk->numrecords == 0)
{
}
else if (!strcmp("PNTS0000", chunk->id) && chunk->recordsize == sizeof(pskpnts_t))
{
num_pnts = chunk->numrecords;
pnts = (pskpnts_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_pnts; i++)
{
pnts[i].origin[0] = LittleFloat(pnts[i].origin[0]);
pnts[i].origin[1] = LittleFloat(pnts[i].origin[1]);
pnts[i].origin[2] = LittleFloat(pnts[i].origin[2]);
}
}
else if (!strcmp("VTXW0000", chunk->id) && chunk->recordsize == sizeof(pskvtxw_t))
{
num_vtxw = chunk->numrecords;
vtxw = (pskvtxw_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_vtxw; i++)
{
vtxw[i].pntsindex = LittleShort(vtxw[i].pntsindex);
vtxw[i].texcoord[0] = LittleFloat(vtxw[i].texcoord[0]);
vtxw[i].texcoord[1] = LittleFloat(vtxw[i].texcoord[1]);
}
}
else if (!strcmp("FACE0000", chunk->id) && chunk->recordsize == sizeof(pskface_t))
{
num_face = chunk->numrecords;
face = (pskface_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_face; i++)
{
face[i].vtxwindex[0] = LittleShort(face[i].vtxwindex[0]);
face[i].vtxwindex[1] = LittleShort(face[i].vtxwindex[1]);
face[i].vtxwindex[2] = LittleShort(face[i].vtxwindex[2]);
}
}
else if (!strcmp("MATT0000", chunk->id) && chunk->recordsize == sizeof(pskmatt_t))
{
num_matt = chunk->numrecords;
matt = (pskmatt_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
}
else if (!strcmp("REFSKELT", chunk->id) && chunk->recordsize == sizeof(pskboneinfo_t))
{
num_boneinfo = chunk->numrecords;
boneinfo = (pskboneinfo_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_boneinfo; i++)
{
boneinfo[i].parent = LittleLong(boneinfo[i].parent);
boneinfo[i].basepose.origin[0] = LittleFloat(boneinfo[i].basepose.origin[0]);
boneinfo[i].basepose.origin[1] = LittleFloat(boneinfo[i].basepose.origin[1]);
boneinfo[i].basepose.origin[2] = LittleFloat(boneinfo[i].basepose.origin[2]);
boneinfo[i].basepose.quat[0] = LittleFloat(boneinfo[i].basepose.quat[0]);
boneinfo[i].basepose.quat[1] = LittleFloat(boneinfo[i].basepose.quat[1]);
boneinfo[i].basepose.quat[2] = LittleFloat(boneinfo[i].basepose.quat[2]);
boneinfo[i].basepose.quat[3] = LittleFloat(boneinfo[i].basepose.quat[3]);
boneinfo[i].basepose.size[0] = LittleFloat(boneinfo[i].basepose.size[0]);
boneinfo[i].basepose.size[1] = LittleFloat(boneinfo[i].basepose.size[1]);
boneinfo[i].basepose.size[2] = LittleFloat(boneinfo[i].basepose.size[2]);
/*not sure if this is needed, but mimic DP*/
if (i)
{
boneinfo[i].basepose.quat[0] *= -1;
boneinfo[i].basepose.quat[2] *= -1;
}
boneinfo[i].basepose.quat[1] *= -1;
}
}
else if (!strcmp("RAWWEIGHTS", chunk->id) && chunk->recordsize == sizeof(pskrawweights_t))
{
num_rawweights = chunk->numrecords;
rawweights = (pskrawweights_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_rawweights; i++)
{
rawweights[i].boneindex = LittleLong(rawweights[i].boneindex);
rawweights[i].pntsindex = LittleLong(rawweights[i].pntsindex);
rawweights[i].weight = LittleFloat(rawweights[i].weight);
}
}
else
{
Con_Printf(CON_ERROR "%s has unsupported chunk %s of %i size with version %i.\n", mod->name, chunk->id, chunk->recordsize, chunk->version);
fail = true;
}
}
if (!num_matt)
fail = true;
if (!pnts || !vtxw || !face || !matt || !boneinfo || !rawweights)
fail = true;
/*attempt to load a psa file. don't die if we can't find one*/
COM_StripExtension(mod->name, basename, sizeof(basename));
buffer = COM_LoadTempMoreFile(va("%s.psa", basename));
if (buffer)
{
pos = 0;
while (pos < com_filesize && !fail)
{
chunk = (pskchunk_t*)((char*)buffer + pos);
chunk->version = LittleLong(chunk->version);
chunk->recordsize = LittleLong(chunk->recordsize);
chunk->numrecords = LittleLong(chunk->numrecords);
pos += sizeof(*chunk);
if (!strcmp("ANIMHEAD", chunk->id) && chunk->recordsize == 0 && chunk->numrecords == 0)
{
}
else if (!strcmp("BONENAMES", chunk->id) && chunk->recordsize == sizeof(pskboneinfo_t))
{
/*parsed purely to ensure that the bones match the main model*/
pskboneinfo_t *animbones = (pskboneinfo_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
if (num_boneinfo != chunk->numrecords)
{
fail = true;
Con_Printf("PSK/PSA bone counts do not match\n");
}
else
{
for (i = 0; i < num_boneinfo; i++)
{
/*assumption: 1:1 mapping will be common*/
if (!strcmp(boneinfo[i].name, animbones[i].name))
bonemap[i] = i;
else
{
/*non 1:1 mapping*/
for (j = 0; j < chunk->numrecords; j++)
{
if (!strcmp(boneinfo[i].name, animbones[j].name))
{
bonemap[i] = j;
break;
}
}
if (j == chunk->numrecords)
{
fail = true;
Con_Printf("PSK bone %s does not exist in PSA %s\n", boneinfo[i].name, basename);
break;
}
}
}
}
}
else if (!strcmp("ANIMINFO", chunk->id) && chunk->recordsize == sizeof(pskaniminfo_t))
{
num_animinfo = chunk->numrecords;
animinfo = (pskaniminfo_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_animinfo; i++)
{
animinfo[i].firstframe = LittleLong(animinfo[i].firstframe);
animinfo[i].numframes = LittleLong(animinfo[i].numframes);
animinfo[i].numbones = LittleLong(animinfo[i].numbones);
animinfo[i].fps = LittleFloat(animinfo[i].fps);
animinfo[i].playtime = LittleFloat(animinfo[i].playtime);
}
}
else if (!strcmp("ANIMKEYS", chunk->id) && chunk->recordsize == sizeof(pskanimkeys_t))
{
num_animkeys = chunk->numrecords;
animkeys = (pskanimkeys_t*)((char*)buffer + pos);
pos += chunk->recordsize * chunk->numrecords;
for (i = 0; i < num_animkeys; i++)
{
animkeys[i].origin[0] = LittleFloat(animkeys[i].origin[0]);
animkeys[i].origin[1] = LittleFloat(animkeys[i].origin[1]);
animkeys[i].origin[2] = LittleFloat(animkeys[i].origin[2]);
animkeys[i].quat[0] = LittleFloat(animkeys[i].quat[0]);
animkeys[i].quat[1] = LittleFloat(animkeys[i].quat[1]);
animkeys[i].quat[2] = LittleFloat(animkeys[i].quat[2]);
animkeys[i].quat[3] = LittleFloat(animkeys[i].quat[3]);
/*not sure if this is needed, but mimic DP*/
if (i%num_boneinfo)
{
animkeys[i].quat[0] *= -1;
animkeys[i].quat[2] *= -1;
}
animkeys[i].quat[1] *= -1;
}
}
else if (!strcmp("SCALEKEYS", chunk->id) && chunk->recordsize == 16)
{
pos += chunk->recordsize * chunk->numrecords;
}
else
{
Con_Printf(CON_ERROR "%s has unsupported chunk %s of %i size with version %i.\n", va("%s.psa", basename), chunk->id, chunk->recordsize, chunk->version);
fail = true;
}
}
if (fail)
{
animinfo = NULL;
num_animinfo = 0;
animkeys = NULL;
num_animkeys = 0;
fail = false;
}
}
if (fail)
{
return false;
}
gmdl = ZG_Malloc(&loadmodel->memgroup, sizeof(*gmdl)*num_matt);
/*bones!*/
bones = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasbone_t) * num_boneinfo);
for (i = 0; i < num_boneinfo; i++)
{
Q_strncpyz(bones[i].name, boneinfo[i].name, sizeof(bones[i].name));
e = bones[i].name + strlen(bones[i].name);
while(e > bones[i].name && e[-1] == ' ')
*--e = 0;
bones[i].parent = boneinfo[i].parent;
if (i == 0 && bones[i].parent == 0)
bones[i].parent = -1;
else if (bones[i].parent >= i || bones[i].parent < -1)
{
Con_Printf("Invalid bones\n");
break;
}
}
basematrix = ZG_Malloc(&loadmodel->memgroup, num_boneinfo*sizeof(float)*12);
for (i = 0; i < num_boneinfo; i++)
{
float tmp[12];
PSKGenMatrix(
boneinfo[i].basepose.origin[0], boneinfo[i].basepose.origin[1], boneinfo[i].basepose.origin[2],
