1
0
Fork 0
forked from fte/fteqw
fteqw/engine/gl/gl_alias.c
Spoike f33873ddb7 committing for fixes for OMC
some minor changes. Mostly bug fixes and internal reorganisation.
Added code to provide an activex control as part of the npfte.dll plugin. If the dll is registered the regsvr32 way, the plugin can be used with IE as well.
fisheye/panoramic view enable is now controlled by rulesets instead of serverinfo.
server will list all pak files it has loaded. client will probably do the wrong thing and still needs fixing properly.

git-svn-id: https://svn.code.sf.net/p/fteqw/code/trunk@3910 fc73d0e0-1445-4013-8a0c-d673dee63da5
2011-10-27 16:16:29 +00:00

1999 lines
49 KiB
C
Raw Blame History

//a note about dedicated servers:
//In the server-side gamecode, a couple of q1 extensions require knowing something about models.
//So we load models serverside, if required.
//things we need:
//tag/bone names and indexes so we can have reasonable modding with tags. :)
//tag/bone positions so we can shoot from the actual gun or other funky stuff
//vertex positions so we can trace against the mesh rather than the bbox.
//we use the gl renderer's model code because it supports more sorts of models than the sw renderer. Sad but true.
#include "quakedef.h"
#include "glquake.h"
#if defined(GLQUAKE) || defined(D3DQUAKE)
#ifdef _WIN32
#include <malloc.h>
#else
#include <alloca.h>
#endif
#include "com_mesh.h"
//FIXME
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;
extern cvar_t gl_part_flame, r_fullbrightSkins, r_fb_models;
extern cvar_t r_noaliasshadows;
extern char loadname[32]; // for hunk tags
extern cvar_t gl_ati_truform;
extern cvar_t r_vertexdlights;
extern cvar_t mod_md3flags;
extern cvar_t r_skin_overlays;
#ifndef SERVERONLY
static hashtable_t skincolourmapped;
//changes vertex lighting values
#if 0
static void R_GAliasApplyLighting(mesh_t *mesh, vec3_t org, vec3_t angles, float *colormod)
{
int l, v;
vec3_t rel;
vec3_t dir;
float dot, d, a;
if (mesh->colors4f_array)
{
float l;
int temp;
int i;
avec4_t *colours = mesh->colors4f_array;
vec3_t *normals = mesh->normals_array;
vec3_t ambient, shade;
qbyte alphab = bound(0, colormod[3], 1);
if (!mesh->normals_array)
{
mesh->colors4f_array = NULL;
return;
}
VectorCopy(ambientlight, ambient);
VectorCopy(shadelight, shade);
for (i = 0; i < 3; i++)
{
ambient[i] *= colormod[i];
shade[i] *= colormod[i];
}
for (i = mesh->numvertexes-1; i >= 0; i--)
{
l = DotProduct(normals[i], shadevector);
temp = l*ambient[0]+shade[0];
colours[i][0] = temp;
temp = l*ambient[1]+shade[1];
colours[i][1] = temp;
temp = l*ambient[2]+shade[2];
colours[i][2] = temp;
colours[i][3] = alphab;
}
}
if (r_vertexdlights.value && mesh->colors4f_array)
{
//don't include world lights
for (l=rtlights_first ; l<RTL_FIRST; l++)
{
if (cl_dlights[l].radius)
{
VectorSubtract (cl_dlights[l].origin,
org,
dir);
if (Length(dir)>cl_dlights[l].radius+mesh->radius) //far out man!
continue;
rel[0] = -DotProduct(dir, currententity->axis[0]);
rel[1] = -DotProduct(dir, currententity->axis[1]); //quake's crazy.
rel[2] = -DotProduct(dir, currententity->axis[2]);
/*
glBegin(GL_LINES);
glVertex3f(0,0,0);
glVertex3f(rel[0],rel[1],rel[2]);
glEnd();
*/
for (v = 0; v < mesh->numvertexes; v++)
{
VectorSubtract(mesh->xyz_array[v], rel, dir);
dot = DotProduct(dir, mesh->normals_array[v]);
if (dot>0)
{
d = DotProduct(dir, dir);
a = 1/d;
if (a>0)
{
a *= 10000000*dot/sqrt(d);
mesh->colors4f_array[v][0] += a*cl_dlights[l].color[0];
mesh->colors4f_array[v][1] += a*cl_dlights[l].color[1];
mesh->colors4f_array[v][2] += a*cl_dlights[l].color[2];
}
// else
// mesh->colors4f_array[v][1] = 1;
}
// else
// mesh->colors4f_array[v][2] = 1;
}
}
}
}
}
#endif
void GL_GAliasFlushSkinCache(void)
{
int i;
bucket_t *b;
for (i = 0; i < skincolourmapped.numbuckets; i++)
{
while((b = skincolourmapped.bucket[i]))
{
skincolourmapped.bucket[i] = b->next;
BZ_Free(b->data);
}
}
if (skincolourmapped.bucket)
BZ_Free(skincolourmapped.bucket);
skincolourmapped.bucket = NULL;
skincolourmapped.numbuckets = 0;
}
static texnums_t *GL_ChooseSkin(galiasinfo_t *inf, model_t *model, int surfnum, entity_t *e)
{
galiasskin_t *skins;
texnums_t *texnums;
int frame;
unsigned int subframe;
extern int cl_playerindex; //so I don't have to strcmp
unsigned int tc, bc, pc;
qboolean forced;
if (e->skinnum >= 100 && e->skinnum < 110)
{
shader_t *s;
s = R_RegisterSkin(va("gfx/skin%d.lmp", e->skinnum), NULL);
if (!TEXVALID(s->defaulttextures.base))
s->defaulttextures.base = R_LoadHiResTexture(va("gfx/skin%d.lmp", e->skinnum), NULL, 0);
s->defaulttextures.shader = s;
return &s->defaulttextures;
}
if ((e->model->engineflags & MDLF_NOTREPLACEMENTS) && !ruleset_allow_sensative_texture_replacements.ival)
forced = true;
else
forced = false;
if (!gl_nocolors.ival || forced)
{
if (e->scoreboard)
{
if (!e->scoreboard->skin)
Skin_Find(e->scoreboard);
tc = e->scoreboard->ttopcolor;
bc = e->scoreboard->tbottomcolor;
pc = e->scoreboard->h2playerclass;
}
else
{
tc = 1;
bc = 1;
pc = 0;
}
if (forced || tc != 1 || bc != 1 || (e->scoreboard && e->scoreboard->skin))
{
int inwidth, inheight;
int tinwidth, tinheight;
char *skinname;
qbyte *original;
galiascolourmapped_t *cm;
char hashname[512];
// if (e->scoreboard->skin->cachedbpp
/* if (cls.protocol == CP_QUAKE2)
{
if (e->scoreboard && e->scoreboard->skin)
snprintf(hashname, sizeof(hashname), "%s$%s$%i", modelname, e->scoreboard->skin->name, surfnum);
else
snprintf(hashname, sizeof(hashname), "%s$%i", modelname, surfnum);
skinname = hashname;
}
else */
{
if (e->scoreboard && e->scoreboard->skin)
{
snprintf(hashname, sizeof(hashname), "%s$%s$%i", model->name, e->scoreboard->skin->name, surfnum);
skinname = hashname;
}
else if (surfnum)
{
snprintf(hashname, sizeof(hashname), "%s$%i", model->name, surfnum);
skinname = hashname;
}
else
skinname = model->name;
}
if (!skincolourmapped.numbuckets)
{
void *buckets = BZ_Malloc(Hash_BytesForBuckets(256));
memset(buckets, 0, Hash_BytesForBuckets(256));
Hash_InitTable(&skincolourmapped, 256, buckets);
}
if (!inf->numskins)
{
skins = NULL;
subframe = 0;
texnums = NULL;
}
else
{
skins = (galiasskin_t*)((char *)inf + inf->ofsskins);
if (!skins->texnums)
{
skins = NULL;
subframe = 0;
texnums = NULL;
}
else
{
if (e->skinnum >= 0 && e->skinnum < inf->numskins)
skins += e->skinnum;
subframe = cl.time*skins->skinspeed;
subframe = subframe%skins->texnums;
texnums = (texnums_t*)((char *)skins + skins->ofstexnums + subframe*sizeof(texnums_t));
}
}
for (cm = Hash_Get(&skincolourmapped, skinname); cm; cm = Hash_GetNext(&skincolourmapped, skinname, cm))
{
if (cm->tcolour == tc && cm->bcolour == bc && cm->skinnum == e->skinnum && cm->subframe == subframe && cm->pclass == pc)
{
return &cm->texnum;
}
}
//colourmap isn't present yet.