boneinfo[i].basepose.quat[0], boneinfo[i].basepose.quat[1], boneinfo[i].basepose.quat[2], boneinfo[i].basepose.quat[3],
tmp);
if (bones[i].parent < 0)
memcpy(basematrix + i*12, tmp, sizeof(float)*12);
else
R_ConcatTransforms((void*)(basematrix + bones[i].parent*12), (void*)tmp, (void*)(basematrix+i*12));
}
for (i = 0; i < num_boneinfo; i++)
{
Matrix3x4_Invert_Simple(basematrix+i*12, bones[i].inverse);
}
#ifndef PSK_GPU
/*expand the translations*/
num_trans = 0;
for (i = 0; i < num_vtxw; i++)
{
for (j = 0; j < num_rawweights; j++)
{
if (rawweights[j].pntsindex == vtxw[i].pntsindex)
{
num_trans++;
}
}
}
trans = ZG_Malloc(&loadmodel->memgroup, sizeof(*trans)*num_trans);
num_trans = 0;
for (i = 0; i < num_vtxw; i++)
{
// first_trans = num_trans;
for (j = 0; j < num_rawweights; j++)
{
if (rawweights[j].pntsindex == vtxw[i].pntsindex)
{
vec3_t tmp;
trans[num_trans].vertexindex = i;
trans[num_trans].boneindex = rawweights[j].boneindex;
VectorTransform(pnts[rawweights[j].pntsindex].origin, (void*)bones[rawweights[j].boneindex].inverse, tmp);
VectorScale(tmp, rawweights[j].weight, trans[num_trans].org);
trans[num_trans].org[3] = rawweights[j].weight;
num_trans++;
}
}
}
#else
skel_xyz = Hunk_Alloc(sizeof(*skel_xyz) * num_vtxw);
skel_norm = Hunk_Alloc(sizeof(*skel_norm) * num_vtxw);
skel_idx = Hunk_Alloc(sizeof(*skel_idx) * num_vtxw);
skel_weights = Hunk_Alloc(sizeof(*skel_weights) * num_vtxw);
for (i = 0; i < num_vtxw; i++)
{
float t;
*(unsigned int*)skel_idx[i] = ~0;
for (j = 0; j < num_rawweights; j++)
{
if (rawweights[j].pntsindex == vtxw[i].pntsindex)
{
int in, lin = -1;
float liv = rawweights[j].weight;
for (in = 0; in < 4; in++)
{
if (liv > skel_weights[i][in])
{
liv = skel_weights[i][in];
lin = in;
if (!liv)
break;
}
}
if (lin >= 0)
{
skel_idx[i][lin] = rawweights[j].boneindex;
skel_weights[i][lin] = rawweights[j].weight;
}
}
}
t = 0;
for (j = 0; j < 4; j++)
t += skel_weights[i][j];
if (t != 1)
for (j = 0; j < 4; j++)
skel_weights[i][j] *= 1/t;
skel_xyz[i][0] = pnts[vtxw[i].pntsindex].origin[0];
skel_xyz[i][1] = pnts[vtxw[i].pntsindex].origin[1];
skel_xyz[i][2] = pnts[vtxw[i].pntsindex].origin[2];
}
#endif
#ifndef SERVERONLY
/*st coords, all share the same list*/
stcoord = ZG_Malloc(&loadmodel->memgroup, sizeof(vec2_t)*num_vtxw);
for (i = 0; i < num_vtxw; i++)
{
stcoord[i][0] = vtxw[i].texcoord[0];
stcoord[i][1] = vtxw[i].texcoord[1];
}
#endif
/*allocate faces in a single block, as we at least know an upper bound*/
indexes = ZG_Malloc(&loadmodel->memgroup, sizeof(index_t)*num_face*3);
if (animinfo && animkeys)
{
int numgroups = 0;
frameinfo_t *frameinfo = ParseFrameInfo(mod->name, &numgroups);
if (numgroups)
{
/*externally supplied listing of frames. ignore all framegroups in the model and use only the pose info*/
group = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t)*numgroups + num_animkeys*sizeof(float)*12);
animmatrix = (float*)(group+numgroups);
for (j = 0; j < numgroups; j++)
{
//FIXME: bound check
group[j].boneofs = animmatrix + 12*num_boneinfo*frameinfo[j].firstpose;
group[j].numposes = frameinfo[j].posecount;
if (*frameinfo[j].name)
snprintf(group[j].name, sizeof(group[j].name), "%s", frameinfo[j].name);
else
snprintf(group[j].name, sizeof(group[j].name), "frame_%i", j);
group[j].loop = frameinfo[j].loop;
group[j].rate = frameinfo[j].fps;
group[j].isheirachical = true;
}
num_animinfo = numgroups;
}
else if (dpcompat_psa_ungroup.ival)
{
/*unpack each frame of each animation to be a separate framegroup*/
unsigned int iframe; /*individual frame count*/
iframe = 0;
for (i = 0; i < num_animinfo; i++)
iframe += animinfo[i].numframes;
group = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t)*iframe + num_animkeys*sizeof(float)*12);
animmatrix = (float*)(group+iframe);
iframe = 0;
for (j = 0; j < num_animinfo; j++)
{
for (i = 0; i < animinfo[j].numframes; i++)
{
group[iframe].boneofs = animmatrix + 12*num_boneinfo*(animinfo[j].firstframe+i);
group[iframe].numposes = 1;
snprintf(group[iframe].name, sizeof(group[iframe].name), "%s_%i", animinfo[j].name, i);
group[iframe].loop = true;
group[iframe].rate = animinfo[j].fps;
group[iframe].isheirachical = true;
iframe++;
}
}
num_animinfo = iframe;
}
else
{
/*keep each framegroup as a group*/
group = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t)*num_animinfo + num_animkeys*sizeof(float)*12);
animmatrix = (float*)(group+num_animinfo);
for (i = 0; i < num_animinfo; i++)
{
group[i].boneofs = animmatrix + 12*num_boneinfo*animinfo[i].firstframe;
group[i].numposes = animinfo[i].numframes;
Q_strncpyz(group[i].name, animinfo[i].name, sizeof(group[i].name));
group[i].loop = true;
group[i].rate = animinfo[i].fps;
group[i].isheirachical = true;
}
}
for (j = 0; j < num_animkeys; j += num_boneinfo)
{
pskanimkeys_t *sb;
for (i = 0; i < num_boneinfo; i++)
{
sb = &animkeys[j + bonemap[i]];
PSKGenMatrix(
sb->origin[0], sb->origin[1], sb->origin[2],
sb->quat[0], sb->quat[1], sb->quat[2], sb->quat[3],
animmatrix + (j+i)*12);
}
}
}
else
{
num_animinfo = 1;
/*build a base pose*/
group = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t) + num_boneinfo*sizeof(float)*12);
animmatrix = basematrix;
group->boneofs = animmatrix;
group->numposes = 1;
strcpy(group->name, "base");
group->loop = true;
group->rate = 10;
group->isheirachical = false;
}
#ifndef SERVERONLY
skin = ZG_Malloc(&loadmodel->memgroup, num_matt * (sizeof(galiasskin_t) + sizeof(shader_t*)));
gshaders = (shader_t**)(skin + num_matt);
for (i = 0; i < num_matt; i++, skin++)
{
skin->ofsshaders = &gshaders[i];
skin->numshaders = 1;
skin->skinspeed = 10;
Q_strncpyz(skin->name, matt[i].name, sizeof(skin->name));
gshaders[i] = R_RegisterSkin(matt[i].name, mod->name);
R_BuildDefaultTexnums(NULL, gshaders[i]);
if (gshaders[i]->flags & SHADER_NOIMAGE)
Con_Printf("Unable to load texture for shader \"%s\" for model \"%s\"\n", gshaders[i]->name, loadmodel->name);
gmdl[i].ofsskins = skin;
gmdl[i].numskins = 1;
gmdl[i].ofs_st_array = stcoord;
gmdl[i].numverts = num_vtxw;
#else
for (i = 0; i < num_matt; i++)
{
#endif
gmdl[i].groupofs = group;
gmdl[i].groups = num_animinfo;
gmdl[i].baseframeofs = basematrix;
gmdl[i].numindexes = 0;
for (j = 0; j < num_face; j++)
{
if (face[j].mattindex%num_matt == i)
{
indexes[gmdl[i].numindexes+0] = face[j].vtxwindex[0];
indexes[gmdl[i].numindexes+1] = face[j].vtxwindex[1];
indexes[gmdl[i].numindexes+2] = face[j].vtxwindex[2];
gmdl[i].numindexes += 3;
}
}
gmdl[i].ofs_indexes = indexes;
indexes += gmdl[i].numindexes;
gmdl[i].ofsbones = bones;
gmdl[i].numbones = num_boneinfo;
#ifndef PSK_GPU
gmdl[i].ofsswtransforms = trans;
gmdl[i].numswtransforms = num_trans;
#else
gmdl[i].ofs_skel_idx = skel_idx;
gmdl[i].ofs_skel_weight = skel_weights;
gmdl[i].ofs_skel_xyz = skel_xyz;
gmdl[i].ofs_skel_norm = skel_norm;
#endif
gmdl[i].shares_verts = 0;
gmdl[i].shares_bones = 0;
gmdl[i].nextsurf = (i != num_matt-1)?&gmdl[i+1]:NULL;
}
if (fail)
{
return false;
}
vrad = Alias_CalculateSkeletalNormals(gmdl);
mod->mins[0] = mod->mins[1] = mod->mins[2] = -vrad;
mod->maxs[0] = mod->maxs[1] = mod->maxs[2] = vrad;
mod->radius = vrad;
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0); //file replacement - inherit flags from any defunc mdl files.