cm = BZ_Malloc(sizeof(*cm));
Q_strncpyz(cm->name, skinname, sizeof(cm->name));
Hash_Add(&skincolourmapped, cm->name, cm, &cm->bucket);
cm->tcolour = tc;
cm->bcolour = bc;
cm->pclass = pc;
cm->skinnum = e->skinnum;
cm->subframe = subframe;
cm->texnum.fullbright = r_nulltex;
cm->texnum.base = r_nulltex;
cm->texnum.loweroverlay = r_nulltex;
cm->texnum.upperoverlay = r_nulltex;
cm->texnum.shader = texnums?texnums->shader:R_RegisterSkin(skinname, NULL);
if (!texnums)
{ //load just the skin
if (e->scoreboard && e->scoreboard->skin)
{
if (cls.protocol == CP_QUAKE2)
{
original = Skin_Cache32(e->scoreboard->skin);
if (original)
{
inwidth = e->scoreboard->skin->width;
inheight = e->scoreboard->skin->height;
cm->texnum.base = R_LoadTexture32(e->scoreboard->skin->name, inwidth, inheight, (unsigned int*)original, IF_NOALPHA|IF_NOGAMMA);
return &cm->texnum;
}
}
else
{
original = Skin_Cache8(e->scoreboard->skin);
if (original)
{
inwidth = e->scoreboard->skin->width;
inheight = e->scoreboard->skin->height;
cm->texnum.base = R_LoadTexture8(e->scoreboard->skin->name, inwidth, inheight, original, IF_NOALPHA|IF_NOGAMMA, 1);
return &cm->texnum;
}
}
if (TEXVALID(e->scoreboard->skin->tex_base))
{
texnums = &cm->texnum;
texnums->loweroverlay = e->scoreboard->skin->tex_lower;
texnums->upperoverlay = e->scoreboard->skin->tex_upper;
texnums->base = e->scoreboard->skin->tex_base;
return texnums;
}
cm->texnum.base = R_LoadHiResTexture(e->scoreboard->skin->name, "skins", IF_NOALPHA);
return &cm->texnum;
}
return NULL;
}
cm->texnum.bump = texnums[cm->skinnum].bump; //can't colour bumpmapping
if (cls.protocol != CP_QUAKE2 && ((!texnums || (model==cl.model_precache[cl_playerindex] || model==cl.model_precache_vwep[0])) && e->scoreboard && e->scoreboard->skin))
{
original = Skin_Cache8(e->scoreboard->skin);
inwidth = e->scoreboard->skin->width;
inheight = e->scoreboard->skin->height;
if (!original && TEXVALID(e->scoreboard->skin->tex_base))
{
texnums = &cm->texnum;
texnums->loweroverlay = e->scoreboard->skin->tex_lower;
texnums->upperoverlay = e->scoreboard->skin->tex_upper;
texnums->base = e->scoreboard->skin->tex_base;
return texnums;
}
}
else
{
original = NULL;
inwidth = 0;
inheight = 0;
}
if (!original)
{
if (skins->ofstexels)
{
original = (qbyte *)skins + skins->ofstexels;
inwidth = skins->skinwidth;
inheight = skins->skinheight;
}
else
{
original = NULL;
inwidth = 0;
inheight = 0;
}
}
tinwidth = skins->skinwidth;
tinheight = skins->skinheight;
if (original)
{
int i, j;
unsigned translate32[256];
static unsigned pixels[512*512];
unsigned *out;
unsigned frac, fracstep;
unsigned scaled_width, scaled_height;
qbyte *inrow;
texnums = &cm->texnum;
texnums->base = r_nulltex;
texnums->fullbright = r_nulltex;
scaled_width = gl_max_size.value < 512 ? gl_max_size.value : 512;
scaled_height = gl_max_size.value < 512 ? gl_max_size.value : 512;
//handle the case of an external skin being smaller than the texture that its meant to replace
//(to support the evil hackage of the padding on the outside of common qw skins)
if (tinwidth > inwidth)
tinwidth = inwidth;
if (tinheight > inheight)
tinheight = inheight;
//don't make scaled width any larger than it needs to be