Mod_ClampModelSize(mod);
mod->meshinfo = gmdl;
mod->type = mod_alias;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif
//////////////////////////////////////////////////////////////
//dpm
#ifdef DPMMODELS
// header for the entire file
typedef struct dpmheader_s
{
char id[16]; // "DARKPLACESMODEL\0", length 16
unsigned int type; // 2 (hierarchical skeletal pose)
unsigned int filesize; // size of entire model file
float mins[3], maxs[3], yawradius, allradius; // for clipping uses
// these offsets are relative to the file
unsigned int num_bones;
unsigned int num_meshs;
unsigned int num_frames;
unsigned int ofs_bones; // dpmbone_t bone[num_bones];
unsigned int ofs_meshs; // dpmmesh_t mesh[num_meshs];
unsigned int ofs_frames; // dpmframe_t frame[num_frames];
} dpmheader_t;
// there may be more than one of these
typedef struct dpmmesh_s
{
// these offsets are relative to the file
char shadername[32]; // name of the shader to use
unsigned int num_verts;
unsigned int num_tris;
unsigned int ofs_verts; // dpmvertex_t vert[numvertices]; // see vertex struct
unsigned int ofs_texcoords; // float texcoords[numvertices][2];
unsigned int ofs_indices; // unsigned int indices[numtris*3]; // designed for glDrawElements (each triangle is 3 unsigned int indices)
unsigned int ofs_groupids; // unsigned int groupids[numtris]; // the meaning of these values is entirely up to the gamecode and modeler
} dpmmesh_t;
// if set on a bone, it must be protected from removal
#define DPMBONEFLAG_ATTACHMENT 1
// one per bone
typedef struct dpmbone_s
{
// name examples: upperleftarm leftfinger1 leftfinger2 hand, etc
char name[32];
// parent bone number
signed int parent;
// flags for the bone
unsigned int flags;
} dpmbone_t;
// a bonepose matrix is intended to be used like this:
// (n = output vertex, v = input vertex, m = matrix, f = influence)
// n[0] = v[0] * m[0][0] + v[1] * m[0][1] + v[2] * m[0][2] + f * m[0][3];
// n[1] = v[0] * m[1][0] + v[1] * m[1][1] + v[2] * m[1][2] + f * m[1][3];
// n[2] = v[0] * m[2][0] + v[1] * m[2][1] + v[2] * m[2][2] + f * m[2][3];
typedef struct dpmbonepose_s
{
float matrix[3][4];
} dpmbonepose_t;
// immediately followed by bone positions for the frame
typedef struct dpmframe_s
{
// name examples: idle_1 idle_2 idle_3 shoot_1 shoot_2 shoot_3, etc
char name[32];
float mins[3], maxs[3], yawradius, allradius;
int ofs_bonepositions; // dpmbonepose_t bonepositions[bones];
} dpmframe_t;
// one or more of these per vertex
typedef struct dpmbonevert_s
{
float origin[3]; // vertex location (these blend)
float influence; // influence fraction (these must add up to 1)
float normal[3]; // surface normal (these blend)
unsigned int bonenum; // number of the bone
} dpmbonevert_t;
// variable size, parsed sequentially
typedef struct dpmvertex_s
{
unsigned int numbones;
// immediately followed by 1 or more dpmbonevert_t structures
} dpmvertex_t;
qboolean QDECL Mod_LoadDarkPlacesModel(model_t *mod, void *buffer)
{
#ifndef SERVERONLY
galiasskin_t *skin;
shader_t **shaders;
int skinfiles;
float *inst;
float *outst;
#endif
int i, j, k;
dpmheader_t *header;
galiasinfo_t *root, *m;
dpmmesh_t *mesh;
dpmvertex_t *vert;
dpmbonevert_t *bonevert;
galisskeletaltransforms_t *transforms;
galiasbone_t *outbone;
dpmbone_t *inbone;
float *outposedata;
galiasgroup_t *outgroups;
float *inposedata;
dpmframe_t *inframes;
unsigned int *index; index_t *outdex; // groan...
int numtransforms;
int numverts;
loadmodel=mod;
header = buffer;
if (memcmp(header->id, "DARKPLACESMODEL\0", 16))
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: %s, doesn't appear to be a darkplaces model!\n", mod->name);
return false;
}
if (BigLong(header->type) != 2)
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: %s, only type 2 is supported\n", mod->name);
return false;
}
for (i = 0; i < sizeof(dpmheader_t)/4; i++)
((int*)header)[i] = BigLong(((int*)header)[i]);
if (!header->num_bones)
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: %s, no bones\n", mod->name);
return false;
}
if (!header->num_frames)
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: %s, no frames\n", mod->name);
return false;
}
if (!header->num_meshs)
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: %s, no surfaces\n", mod->name);
return false;
}
VectorCopy(header->mins, mod->mins);
VectorCopy(header->maxs, mod->maxs);
root = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasinfo_t)*header->num_meshs);
mesh = (dpmmesh_t*)((char*)buffer + header->ofs_meshs);
for (i = 0; i < header->num_meshs; i++, mesh++)
{
//work out how much memory we need to allocate
mesh->num_verts = BigLong(mesh->num_verts);
mesh->num_tris = BigLong(mesh->num_tris);
mesh->ofs_verts = BigLong(mesh->ofs_verts);
mesh->ofs_texcoords = BigLong(mesh->ofs_texcoords);
mesh->ofs_indices = BigLong(mesh->ofs_indices);
mesh->ofs_groupids = BigLong(mesh->ofs_groupids);
numverts = mesh->num_verts;
numtransforms = 0;
//count and byteswap the transformations
vert = (dpmvertex_t*)((char *)buffer+mesh->ofs_verts);
for (j = 0; j < mesh->num_verts; j++)
{
vert->numbones = BigLong(vert->numbones);
numtransforms += vert->numbones;
bonevert = (dpmbonevert_t*)(vert+1);
vert = (dpmvertex_t*)(bonevert+vert->numbones);
}
m = &root[i];
#ifdef SERVERONLY
transforms = ZG_Malloc(&loadmodel->memgroup, numtransforms*sizeof(galisskeletaltransforms_t) + mesh->num_tris*3*sizeof(index_t));
#else
outst = ZG_Malloc(&loadmodel->memgroup, numverts*sizeof(vec2_t) + numtransforms*sizeof(galisskeletaltransforms_t) + mesh->num_tris*3*sizeof(index_t));
m->ofs_st_array = (vec2_t*)outst;
m->numverts = mesh->num_verts;
inst = (float*)((char*)buffer + mesh->ofs_texcoords);
for (j = 0; j < numverts; j++, outst+=2, inst+=2)
{
outst[0] = BigFloat(inst[0]);
outst[1] = BigFloat(inst[1]);
}
transforms = (galisskeletaltransforms_t*)outst;
#endif
//build the transform list.
m->ofsswtransforms = transforms;
m->numswtransforms = numtransforms;
vert = (dpmvertex_t*)((char *)buffer+mesh->ofs_verts);
for (j = 0; j < mesh->num_verts; j++)
{
bonevert = (dpmbonevert_t*)(vert+1);
for (k = 0; k < vert->numbones; k++, bonevert++, transforms++)
{
transforms->boneindex = BigLong(bonevert->bonenum);
transforms->vertexindex = j;
transforms->org[0] = BigFloat(bonevert->origin[0]);
transforms->org[1] = BigFloat(bonevert->origin[1]);
transforms->org[2] = BigFloat(bonevert->origin[2]);
transforms->org[3] = BigFloat(bonevert->influence);
//do nothing with the normals. :(
}
vert = (dpmvertex_t*)bonevert;
}
index = (unsigned int*)((char*)buffer + mesh->ofs_indices);
outdex = (index_t *)transforms;
m->ofs_indexes = outdex;
m->numindexes = mesh->num_tris*3;
for (j = 0; j < m->numindexes; j++)
{
*outdex++ = BigLong(*index++);
}
}
outbone = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasbone_t)*header->num_bones);
inbone = (dpmbone_t*)((char*)buffer + header->ofs_bones);
for (i = 0; i < header->num_bones; i++)
{
outbone[i].parent = BigLong(inbone[i].parent);
if (outbone[i].parent >= i || outbone[i].parent < -1)
{
Con_Printf(CON_ERROR "Mod_LoadDarkPlacesModel: bad bone index in %s\n", mod->name);
return false;
}
Q_strncpyz(outbone[i].name, inbone[i].name, sizeof(outbone[i].name));
//throw away the flags.
}
outgroups = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t)*header->num_frames + sizeof(float)*header->num_frames*header->num_bones*12);
outposedata = (float*)(outgroups+header->num_frames);
inframes = (dpmframe_t*)((char*)buffer + header->ofs_frames);
for (i = 0; i < header->num_frames; i++)
{
inframes[i].ofs_bonepositions = BigLong(inframes[i].ofs_bonepositions);
inframes[i].allradius = BigLong(inframes[i].allradius);
inframes[i].yawradius = BigLong(inframes[i].yawradius);
inframes[i].mins[0] = BigLong(inframes[i].mins[0]);
inframes[i].mins[1] = BigLong(inframes[i].mins[1]);
inframes[i].mins[2] = BigLong(inframes[i].mins[2]);
inframes[i].maxs[0] = BigLong(inframes[i].maxs[0]);
inframes[i].maxs[1] = BigLong(inframes[i].maxs[1]);
inframes[i].maxs[2] = BigLong(inframes[i].maxs[2]);
Q_strncpyz(outgroups[i].name, inframes[i].name, sizeof(outgroups[i].name));
outgroups[i].rate = 10;
outgroups[i].numposes = 1;
outgroups[i].isheirachical = true;
outgroups[i].boneofs = outposedata;
inposedata = (float*)((char*)buffer + inframes[i].ofs_bonepositions);
for (j = 0; j < header->num_bones*12; j++)
*outposedata++ = BigFloat(*inposedata++);
}
#ifndef SERVERONLY
skinfiles = Mod_BuildSkinFileList(loadmodel->name);
if (skinfiles < 1)
skinfiles = 1;
#endif
mesh = (dpmmesh_t*)((char*)buffer + header->ofs_meshs);
for (i = 0; i < header->num_meshs; i++, mesh++)
{
m = &root[i];
if (i < header->num_meshs-1)
m->nextsurf = &root[i+1];
m->shares_bones = 0;
m->ofsbones = outbone;
m->numbones = header->num_bones;
m->groups = header->num_frames;
m->groupofs = outgroups;
#ifdef SERVERONLY
m->numskins = 1;
#else
m->numskins = skinfiles;
skin = ZG_Malloc(&loadmodel->memgroup, (sizeof(galiasskin_t)+sizeof(shader_t*))*skinfiles);
shaders = (shader_t**)(skin+skinfiles);
for (j = 0; j < skinfiles; j++, shaders++)
{
skin[j].numshaders = 1; //non-sequenced skins.
skin[j].ofsshaders = shaders;
shaders[0] = Mod_LoadSkinFile(mesh->shadername, j, NULL, 0, 0, NULL);
}
m->ofsskins = skin;
#endif
}
Alias_CalculateSkeletalNormals(root);
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0); //file replacement - inherit flags from any defunc mdl files.