for (i = 0; i < 10; i++)
{
scaled_width = (1<<i);
if (scaled_width >= tinwidth)
break; //its covered
}
if (scaled_width > gl_max_size.value)
scaled_width = gl_max_size.value; //whoops, we made it too big
for (i = 0; i < 10; i++)
{
scaled_height = (1<<i);
if (scaled_height >= tinheight)
break; //its covered
}
if (scaled_height > gl_max_size.value)
scaled_height = gl_max_size.value; //whoops, we made it too big
if (scaled_width < 4)
scaled_width = 4;
if (scaled_height < 4)
scaled_height = 4;
if (h2playertranslations && pc)
{
unsigned int color_offsets[5] = {2*14*256,0,1*14*256,2*14*256,2*14*256};
unsigned char *colorA, *colorB, *sourceA, *sourceB;
colorA = h2playertranslations + 256 + color_offsets[pc-1];
colorB = colorA + 256;
sourceA = colorB + (tc * 256);
sourceB = colorB + (bc * 256);
for(i=0;i<256;i++)
{
translate32[i] = d_8to24rgbtable[i];
if (tc > 0 && (colorA[i] != 255))
translate32[i] = d_8to24rgbtable[sourceA[i]];
if (bc > 0 && (colorB[i] != 255))
translate32[i] = d_8to24rgbtable[sourceB[i]];
}
translate32[0] = 0;
}
else
{
for (i=0 ; i<256 ; i++)
translate32[i] = d_8to24rgbtable[i];
for (i = 0; i < 16; i++)
{
if (tc >= 16)
{
//assumption: row 0 is pure white.
*((unsigned char*)&translate32[TOP_RANGE+i]+0) = (((tc&0xff0000)>>16)**((unsigned char*)&d_8to24rgbtable[i]+0))>>8;
*((unsigned char*)&translate32[TOP_RANGE+i]+1) = (((tc&0x00ff00)>> 8)**((unsigned char*)&d_8to24rgbtable[i]+1))>>8;
*((unsigned char*)&translate32[TOP_RANGE+i]+2) = (((tc&0x0000ff)>> 0)**((unsigned char*)&d_8to24rgbtable[i]+2))>>8;
*((unsigned char*)&translate32[TOP_RANGE+i]+3) = 0xff;
}
else
{
if (tc < 8)
translate32[TOP_RANGE+i] = d_8to24rgbtable[(tc<<4)+i];
else
translate32[TOP_RANGE+i] = d_8to24rgbtable[(tc<<4)+15-i];
}
if (bc >= 16)
{
*((unsigned char*)&translate32[BOTTOM_RANGE+i]+0) = (((bc&0xff0000)>>16)**((unsigned char*)&d_8to24rgbtable[i]+0))>>8;
*((unsigned char*)&translate32[BOTTOM_RANGE+i]+1) = (((bc&0x00ff00)>> 8)**((unsigned char*)&d_8to24rgbtable[i]+1))>>8;
*((unsigned char*)&translate32[BOTTOM_RANGE+i]+2) = (((bc&0x0000ff)>> 0)**((unsigned char*)&d_8to24rgbtable[i]+2))>>8;
*((unsigned char*)&translate32[BOTTOM_RANGE+i]+3) = 0xff;
}
else
{
if (bc < 8)
translate32[BOTTOM_RANGE+i] = d_8to24rgbtable[(bc<<4)+i];
else
translate32[BOTTOM_RANGE+i] = d_8to24rgbtable[(bc<<4)+15-i];
}
}
}
out = pixels;
fracstep = tinwidth*0x10000/scaled_width;
for (i=0 ; i<scaled_height ; i++, out += scaled_width)
{
inrow = original + inwidth*(i*inheight/scaled_height);
frac = fracstep >> 1;
for (j=0 ; j<scaled_width ; j+=4)
{
out[j] = translate32[inrow[frac>>16]];
frac += fracstep;
out[j+1] = translate32[inrow[frac>>16]];
frac += fracstep;
out[j+2] = translate32[inrow[frac>>16]];
frac += fracstep;
out[j+3] = translate32[inrow[frac>>16]];
frac += fracstep;
}
}
if (qrenderer == QR_OPENGL)
{
texnums->base = R_AllocNewTexture(cm->name, scaled_width, scaled_height);
R_Upload(texnums->base, cm->name, h2playertranslations?TF_RGBA32:TF_RGBX32, pixels, NULL, scaled_width, scaled_height, IF_NOMIPMAP);
}
else
{
texnums->base = R_LoadTexture(cm->name, scaled_width, scaled_height, h2playertranslations?TF_RGBA32:TF_RGBX32, pixels, 0);
}
if (!h2playertranslations)
{
//now do the fullbrights.