Mod_ClampModelSize(mod);
mod->meshinfo = root;
mod->type = mod_alias;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif //DPMMODELS
#ifdef INTERQUAKEMODELS
#define IQM_MAGIC "INTERQUAKEMODEL"
#define IQM_VERSION1 1
#define IQM_VERSION2 2
struct iqmheader
{
char magic[16];
unsigned int version;
unsigned int filesize;
unsigned int flags;
unsigned int num_text, ofs_text;
unsigned int num_meshes, ofs_meshes;
unsigned int num_vertexarrays, num_vertexes, ofs_vertexarrays;
unsigned int num_triangles, ofs_triangles, ofs_adjacency;
unsigned int num_joints, ofs_joints;
unsigned int num_poses, ofs_poses;
unsigned int num_anims, ofs_anims;
unsigned int num_frames, num_framechannels, ofs_frames, ofs_bounds;
unsigned int num_comment, ofs_comment;
unsigned int num_extensions, ofs_extensions;
};
struct iqmmesh
{
unsigned int name;
unsigned int material;
unsigned int first_vertex, num_vertexes;
unsigned int first_triangle, num_triangles;
};
enum
{
IQM_POSITION = 0,
IQM_TEXCOORD = 1,
IQM_NORMAL = 2,
IQM_TANGENT = 3,
IQM_BLENDINDEXES = 4,
IQM_BLENDWEIGHTS = 5,
IQM_COLOR = 6,
IQM_CUSTOM = 0x10
};
enum
{
IQM_BYTE = 0,
IQM_UBYTE = 1,
IQM_SHORT = 2,
IQM_USHORT = 3,
IQM_INT = 4,
IQM_UINT = 5,
IQM_HALF = 6,
IQM_FLOAT = 7,
IQM_DOUBLE = 8,
};
struct iqmtriangle
{
unsigned int vertex[3];
};
struct iqmjoint1
{
unsigned int name;
int parent;
float translate[3], rotate[3], scale[3];
};
struct iqmjoint2
{
unsigned int name;
int parent;
float translate[3], rotate[4], scale[3];
};
struct iqmpose1
{
int parent;
unsigned int mask;
float channeloffset[9];
float channelscale[9];
};
struct iqmpose2
{
int parent;
unsigned int mask;
float channeloffset[10];
float channelscale[10];
};
struct iqmanim
{
unsigned int name;
unsigned int first_frame, num_frames;
float framerate;
unsigned int flags;
};
enum
{
IQM_LOOP = 1<<0
};
struct iqmvertexarray
{
unsigned int type;
unsigned int flags;
unsigned int format;
unsigned int size;
unsigned int offset;
};
struct iqmbounds
{
float bbmin[3], bbmax[3];
float xyradius, radius;
};
/*
galisskeletaltransforms_t *IQM_ImportTransforms(int *resultcount, int inverts, float *vpos, float *tcoord, float *vnorm, float *vtang, unsigned char *vbone, unsigned char *vweight)
{
galisskeletaltransforms_t *t, *r;
unsigned int num_t = 0;
unsigned int v, j;
for (v = 0; v < inverts*4; v++)
{
if (vweight[v])
num_t++;
}
t = r = Hunk_Alloc(sizeof(*r)*num_t);
for (v = 0; v < inverts; v++)
{
for (j = 0; j < 4; j++)
{
if (vweight[(v<<2)+j])
{
t->boneindex = vbone[(v<<2)+j];
t->vertexindex = v;
VectorScale(vpos, vweight[(v<<2)+j]/255.0, t->org);
VectorScale(vnorm, vweight[(v<<2)+j]/255.0, t->normal);
t++;
}
}
}
return r;
}
*/
galiasinfo_t *Mod_ParseIQMMeshModel(model_t *mod, char *buffer)
{
struct iqmheader *h = (struct iqmheader *)buffer;
struct iqmmesh *mesh;
struct iqmvertexarray *varray;
struct iqmtriangle *tris;
struct iqmanim *anim;
unsigned int i, j, t, nt;
char *strings;
float *vpos = NULL, *tcoord = NULL, *vnorm = NULL, *vtang = NULL;
unsigned char *vbone = NULL, *vweight = NULL;
unsigned int type, fmt, size, offset;
unsigned short *framedata;
vecV_t *opos;
vec3_t *onorm1, *onorm2, *onorm3;
vec4_t *oweight;
byte_vec4_t *oindex;
float *opose,*oposebase;
vec2_t *otcoords;
int memsize;
galiasinfo_t *gai;
#ifndef SERVERONLY
galiasskin_t *skin;
shader_t **shaders;
#endif
galiasgroup_t *fgroup;
galiasbone_t *bones;
index_t *idx;
float basepose[12 * MAX_BONES];
qboolean noweights;
frameinfo_t *framegroups;
int numgroups;
if (memcmp(h->magic, IQM_MAGIC, sizeof(h->magic)))
{
Con_Printf("%s: format not recognised\n", mod->name);
return NULL;
}
if (h->version != IQM_VERSION1 && h->version != IQM_VERSION2)
{
Con_Printf("%s: unsupported IQM version\n", mod->name);
return NULL;
}
if (h->filesize != com_filesize)
{
Con_Printf("%s: size (%u != %u)\n", mod->name, h->filesize, com_filesize);
return NULL;
}
varray = (struct iqmvertexarray*)(buffer + h->ofs_vertexarrays);
for (i = 0; i < h->num_vertexarrays; i++)
{
type = LittleLong(varray[i].type);
fmt = LittleLong(varray[i].format);
size = LittleLong(varray[i].size);
offset = LittleLong(varray[i].offset);
if (type == IQM_POSITION && fmt == IQM_FLOAT && size == 3)
vpos = (float*)(buffer + offset);
else if (type == IQM_TEXCOORD && fmt == IQM_FLOAT && size == 2)
tcoord = (float*)(buffer + offset);
else if (type == IQM_NORMAL && fmt == IQM_FLOAT && size == 3)
vnorm = (float*)(buffer + offset);
else if (type == IQM_TANGENT && fmt == IQM_FLOAT && size == 4) /*yup, 4*/
vtang = (float*)(buffer + offset);
else if (type == IQM_BLENDINDEXES && fmt == IQM_UBYTE && size == 4)
vbone = (unsigned char *)(buffer + offset);
else if (type == IQM_BLENDWEIGHTS && fmt == IQM_UBYTE && size == 4)
vweight = (unsigned char *)(buffer + offset);
else
Con_Printf("Unrecognised iqm info\n");
}
if (!h->num_meshes)
return NULL;
//a mesh must contain vertex coords or its not much of a mesh.
//we also require texcoords because we can.
//we don't require normals
//we don't require weights, but such models won't animate.