out = pixels;
fracstep = tinwidth*0x10000/scaled_width;
for (i=0 ; i<scaled_height ; i++, out += scaled_width)
{
inrow = original + inwidth*(i*inheight/scaled_height);
frac = fracstep >> 1;
for (j=0 ; j<scaled_width ; j+=1)
{
if (inrow[frac>>16] < 255-vid.fullbright)
((char *) (&out[j]))[3] = 0; //alpha 0
frac += fracstep;
}
}
if (qrenderer == QR_OPENGL)
{
texnums->fullbright = R_AllocNewTexture(cm->name, scaled_width, scaled_height);
R_Upload(texnums->fullbright, cm->name, TF_RGBA32, pixels, NULL, scaled_width, scaled_height, IF_NOMIPMAP);
}
else
{
texnums->fullbright = R_LoadTexture(cm->name, scaled_width, scaled_height, h2playertranslations?TF_RGBA32:TF_RGBX32, pixels, 0);
}
}
}
else
{
skins = (galiasskin_t*)((char *)inf + inf->ofsskins);
if (e->skinnum >= 0 && e->skinnum < inf->numskins)
skins += e->skinnum;
if (!inf->numskins || !skins->texnums)
return NULL;
frame = cl.time*skins->skinspeed;
frame = frame%skins->texnums;
texnums = (texnums_t*)((char *)skins + skins->ofstexnums + frame*sizeof(texnums_t));
memcpy(&cm->texnum, texnums, sizeof(cm->texnum));
}
return &cm->texnum;
}
}
if (!inf->numskins)
return NULL;
skins = (galiasskin_t*)((char *)inf + inf->ofsskins);
if (e->skinnum >= 0 && e->skinnum < inf->numskins)
skins += e->skinnum;
else
{
Con_DPrintf("Skin number out of range\n");
if (!inf->numskins)
return NULL;
}
if (!skins->texnums)
return NULL;
frame = cl.time*skins->skinspeed;
frame = frame%skins->texnums;
texnums = (texnums_t*)((char *)skins + skins->ofstexnums + frame*sizeof(texnums_t));
return texnums;
}
#if defined(RTLIGHTS) && defined(GLQUAKE)
static int numFacing;
static qbyte *triangleFacing;
static void R_CalcFacing(mesh_t *mesh, vec3_t lightpos)
{
float *v1, *v2, *v3;
vec3_t d1, d2, norm;
int i;
index_t *indexes = mesh->indexes;
int numtris = mesh->numindexes/3;
if (numFacing < numtris)
{
if (triangleFacing)
BZ_Free(triangleFacing);
triangleFacing = BZ_Malloc(sizeof(*triangleFacing)*numtris);
numFacing = numtris;
}
for (i = 0; i < numtris; i++, indexes+=3)
{
v1 = (float *)(mesh->xyz_array + indexes[0]);
v2 = (float *)(mesh->xyz_array + indexes[1]);
v3 = (float *)(mesh->xyz_array + indexes[2]);
VectorSubtract(v1, v2, d1);
VectorSubtract(v3, v2, d2);
CrossProduct(d1, d2, norm);
triangleFacing[i] = (( lightpos[0] - v1[0] ) * norm[0] + ( lightpos[1] - v1[1] ) * norm[1] + ( lightpos[2] - v1[2] ) * norm[2]) > 0;