if (h->num_vertexes > 0 && (!vpos || !tcoord))
{
Con_Printf("%s is missing vertex array data\n", loadmodel->name);
return NULL;
}
noweights = !vbone || !vweight;
if (noweights)
{
if (h->num_frames || h->num_anims || h->num_joints)
return NULL;
}
strings = buffer + h->ofs_text;
/*try to completely disregard all the info the creator carefully added to their model...*/
numgroups = 0;
framegroups = NULL;
if (!numgroups)
framegroups = ParseFrameInfo(loadmodel->name, &numgroups);
if (!numgroups && h->num_anims)
{
/*use the model's framegroups*/
numgroups = h->num_anims;
framegroups = malloc(sizeof(*framegroups)*numgroups);
anim = (struct iqmanim*)(buffer + h->ofs_anims);
for (i = 0; i < numgroups; i++)
{
framegroups[i].firstpose = LittleLong(anim[i].first_frame);
framegroups[i].posecount = LittleLong(anim[i].num_frames);
framegroups[i].fps = LittleFloat(anim[i].framerate);
framegroups[i].loop = !!(LittleLong(anim[i].flags) & IQM_LOOP);
Q_strncpyz(framegroups[i].name, strings+anim[i].name, sizeof(fgroup[i].name));
}
}
if (!numgroups)
{ /*base frame only*/
numgroups = 1;
framegroups = malloc(sizeof(*framegroups));
framegroups->firstpose = -1;
framegroups->posecount = 1;
framegroups->fps = 10;
framegroups->loop = 1;
strcpy(framegroups->name, "base");
}
mesh = (struct iqmmesh*)(buffer + h->ofs_meshes);
memsize = sizeof(*gai)*h->num_meshes;
memsize += sizeof(*fgroup)*numgroups + sizeof(float)*12*(h->num_joints + (h->num_poses*h->num_frames)) + sizeof(*bones)*h->num_joints;
memsize += (sizeof(*opos) + sizeof(*onorm1) + sizeof(*onorm2) + sizeof(*onorm3) + sizeof(*otcoords) + (noweights?0:(sizeof(*oindex)+sizeof(*oweight)))) * h->num_vertexes;
#ifndef SERVERONLY
memsize += sizeof(*skin)*h->num_meshes + sizeof(*shaders)*h->num_meshes;
#endif
/*allocate a nice big block of memory and figure out where stuff is*/
gai = ZG_Malloc(&loadmodel->memgroup, memsize);
bones = (galiasbone_t*)(gai + h->num_meshes);
opos = (vecV_t*)(bones + h->num_joints);
onorm3 = (vec3_t*)(opos + h->num_vertexes);
onorm2 = (vec3_t*)(onorm3 + h->num_vertexes);
onorm1 = (vec3_t*)(onorm2 + h->num_vertexes);
if (noweights)
{
oindex = NULL;
oweight = NULL;
otcoords = (vec2_t*)(onorm1 + h->num_vertexes);
}
else
{
oindex = (byte_vec4_t*)(onorm1 + h->num_vertexes);
oweight = (vec4_t*)(oindex + h->num_vertexes);
otcoords = (vec2_t*)(oweight + h->num_vertexes);
}
fgroup = (galiasgroup_t*)(otcoords + h->num_vertexes);
oposebase = (float*)(fgroup + numgroups);
opose = oposebase + 12*h->num_joints;
#ifndef SERVERONLY
skin = (galiasskin_t*)(opose + 12*(h->num_poses*h->num_frames));
shaders = (shader_t**)(skin + h->num_meshes);
#endif
//no code to load animations or bones
framedata = (unsigned short*)(buffer + h->ofs_frames);
/*Version 1 supports only normalized quaternions, version 2 uses complete quaternions. Some struct sizes change for this, otherwise functionally identical.*/
if (h->version == IQM_VERSION1)
{
struct iqmpose1 *p, *ipose = (struct iqmpose1*)(buffer + h->ofs_poses);
struct iqmjoint1 *ijoint = (struct iqmjoint1*)(buffer + h->ofs_joints);
vec3_t pos;
vec4_t quat;
vec3_t scale;
float mat[12];
//joint info (mesh)
for (i = 0; i < h->num_joints; i++)
{
Q_strncpyz(bones[i].name, strings+ijoint[i].name, sizeof(bones[i].name));
bones[i].parent = ijoint[i].parent;
GenMatrixPosQuat3Scale(ijoint[i].translate, ijoint[i].rotate, ijoint[i].scale, mat);
if (ijoint[i].parent >= 0)
Matrix3x4_Multiply(mat, &basepose[ijoint[i].parent*12], &basepose[i*12]);
else
memcpy(&basepose[i*12], mat, sizeof(mat));
Matrix3x4_Invert_Simple(&basepose[i*12], bones[i].inverse);
}
//pose info (anim)
for (i = 0; i < h->num_frames; i++)
{
for (j = 0, p = ipose; j < h->num_poses; j++, p++)
{
pos[0] = p->channeloffset[0]; if (p->mask & 1) pos[0] += *framedata++ * p->channelscale[0];
pos[1] = p->channeloffset[1]; if (p->mask & 2) pos[1] += *framedata++ * p->channelscale[1];
pos[2] = p->channeloffset[2]; if (p->mask & 4) pos[2] += *framedata++ * p->channelscale[2];
quat[0] = p->channeloffset[3]; if (p->mask & 8) quat[0] += *framedata++ * p->channelscale[3];
quat[1] = p->channeloffset[4]; if (p->mask & 16) quat[1] += *framedata++ * p->channelscale[4];
quat[2] = p->channeloffset[5]; if (p->mask & 32) quat[2] += *framedata++ * p->channelscale[5];
scale[0] = p->channeloffset[6]; if (p->mask & 64) scale[0] += *framedata++ * p->channelscale[6];
scale[1] = p->channeloffset[7]; if (p->mask & 128) scale[1] += *framedata++ * p->channelscale[7];
scale[2] = p->channeloffset[8]; if (p->mask & 256) scale[2] += *framedata++ * p->channelscale[8];
quat[3] = -sqrt(max(1.0 - pow(VectorLength(quat),2), 0.0));
GenMatrixPosQuat3Scale(pos, quat, scale, &opose[(i*h->num_poses+j)*12]);
}
}
}
else
{
struct iqmpose2 *p, *ipose = (struct iqmpose2*)(buffer + h->ofs_poses);
struct iqmjoint2 *ijoint = (struct iqmjoint2*)(buffer + h->ofs_joints);
vec3_t pos;
vec4_t quat;
vec3_t scale;
float mat[12];
//joint info (mesh)
for (i = 0; i < h->num_joints; i++)
{
Q_strncpyz(bones[i].name, strings+ijoint[i].name, sizeof(bones[i].name));
bones[i].parent = ijoint[i].parent;
GenMatrixPosQuat4Scale(ijoint[i].translate, ijoint[i].rotate, ijoint[i].scale, mat);
if (ijoint[i].parent >= 0)
Matrix3x4_Multiply(mat, &basepose[ijoint[i].parent*12], &basepose[i*12]);
else
memcpy(&basepose[i*12], mat, sizeof(mat));
Matrix3x4_Invert_Simple(&basepose[i*12], bones[i].inverse);
}
//pose info (anim)
for (i = 0; i < h->num_frames; i++)
{
for (j = 0, p = ipose; j < h->num_poses; j++, p++)
{
pos[0] = p->channeloffset[0]; if (p->mask & 1) pos[0] += *framedata++ * p->channelscale[0];
pos[1] = p->channeloffset[1]; if (p->mask & 2) pos[1] += *framedata++ * p->channelscale[1];
pos[2] = p->channeloffset[2]; if (p->mask & 4) pos[2] += *framedata++ * p->channelscale[2];
quat[0] = p->channeloffset[3]; if (p->mask & 8) quat[0] += *framedata++ * p->channelscale[3];
quat[1] = p->channeloffset[4]; if (p->mask & 16) quat[1] += *framedata++ * p->channelscale[4];
quat[2] = p->channeloffset[5]; if (p->mask & 32) quat[2] += *framedata++ * p->channelscale[5];
quat[3] = p->channeloffset[6]; if (p->mask & 64) quat[3] += *framedata++ * p->channelscale[6];
scale[0] = p->channeloffset[7]; if (p->mask & 128) scale[0] += *framedata++ * p->channelscale[7];
scale[1] = p->channeloffset[8]; if (p->mask & 256) scale[1] += *framedata++ * p->channelscale[8];
scale[2] = p->channeloffset[9]; if (p->mask & 512) scale[2] += *framedata++ * p->channelscale[9];
GenMatrixPosQuat4Scale(pos, quat, scale, &opose[(i*h->num_poses+j)*12]);
}
}
}
//basepose
memcpy(oposebase, basepose, sizeof(float)*12 * h->num_joints);
//now generate the animations.
for (i = 0; i < numgroups; i++)
{
if ((unsigned)framegroups[i].firstpose >= h->num_frames)
{
//invalid/basepose
fgroup[i].isheirachical = false;
fgroup[i].boneofs = oposebase;
fgroup[i].numposes = 1;
}
else
{
fgroup[i].isheirachical = true;
fgroup[i].boneofs = opose + framegroups[i].firstpose*12*h->num_poses;
fgroup[i].numposes = framegroups[i].posecount;
}
fgroup[i].loop = framegroups[i].loop;
fgroup[i].rate = framegroups[i].fps;
Q_strncpyz(fgroup[i].name, framegroups[i].name, sizeof(fgroup[i].name));
if (fgroup[i].rate <= 0)
fgroup[i].rate = 10;
}
free(framegroups);
for (i = 0; i < h->num_meshes; i++)
{
gai[i].nextsurf = (i == (h->num_meshes-1))?NULL:&gai[i+1];
/*animation info*/
gai[i].shares_bones = 0;
gai[i].numbones = h->num_joints;
gai[i].ofsbones = bones;
gai[i].groups = numgroups;
gai[i].groupofs = fgroup;
offset = LittleLong(mesh[i].first_vertex);
#ifndef SERVERONLY
/*skins*/
gai[i].numskins = 1;
gai[i].ofsskins = &skin[i];
Q_strncpyz(skin[i].name, strings+mesh[i].material, sizeof(skin[i].name));
skin[i].skinwidth = 1;
skin[i].skinheight = 1;
skin[i].ofstexels = 0; /*doesn't support 8bit colourmapping*/
skin[i].skinspeed = 10; /*something to avoid div by 0*/
skin[i].numshaders = 1;
skin[i].ofsshaders = &shaders[i];
shaders[i] = R_RegisterSkin(skin[i].name, mod->name);
R_BuildDefaultTexnums(NULL, shaders[i]);
if (shaders[i]->flags & SHADER_NOIMAGE)
Con_Printf("Unable to load texture for shader \"%s\" for model \"%s\"\n", shaders[i]->name, loadmodel->name);
gai[i].ofs_st_array = (otcoords+offset);
#endif
nt = LittleLong(mesh[i].num_triangles);
tris = (struct iqmtriangle*)(buffer + LittleLong(h->ofs_triangles));
tris += LittleLong(mesh[i].first_triangle);
gai[i].numindexes = nt*3;
idx = ZG_Malloc(&loadmodel->memgroup, sizeof(*idx)*gai[i].numindexes);
gai[i].ofs_indexes = idx;
for (t = 0; t < nt; t++)
{
*idx++ = LittleShort(tris[t].vertex[0]) - offset;
*idx++ = LittleShort(tris[t].vertex[1]) - offset;
*idx++ = LittleShort(tris[t].vertex[2]) - offset;
}
/*verts*/
gai[i].shares_verts = i;
gai[i].numverts = LittleLong(mesh[i].num_vertexes);
gai[i].ofs_skel_xyz = (opos+offset);
gai[i].ofs_skel_norm = vnorm?(onorm1+offset):NULL;
gai[i].ofs_skel_svect = (vnorm&&vtang)?(onorm2+offset):NULL;
gai[i].ofs_skel_tvect = (vnorm&&vtang)?(onorm3+offset):NULL;
gai[i].ofs_skel_idx = oindex?(oindex+offset):NULL;
gai[i].ofs_skel_weight = oweight?(oweight+offset):NULL;
}
if (!noweights)
{
for (i = 0; i < h->num_vertexes; i++)
{
Vector4Copy(vbone+i*4, oindex[i]);
Vector4Scale(vweight+i*4, 1/255.0, oweight[i]);
//FIXME: should we be normalising?
if (!oweight[i][0] && !oweight[i][1] && !oweight[i][2] && !oweight[i][3])
oweight[i][0] = 1;
}
}
for (i = 0; i < h->num_vertexes; i++)
{
Vector2Copy(tcoord+i*2, otcoords[i]);
VectorCopy(vpos+i*3, opos[i]);
if (vnorm)
{
VectorCopy(vnorm+i*3, onorm1[i]);
}
if (vnorm && vtang)
{
VectorCopy(vtang+i*4, onorm2[i]);
if(LittleFloat(vtang[i*4 + 3]) < 0)
CrossProduct(onorm2[i], onorm1[i], onorm3[i]);
else
CrossProduct(onorm1[i], onorm2[i], onorm3[i]);
}
}
return gai;
}
qboolean Mod_ParseIQMAnim(char *buffer, galiasinfo_t *prototype, void**poseofs, galiasgroup_t *gat)
{
return false;
}
qboolean Mod_LoadInterQuakeModel(model_t *mod, void *buffer)
{
int i;
galiasinfo_t *root;
struct iqmheader *h = (struct iqmheader *)buffer;
root = Mod_ParseIQMMeshModel(mod, buffer);
if (!root)
{
return false;
}
mod->flags = h->flags;
ClearBounds(mod->mins, mod->maxs);
for (i = 0; i < root->numverts; i++)
AddPointToBounds(root->ofs_skel_xyz[i], mod->mins, mod->maxs);
Mod_ClampModelSize(mod);
mod->meshinfo = root;
mod->type = mod_alias;
return true;
}
#endif
#ifdef MD5MODELS
qboolean Mod_ParseMD5Anim(char *buffer, galiasinfo_t *prototype, void**poseofs, galiasgroup_t *gat)
{
#define MD5ERROR0PARAM(x) { Con_Printf(CON_ERROR x "\n"); return false; }
#define MD5ERROR1PARAM(x, y) { Con_Printf(CON_ERROR x "\n", y); return false; }
#define EXPECT(x) buffer = COM_Parse(buffer); if (strcmp(com_token, x)) MD5ERROR1PARAM("MD5ANIM: expected %s", x);
unsigned int i, j;
galiasgroup_t grp;
unsigned int parent;
unsigned int numframes;
unsigned int numjoints;
float framespersecond;
unsigned int numanimatedparts;
galiasbone_t *bonelist;
unsigned char *boneflags;
unsigned int *firstanimatedcomponents;
float *animatedcomponents;
float *baseframe; //6 components.
float *posedata;
float tx, ty, tz, qx, qy, qz;
int fac, flags;
float f;
char com_token[8192];
EXPECT("MD5Version");
EXPECT("10");
EXPECT("commandline");
buffer = COM_Parse(buffer);
EXPECT("numFrames");
buffer = COM_Parse(buffer);
numframes = atoi(com_token);
EXPECT("numJoints");
buffer = COM_Parse(buffer);
numjoints = atoi(com_token);
EXPECT("frameRate");
buffer = COM_Parse(buffer);
framespersecond = atof(com_token);
EXPECT("numAnimatedComponents");
buffer = COM_Parse(buffer);
numanimatedparts = atoi(com_token);
firstanimatedcomponents = BZ_Malloc(sizeof(int)*numjoints);
animatedcomponents = BZ_Malloc(sizeof(float)*numanimatedparts);
boneflags = BZ_Malloc(sizeof(unsigned char)*numjoints);
baseframe = BZ_Malloc(sizeof(float)*12*numjoints);
*poseofs = posedata = ZG_Malloc(&loadmodel->memgroup, sizeof(float)*12*numjoints*numframes);
if (prototype->numbones)
{
if (prototype->numbones != numjoints)
MD5ERROR0PARAM("MD5ANIM: number of bones doesn't match");
bonelist = prototype->ofsbones;
}
else
{
bonelist = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasbone_t)*numjoints);
prototype->ofsbones = bonelist;
}
EXPECT("hierarchy");
EXPECT("{");
for (i = 0; i < numjoints; i++, bonelist++)
{
buffer = COM_Parse(buffer);
if (prototype->numbones)
{
if (strcmp(bonelist->name, com_token))
MD5ERROR1PARAM("MD5ANIM: bone name doesn't match (%s)", com_token);
}
else
Q_strncpyz(bonelist->name, com_token, sizeof(bonelist->name));
buffer = COM_Parse(buffer);
parent = atoi(com_token);
if (prototype->numbones)
{
if (bonelist->parent != parent)
MD5ERROR1PARAM("MD5ANIM: bone name doesn't match (%s)", com_token);
}
else
bonelist->parent = parent;
buffer = COM_Parse(buffer);
boneflags[i] = atoi(com_token);
buffer = COM_Parse(buffer);
firstanimatedcomponents[i] = atoi(com_token);
}
EXPECT("}");
if (!prototype->numbones)
prototype->numbones = numjoints;
EXPECT("bounds");
EXPECT("{");
for (i = 0; i < numframes; i++)
{
EXPECT("(");
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f < loadmodel->mins[0]) loadmodel->mins[0] = f;
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f < loadmodel->mins[1]) loadmodel->mins[1] = f;
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f < loadmodel->mins[2]) loadmodel->mins[2] = f;
EXPECT(")");
EXPECT("(");
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f > loadmodel->maxs[0]) loadmodel->maxs[0] = f;
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f > loadmodel->maxs[1]) loadmodel->maxs[1] = f;
buffer = COM_Parse(buffer);f=atoi(com_token);
if (f > loadmodel->maxs[2]) loadmodel->maxs[2] = f;
EXPECT(")");
}
EXPECT("}");
EXPECT("baseframe");
EXPECT("{");
for (i = 0; i < numjoints; i++)
{
EXPECT("(");
buffer = COM_Parse(buffer);
baseframe[i*6+0] = atof(com_token);
buffer = COM_Parse(buffer);
baseframe[i*6+1] = atof(com_token);
buffer = COM_Parse(buffer);
baseframe[i*6+2] = atof(com_token);
EXPECT(")");
EXPECT("(");
buffer = COM_Parse(buffer);
baseframe[i*6+3] = atof(com_token);
buffer = COM_Parse(buffer);
baseframe[i*6+4] = atof(com_token);
buffer = COM_Parse(buffer);
baseframe[i*6+5] = atof(com_token);
EXPECT(")");
}
EXPECT("}");
for (i = 0; i < numframes; i++)
{
EXPECT("frame");
EXPECT(va("%i", i));
EXPECT("{");
for (j = 0; j < numanimatedparts; j++)
{
buffer = COM_Parse(buffer);
animatedcomponents[j] = atof(com_token);
}
EXPECT("}");
for (j = 0; j < numjoints; j++)
{
fac = firstanimatedcomponents[j];
flags = boneflags[j];
if (flags&1)
tx = animatedcomponents[fac++];
else
tx = baseframe[j*6+0];
if (flags&2)
ty = animatedcomponents[fac++];
else
ty = baseframe[j*6+1];
if (flags&4)
tz = animatedcomponents[fac++];
else
tz = baseframe[j*6+2];
if (flags&8)
qx = animatedcomponents[fac++];
else
qx = baseframe[j*6+3];
if (flags&16)
qy = animatedcomponents[fac++];
else
qy = baseframe[j*6+4];
if (flags&32)
qz = animatedcomponents[fac++];
else
qz = baseframe[j*6+5];
GenMatrix(tx, ty, tz, qx, qy, qz, posedata+12*(j+numjoints*i));
}
}
BZ_Free(firstanimatedcomponents);
BZ_Free(animatedcomponents);
BZ_Free(boneflags);
BZ_Free(baseframe);
Q_strncpyz(grp.name, "", sizeof(grp.name));
grp.isheirachical = true;
grp.numposes = numframes;
grp.rate = framespersecond;
grp.loop = true;
*gat = grp;
return true;
#undef MD5ERROR0PARAM
#undef MD5ERROR1PARAM
#undef EXPECT
}
galiasinfo_t *Mod_ParseMD5MeshModel(char *buffer, char *modname)
{
#define MD5ERROR0PARAM(x) { Con_Printf(CON_ERROR x "\n"); return NULL; }
#define MD5ERROR1PARAM(x, y) { Con_Printf(CON_ERROR x "\n", y); return NULL; }
#define EXPECT(x) buffer = COM_Parse(buffer); if (strcmp(com_token, x)) Sys_Error("MD5MESH: expected %s", x);
int numjoints = 0;
int nummeshes = 0;
qboolean foundjoints = false;
int i;
galiasbone_t *bones = NULL;
galiasgroup_t *pose = NULL;
galiasinfo_t *inf, *root, *lastsurf;
float *posedata;
#ifndef SERVERONLY
galiasskin_t *skin;
shader_t **shaders;
#endif
char *filestart = buffer;
float x, y, z, qx, qy, qz;
buffer = COM_Parse(buffer);
if (strcmp(com_token, "MD5Version"))
MD5ERROR0PARAM("MD5 model without MD5Version identifier first");
buffer = COM_Parse(buffer);
if (atoi(com_token) != 10)
MD5ERROR0PARAM("MD5 model with unsupported MD5Version");
root = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasinfo_t));
lastsurf = NULL;
for(;;)
{
buffer = COM_Parse(buffer);
if (!buffer)
break;
if (!strcmp(com_token, "numFrames"))
{
void *poseofs;
galiasgroup_t *grp = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t));
Mod_ParseMD5Anim(filestart, root, &poseofs, grp);
root->groupofs = grp;
root->groups = 1;
grp->poseofs = poseofs;
return root;
}
else if (!strcmp(com_token, "commandline"))
{ //we don't need this
buffer = strchr(buffer, '\"');
buffer = strchr((char*)buffer+1, '\"')+1;
// buffer = COM_Parse(buffer);
}
else if (!strcmp(com_token, "numJoints"))
{
if (numjoints)
MD5ERROR0PARAM("MD5MESH: numMeshes was already declared");
buffer = COM_Parse(buffer);
numjoints = atoi(com_token);
if (numjoints <= 0)
MD5ERROR0PARAM("MD5MESH: Needs some joints");
}
else if (!strcmp(com_token, "numMeshes"))