}
}
#define PROJECTION_DISTANCE 30000
static int numProjectedShadowVerts;
static vec3_t *ProjectedShadowVerts;
static void R_ProjectShadowVolume(mesh_t *mesh, vec3_t lightpos)
{
int numverts = mesh->numvertexes;
int i;
vecV_t *input = mesh->xyz_array;
vec3_t *projected;
if (numProjectedShadowVerts < numverts)
{
if (ProjectedShadowVerts)
BZ_Free(ProjectedShadowVerts);
ProjectedShadowVerts = BZ_Malloc(sizeof(*ProjectedShadowVerts)*numverts);
numProjectedShadowVerts = numverts;
}
projected = ProjectedShadowVerts;
for (i = 0; i < numverts; i++)
{
projected[i][0] = input[i][0] + (input[i][0]-lightpos[0])*PROJECTION_DISTANCE;
projected[i][1] = input[i][1] + (input[i][1]-lightpos[1])*PROJECTION_DISTANCE;
projected[i][2] = input[i][2] + (input[i][2]-lightpos[2])*PROJECTION_DISTANCE;
}
}
static void R_DrawShadowVolume(mesh_t *mesh)
{
int t;
vec3_t *proj = ProjectedShadowVerts;
vecV_t *verts = mesh->xyz_array;
index_t *indexes = mesh->indexes;
int *neighbours = mesh->trneighbors;
int numtris = mesh->numindexes/3;
qglBegin(GL_TRIANGLES);
for (t = 0; t < numtris; t++)
{
if (triangleFacing[t])
{
//draw front
qglVertex3fv(verts[indexes[t*3+0]]);
qglVertex3fv(verts[indexes[t*3+1]]);
qglVertex3fv(verts[indexes[t*3+2]]);
//draw back
qglVertex3fv(proj[indexes[t*3+1]]);
qglVertex3fv(proj[indexes[t*3+0]]);
qglVertex3fv(proj[indexes[t*3+2]]);
//draw side caps
if (neighbours[t*3+0] < 0 || !triangleFacing[neighbours[t*3+0]])
{
qglVertex3fv(verts[indexes[t*3+1]]);
qglVertex3fv(verts[indexes[t*3+0]]);
qglVertex3fv(proj [indexes[t*3+0]]);
qglVertex3fv(verts[indexes[t*3+1]]);
qglVertex3fv(proj [indexes[t*3+0]]);
qglVertex3fv(proj [indexes[t*3+1]]);
}
if (neighbours[t*3+1] < 0 || !triangleFacing[neighbours[t*3+1]])
{
qglVertex3fv(verts[indexes[t*3+2]]);
qglVertex3fv(verts[indexes[t*3+1]]);
qglVertex3fv(proj [indexes[t*3+1]]);
qglVertex3fv(verts[indexes[t*3+2]]);
qglVertex3fv(proj [indexes[t*3+1]]);
qglVertex3fv(proj [indexes[t*3+2]]);
}
if (neighbours[t*3+2] < 0 || !triangleFacing[neighbours[t*3+2]])
{
qglVertex3fv(verts[indexes[t*3+0]]);
qglVertex3fv(verts[indexes[t*3+2]]);
qglVertex3fv(proj [indexes[t*3+2]]);
qglVertex3fv(verts[indexes[t*3+0]]);
qglVertex3fv(proj [indexes[t*3+2]]);
qglVertex3fv(proj [indexes[t*3+0]]);
}
}
}
qglEnd();
}
#endif
//true if no shading is to be used.
static qboolean R_CalcModelLighting(entity_t *e, model_t *clmodel)
{
vec3_t lightdir;
int i;
vec3_t dist;
float add;
vec3_t shadelight, ambientlight;
if (e->light_known)
return e->light_known-1;
e->light_dir[0] = 0; e->light_dir[1] = 1; e->light_dir[2] = 0;
if (clmodel->engineflags & MDLF_FLAME || r_fullbright.ival)
{
e->light_avg[0] = e->light_avg[1] = e->light_avg[2] = 1;
e->light_range[0] = e->light_range[1] = e->light_range[2] = 0;
e->light_known = 2;
return e->light_known-1;
}
if ((e->drawflags & MLS_MASKIN) == MLS_FULLBRIGHT || (e->flags & Q2RF_FULLBRIGHT))
{
e->light_avg[0] = e->light_avg[1] = e->light_avg[2] = 1;
e->light_range[0] = e->light_range[1] = e->light_range[2] = 0;
e->light_known = 2;
return e->light_known-1;
}
if (!(r_refdef.flags & Q2RDF_NOWORLDMODEL))
{
if (e->flags & Q2RF_WEAPONMODEL)
{
cl.worldmodel->funcs.LightPointValues(cl.worldmodel, r_refdef.vieworg, shadelight, ambientlight, lightdir);
for (i = 0; i < 3; i++)
{ /*viewmodels may not be pure black*/
if (ambientlight[i] < 24)
ambientlight[i] = 24;
}
}
else
{
vec3_t center;
#if 0 /*hexen2*/
VectorAvg(clmodel->mins, clmodel->maxs, center);
VectorAdd(e->origin, center, center);
#else
VectorCopy(e->origin, center);
center[2] += 8;
#endif
cl.worldmodel->funcs.LightPointValues(cl.worldmodel, center, shadelight, ambientlight, lightdir);
}
}
else
{
ambientlight[0] = ambientlight[1] = ambientlight[2] = shadelight[0] = shadelight[1] = shadelight[2] = 128;
lightdir[0] = 0;
lightdir[1] = 1;
lightdir[2] = 1;
}
if (!r_vertexdlights.ival && r_dynamic.ival)
{
//don't do world lights, although that might be funny
for (i=rtlights_first; i<RTL_FIRST; i++)
{
if (cl_dlights[i].radius)
{
VectorSubtract (e->origin,
cl_dlights[i].origin,
dist);
add = cl_dlights[i].radius - Length(dist);
if (add > 0)
{
add*=5;
ambientlight[0] += add * cl_dlights[i].color[0];
ambientlight[1] += add * cl_dlights[i].color[1];
ambientlight[2] += add * cl_dlights[i].color[2];
//ZOID models should be affected by dlights as well
shadelight[0] += add * cl_dlights[i].color[0];
shadelight[1] += add * cl_dlights[i].color[1];
shadelight[2] += add * cl_dlights[i].color[2];
}
}
}
}
for (i = 0; i < 3; i++) //clamp light so it doesn't get vulgar.
{
if (ambientlight[i] > 128)
ambientlight[i] = 128;
if (shadelight[i] > 192)
shadelight[i] = 192;
}
//MORE HUGE HACKS! WHEN WILL THEY CEASE!
// clamp lighting so it doesn't overbright as much
// ZOID: never allow players to go totally black
if (clmodel->engineflags & MDLF_PLAYER)
{
float fb = r_fullbrightSkins.value;
if (fb > cls.allow_fbskins)
fb = cls.allow_fbskins;
if (fb < 0)
fb = 0;
if (fb)
{
extern cvar_t r_fb_models;
if (fb >= 1 && r_fb_models.value)
{
ambientlight[0] = ambientlight[1] = ambientlight[2] = 1;
shadelight[0] = shadelight[1] = shadelight[2] = 1;
e->light_known = 2;
return e->light_known-1;
}
else
{
for (i = 0; i < 3; i++)
{
ambientlight[i] = max(ambientlight[i], 8 + fb * 120);
shadelight[i] = max(shadelight[i], 8 + fb * 120);
}
}
}
for (i = 0; i < 3; i++)
{
if (ambientlight[i] < 8)
ambientlight[i] = 8;
}
}
for (i = 0; i < 3; i++)
{
if (ambientlight[i] > 128)