{
if (nummeshes)
MD5ERROR0PARAM("MD5MESH: numMeshes was already declared");
buffer = COM_Parse(buffer);
nummeshes = atoi(com_token);
if (nummeshes <= 0)
MD5ERROR0PARAM("MD5MESH: Needs some meshes");
}
else if (!strcmp(com_token, "joints"))
{
if (foundjoints)
MD5ERROR0PARAM("MD5MESH: Duplicate joints section");
foundjoints=true;
if (!numjoints)
MD5ERROR0PARAM("MD5MESH: joints section before (or without) numjoints");
bones = ZG_Malloc(&loadmodel->memgroup, sizeof(*bones) * numjoints);
pose = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t));
posedata = ZG_Malloc(&loadmodel->memgroup, sizeof(float)*12 * numjoints);
pose->isheirachical = false;
pose->rate = 1;
pose->numposes = 1;
pose->boneofs = posedata;
Q_strncpyz(pose->name, "base", sizeof(pose->name));
EXPECT("{");
//"name" parent (x y z) (s t u)
//stu are a normalized quaternion, which we will convert to a 3*4 matrix for no apparent reason
for (i = 0; i < numjoints; i++)
{
buffer = COM_Parse(buffer);
Q_strncpyz(bones[i].name, com_token, sizeof(bones[i].name));
buffer = COM_Parse(buffer);
bones[i].parent = atoi(com_token);
if (bones[i].parent >= i)
MD5ERROR0PARAM("MD5MESH: joints parent's must be lower");
if ((bones[i].parent < 0 && i) || (!i && bones[i].parent!=-1))
MD5ERROR0PARAM("MD5MESH: Only the root joint may have a negative parent");
EXPECT("(");
buffer = COM_Parse(buffer);
x = atof(com_token);
buffer = COM_Parse(buffer);
y = atof(com_token);
buffer = COM_Parse(buffer);
z = atof(com_token);
EXPECT(")");
EXPECT("(");
buffer = COM_Parse(buffer);
qx = atof(com_token);
buffer = COM_Parse(buffer);
qy = atof(com_token);
buffer = COM_Parse(buffer);
qz = atof(com_token);
EXPECT(")");
GenMatrix(x, y, z, qx, qy, qz, posedata+i*12);
}
EXPECT("}");
}
else if (!strcmp(com_token, "mesh"))
{
int numverts = 0;
int numweights = 0;
int numtris = 0;
int num;
int vnum;
int numusableweights = 0;
int *firstweightlist = NULL;
int *numweightslist = NULL;
galisskeletaltransforms_t *trans;
#ifndef SERVERONLY
float *stcoord = NULL;
#endif
index_t *indexes = NULL;
float w;
vec4_t *rawweight = NULL;
int *rawweightbone = NULL;
if (!nummeshes)
MD5ERROR0PARAM("MD5MESH: mesh section before (or without) nummeshes");
if (!foundjoints || !bones || !pose)
MD5ERROR0PARAM("MD5MESH: mesh must come after joints");
if (!lastsurf)
{
lastsurf = root;
inf = root;
}
else
{
inf = ZG_Malloc(&loadmodel->memgroup, sizeof(*inf));
lastsurf->nextsurf = inf;
lastsurf = inf;
}
inf->ofsbones = bones;
inf->numbones = numjoints;
inf->groups = 1;
inf->groupofs = pose;
inf->baseframeofs = pose->boneofs;
#ifndef SERVERONLY
skin = ZG_Malloc(&loadmodel->memgroup, sizeof(*skin));
shaders = ZG_Malloc(&loadmodel->memgroup, sizeof(*shaders));
inf->numskins = 1;
inf->ofsskins = skin;
skin->numshaders = 1;
skin->skinspeed = 1;
skin->ofsshaders = shaders;
#endif
EXPECT("{");
for(;;)
{
buffer = COM_Parse(buffer);
if (!buffer)
MD5ERROR0PARAM("MD5MESH: unexpected eof");
if (!strcmp(com_token, "shader"))
{
buffer = COM_Parse(buffer);
#ifndef SERVERONLY
//FIXME: we probably want to support multiple skins some time
shaders[0] = R_RegisterSkin(com_token, modname);
R_BuildDefaultTexnums(NULL, shaders[0]);
if (shaders[0]->flags & SHADER_NOIMAGE)
Con_Printf("Unable to load texture for shader \"%s\" for model \"%s\"\n", shaders[0]->name, loadmodel->name);
#endif
}
else if (!strcmp(com_token, "numverts"))
{
if (numverts)
MD5ERROR0PARAM("MD5MESH: numverts was already specified");
buffer = COM_Parse(buffer);
numverts = atoi(com_token);
if (numverts < 0)
MD5ERROR0PARAM("MD5MESH: numverts cannot be negative");
firstweightlist = Z_Malloc(sizeof(*firstweightlist) * numverts);
numweightslist = Z_Malloc(sizeof(*numweightslist) * numverts);
#ifndef SERVERONLY
stcoord = ZG_Malloc(&loadmodel->memgroup, sizeof(float)*2*numverts);
inf->ofs_st_array = (vec2_t*)stcoord;
inf->numverts = numverts;
#endif
}
else if (!strcmp(com_token, "vert"))
{ //vert num ( s t ) firstweight numweights
buffer = COM_Parse(buffer);
num = atoi(com_token);
if (num < 0 || num >= numverts)
MD5ERROR0PARAM("MD5MESH: vertex out of range");
EXPECT("(");
buffer = COM_Parse(buffer);
#ifndef SERVERONLY
if (!stcoord)
MD5ERROR0PARAM("MD5MESH: vertex out of range");
stcoord[num*2+0] = atof(com_token);
#endif
buffer = COM_Parse(buffer);
#ifndef SERVERONLY
stcoord[num*2+1] = atof(com_token);
#endif
EXPECT(")");
buffer = COM_Parse(buffer);
firstweightlist[num] = atoi(com_token);
buffer = COM_Parse(buffer);
numweightslist[num] = atoi(com_token);
numusableweights += numweightslist[num];
}
else if (!strcmp(com_token, "numtris"))
{
if (numtris)
MD5ERROR0PARAM("MD5MESH: numtris was already specified");
buffer = COM_Parse(buffer);
numtris = atoi(com_token);
if (numtris < 0)
MD5ERROR0PARAM("MD5MESH: numverts cannot be negative");
indexes = ZG_Malloc(&loadmodel->memgroup, sizeof(int)*3*numtris);
inf->ofs_indexes = indexes;
inf->numindexes = numtris*3;
}
else if (!strcmp(com_token, "tri"))
{
buffer = COM_Parse(buffer);
num = atoi(com_token);
if (num < 0 || num >= numtris)
MD5ERROR0PARAM("MD5MESH: vertex out of range");
buffer = COM_Parse(buffer);
indexes[num*3+0] = atoi(com_token);
buffer = COM_Parse(buffer);
indexes[num*3+1] = atoi(com_token);
buffer = COM_Parse(buffer);
indexes[num*3+2] = atoi(com_token);
}
else if (!strcmp(com_token, "numweights"))
{
if (numweights)
MD5ERROR0PARAM("MD5MESH: numweights was already specified");
buffer = COM_Parse(buffer);
numweights = atoi(com_token);
rawweight = Z_Malloc(sizeof(*rawweight)*numweights);
rawweightbone = Z_Malloc(sizeof(*rawweightbone)*numweights);
}
else if (!strcmp(com_token, "weight"))
{
//weight num bone scale ( x y z )
buffer = COM_Parse(buffer);
num = atoi(com_token);
if (num < 0 || num >= numweights)
MD5ERROR0PARAM("MD5MESH: weight out of range");
buffer = COM_Parse(buffer);
rawweightbone[num] = atoi(com_token);
if (rawweightbone[num] < 0 || rawweightbone[num] >= numjoints)
MD5ERROR0PARAM("MD5MESH: weight specifies bad bone");
buffer = COM_Parse(buffer);
w = atof(com_token);
EXPECT("(");
buffer = COM_Parse(buffer);
rawweight[num][0] = w*atof(com_token);
buffer = COM_Parse(buffer);
rawweight[num][1] = w*atof(com_token);
buffer = COM_Parse(buffer);
rawweight[num][2] = w*atof(com_token);
EXPECT(")");
rawweight[num][3] = w;
}
else if (!strcmp(com_token, "}"))
break;
else
MD5ERROR1PARAM("MD5MESH: Unrecognised token inside mesh (%s)", com_token);
}
trans = ZG_Malloc(&loadmodel->memgroup, sizeof(*trans)*numusableweights);
inf->ofsswtransforms = trans;
for (num = 0, vnum = 0; num < numverts; num++)
{
if (numweightslist[num] <= 0)
MD5ERROR0PARAM("MD5MESH: weights not set on vertex");
while(numweightslist[num])
{
trans[vnum].vertexindex = num;
trans[vnum].boneindex = rawweightbone[firstweightlist[num]];
trans[vnum].org[0] = rawweight[firstweightlist[num]][0];
trans[vnum].org[1] = rawweight[firstweightlist[num]][1];
trans[vnum].org[2] = rawweight[firstweightlist[num]][2];
trans[vnum].org[3] = rawweight[firstweightlist[num]][3];
vnum++;
firstweightlist[num]++;
numweightslist[num]--;
}
}
inf->numswtransforms = vnum;
if (firstweightlist)
Z_Free(firstweightlist);
if (numweightslist)
Z_Free(numweightslist);
if (rawweight)
Z_Free(rawweight);
if (rawweightbone)
Z_Free(rawweightbone);
}
else
MD5ERROR1PARAM("Unrecognised token in MD5 model (%s)", com_token);
}
if (!lastsurf)
MD5ERROR0PARAM("MD5MESH: No meshes");
Alias_CalculateSkeletalNormals(root);
return root;
#undef MD5ERROR0PARAM
#undef MD5ERROR1PARAM
#undef EXPECT
}
qboolean QDECL Mod_LoadMD5MeshModel(model_t *mod, void *buffer)
{
galiasinfo_t *root;
loadmodel=mod;
root = Mod_ParseMD5MeshModel(buffer, mod->name);
if (root == NULL)
{
return false;
}
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0); //file replacement - inherit flags from any defunc mdl files.
Mod_ClampModelSize(mod);
mod->type = mod_alias;
mod->meshinfo = root;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
/*
EXTERNALANIM
//File that specifies md5 model/anim stuff.
model test/imp.md5mesh
group test/idle1.md5anim
clampgroup test/idle1.md5anim
frames test/idle1.md5anim
*/
qboolean QDECL Mod_LoadCompositeAnim(model_t *mod, void *buffer)
{
int i;
char *file;
galiasinfo_t *root = NULL, *surf;
int numgroups = 0;
galiasgroup_t *grouplist = NULL;
galiasgroup_t *newgroup = NULL;
float **poseofs;
char com_token[8192];
loadmodel=mod;
buffer = COM_Parse(buffer);
if (strcmp(com_token, "EXTERNALANIM"))
{
Con_Printf (CON_ERROR "EXTERNALANIM: header is not compleate (%s)\n", mod->name);
return false;
}
buffer = COM_Parse(buffer);
if (!strcmp(com_token, "model"))
{
buffer = COM_Parse(buffer);
file = COM_LoadTempMoreFile(com_token);
if (!file) //FIXME: make non fatal somehow..
{
Con_Printf(CON_ERROR "Couldn't open %s (from %s)\n", com_token, mod->name);
return false;
}
root = Mod_ParseMD5MeshModel(file, mod->name);
if (root == NULL)
{
return false;
}
newgroup = root->groupofs;
grouplist = BZ_Malloc(sizeof(galiasgroup_t)*(numgroups+root->groups));
memcpy(grouplist, newgroup, sizeof(galiasgroup_t)*(numgroups+root->groups));
poseofs = BZ_Malloc(sizeof(galiasgroup_t)*(numgroups+root->groups));
for (i = 0; i < root->groups; i++)
{
grouplist[numgroups] = newgroup[i];
poseofs[numgroups] = newgroup[i].boneofs;
numgroups++;
}
}
else
{
Con_Printf (CON_ERROR "EXTERNALANIM: model must be defined immediatly after the header\n");
return false;
}
for (;;)
{
buffer = COM_Parse(buffer);
if (!buffer)
break;
if (!strcmp(com_token, "group"))
{
grouplist = BZ_Realloc(grouplist, sizeof(galiasgroup_t)*(numgroups+1));
poseofs = BZ_Realloc(poseofs, sizeof(*poseofs)*(numgroups+1));
buffer = COM_Parse(buffer);
file = COM_LoadTempMoreFile(com_token);
if (file) //FIXME: make non fatal somehow..
{
char namebkup[MAX_QPATH];
Q_strncpyz(namebkup, com_token, sizeof(namebkup));
if (!Mod_ParseMD5Anim(file, root, &poseofs[numgroups], &grouplist[numgroups]))
{
return false;
}
Q_strncpyz(grouplist[numgroups].name, namebkup, sizeof(grouplist[numgroups].name));
numgroups++;
}
}
else if (!strcmp(com_token, "clampgroup"))
{
grouplist = BZ_Realloc(grouplist, sizeof(galiasgroup_t)*(numgroups+1));
poseofs = BZ_Realloc(poseofs, sizeof(*poseofs)*(numgroups+1));
buffer = COM_Parse(buffer);
file = COM_LoadTempMoreFile(com_token);
if (file) //FIXME: make non fatal somehow..
{
char namebkup[MAX_QPATH];
Q_strncpyz(namebkup, com_token, sizeof(namebkup));
if (!Mod_ParseMD5Anim(file, root, &poseofs[numgroups], &grouplist[numgroups]))
{
return false;
}
Q_strncpyz(grouplist[numgroups].name, namebkup, sizeof(grouplist[numgroups].name));
grouplist[numgroups].loop = false;
numgroups++;
}
}
else if (!strcmp(com_token, "frames"))
{
galiasgroup_t ng;
void *np;
buffer = COM_Parse(buffer);
file = COM_LoadTempMoreFile(com_token);
if (file) //FIXME: make non fatal somehow..
{
char namebkup[MAX_QPATH];
Q_strncpyz(namebkup, com_token, sizeof(namebkup));
if (!Mod_ParseMD5Anim(file, root, &np, &ng))
{
return false;
}
grouplist = BZ_Realloc(grouplist, sizeof(galiasgroup_t)*(numgroups+ng.numposes));
poseofs = BZ_Realloc(poseofs, sizeof(*poseofs)*(numgroups+ng.numposes));
//pull out each frame individually
for (i = 0; i < ng.numposes; i++)
{
grouplist[numgroups].isheirachical = ng.isheirachical;
grouplist[numgroups].loop = false;
grouplist[numgroups].numposes = 1;
grouplist[numgroups].rate = 24;
poseofs[numgroups] = (float*)np + i*12*root->numbones;
Q_snprintfz(grouplist[numgroups].name, sizeof(grouplist[numgroups].name), "%s%i", namebkup, i);
Q_strncpyz(grouplist[numgroups].name, namebkup, sizeof(grouplist[numgroups].name));
grouplist[numgroups].loop = false;
numgroups++;
}
}
}
else
{
Con_Printf(CON_ERROR "EXTERNALANIM: unrecognised token (%s)\n", mod->name);
return false;
}
}
newgroup = grouplist;
grouplist = ZG_Malloc(&loadmodel->memgroup, sizeof(galiasgroup_t)*numgroups);
for(surf = root;;)
{
surf->groupofs = grouplist;
surf->groups = numgroups;
if (!surf->nextsurf)
break;
surf = surf->nextsurf;
}
for (i = 0; i < numgroups; i++)
{
grouplist[i] = newgroup[i];
grouplist[i].boneofs = poseofs[i];
}
mod->flags = Mod_ReadFlagsFromMD1(mod->name, 0); //file replacement - inherit flags from any defunc mdl files.
Mod_ClampModelSize(mod);
mod->type = mod_alias;
mod->meshinfo = root;
mod->funcs.NativeTrace = Mod_Trace;
return true;
}
#endif //MD5MODELS
#else
int Mod_TagNumForName(model_t *model, char *name)
{
return 0;
}
qboolean Mod_GetTag(model_t *model, int tagnum, framestate_t *framestate, float *result)
{
return false;
}
int Mod_GetNumBones(struct model_s *model, qboolean allowtags)
{
return 0;
}
int Mod_GetBoneRelations(model_t *model, int firstbone, int lastbone, framestate_t *fstate, float *result)
{
return 0;
}
int Mod_GetBoneParent(struct model_s *model, int bonenum)
{
return 0;
}
char *Mod_GetBoneName(struct model_s *model, int bonenum)
{
return "";
}
#endif //#if defined(D3DQUAKE) || defined(GLQUAKE)
void Alias_Register(void)
{
Mod_RegisterModelFormatMagic(NULL, "Quake1 Model (mdl)", IDPOLYHEADER, Mod_LoadQ1Model);
Mod_RegisterModelFormatMagic(NULL, "Hexen2 Model (mdl)", RAPOLYHEADER, Mod_LoadQ1Model);
#ifdef MD2MODELS
Mod_RegisterModelFormatMagic(NULL, "Quake2 Model (md2)", MD2IDALIASHEADER, Mod_LoadQ2Model);
#endif
#ifdef MD3MODELS
Mod_RegisterModelFormatMagic(NULL, "Quake3 Model (md3)", MD3_IDENT, Mod_LoadQ3Model);
#endif
#ifdef HALFLIFEMODELS
Mod_RegisterModelFormatMagic(NULL, "Half-Life Model (mdl)", (('T'<<24)+('S'<<16)+('D'<<8)+'I'), Mod_LoadHLModel);
#endif
#ifdef ZYMOTICMODELS
Mod_RegisterModelFormatMagic(NULL, "Zymotic Model (zym)", (('O'<<24)+('M'<<16)+('Y'<<8)+'Z'), Mod_LoadZymoticModel);
#endif
#ifdef DPMMODELS
Mod_RegisterModelFormatMagic(NULL, "DarkPlaces Model (dpm)", (('K'<<24)+('R'<<16)+('A'<<8)+'D'), Mod_LoadDarkPlacesModel);
#endif
#ifdef PSKMODELS
Mod_RegisterModelFormatMagic(NULL, "Unreal Interchange Model (psk)", ('A'<<0)+('C'<<8)+('T'<<16)+('R'<<24), Mod_LoadPSKModel);
#endif
#ifdef INTERQUAKEMODELS
Mod_RegisterModelFormatMagic(NULL, "Inter-Quake Model (iqm)", ('I'<<0)+('N'<<8)+('T'<<16)+('E'<<24), Mod_LoadInterQuakeModel);
#endif
#ifdef MD5MODELS
Mod_RegisterModelFormatText(NULL, "MD5 Mesh/Anim (md5mesh)", "MD5Version", Mod_LoadMD5MeshModel);
Mod_RegisterModelFormatText(NULL, "External Anim", "EXTERNALANIM", Mod_LoadCompositeAnim);
#endif
}