ambientlight[i] = 128;
shadelight[i] /= 200.0/255;
ambientlight[i] /= 200.0/255;
}
if ((e->model->flags & MF_ROTATE) && cl.hexen2pickups)
{
shadelight[0] = shadelight[1] = shadelight[2] =
ambientlight[0] = ambientlight[1] = ambientlight[2] = 128+sin(cl.servertime*4)*64;
}
if ((e->drawflags & MLS_MASKIN) == MLS_ABSLIGHT)
{
shadelight[0] = shadelight[1] = shadelight[2] = e->abslight;
ambientlight[0] = ambientlight[1] = ambientlight[2] = e->abslight;
}
//#define SHOWLIGHTDIR
{ //lightdir is absolute, shadevector is relative
e->light_dir[0] = DotProduct(lightdir, e->axis[0]);
e->light_dir[1] = DotProduct(lightdir, e->axis[1]);
e->light_dir[2] = DotProduct(lightdir, e->axis[2]);
if (e->flags & Q2RF_WEAPONMODEL)
{
vec3_t temp;
temp[0] = DotProduct(e->light_dir, vpn);
temp[1] = -DotProduct(e->light_dir, vright);
temp[2] = DotProduct(e->light_dir, vup);
VectorCopy(temp, e->light_dir);
}
VectorNormalize(e->light_dir);
}
shadelight[0] *= 1/255.0f;
shadelight[1] *= 1/255.0f;
shadelight[2] *= 1/255.0f;
ambientlight[0] *= 1/255.0f;
ambientlight[1] *= 1/255.0f;
ambientlight[2] *= 1/255.0f;
if (e->flags & Q2RF_GLOW)
{
shadelight[0] += sin(cl.time)*0.25;
shadelight[1] += sin(cl.time)*0.25;
shadelight[2] += sin(cl.time)*0.25;
}
VectorMA(ambientlight, 0.5, shadelight, e->light_avg);
VectorSubtract(shadelight, ambientlight, e->light_range);
e->light_known = 1;
return e->light_known-1;
}
void R_GAlias_DrawBatch(batch_t *batch)
{
entity_t *e;
galiasinfo_t *inf;
model_t *clmodel;
int surfnum;
static mesh_t mesh;
static mesh_t *meshl = &mesh;
qboolean needrecolour;
qboolean nolightdir;
e = batch->ent;
clmodel = e->model;
currententity = e;
nolightdir = R_CalcModelLighting(e, clmodel);
inf = RMod_Extradata (clmodel);
if (inf)
{
memset(&mesh, 0, sizeof(mesh));
for(surfnum=0; inf; ((inf->nextsurf)?(inf = (galiasinfo_t*)((char *)inf + inf->nextsurf)):(inf=NULL)), surfnum++)
{
if (batch->surf_first == surfnum)
{
needrecolour = Alias_GAliasBuildMesh(&mesh, inf, surfnum, e, true);
batch->mesh = &meshl;
return;
}
}
}
batch->meshes = 0;
Con_Printf("Broken model surfaces mid-frame\n");
return;
}
void R_GAlias_GenerateBatches(entity_t *e, batch_t **batches)
{
galiasinfo_t *inf;
model_t *clmodel;
shader_t *shader;
batch_t *b;
int surfnum;
shadersort_t sort;
texnums_t *skin;
if (r_refdef.externalview && e->flags & Q2RF_WEAPONMODEL)
return;
/*switch model if we're the player model, and the player skin says a new model*/
{
extern int cl_playerindex;
if (e->scoreboard && e->model == cl.model_precache[cl_playerindex])
{
clmodel = e->scoreboard->model;
if (clmodel && clmodel->type == mod_alias)
e->model = clmodel;
}
}
clmodel = e->model;
if (!(e->flags & Q2RF_WEAPONMODEL))
{
if (R_CullEntityBox (e, clmodel->mins, clmodel->maxs))
return;
#ifdef RTLIGHTS
if (BE_LightCullModel(e->origin, clmodel))
return;
}
else
{
if (BE_LightCullModel(r_origin, clmodel))
return;
#endif
}
if (clmodel->tainted)
{
if (!ruleset_allow_modified_eyes.ival && !strcmp(clmodel->name, "progs/eyes.mdl"))
return;
}
inf = RMod_Extradata (clmodel);
for(surfnum=0; inf; ((inf->nextsurf)?(inf = (galiasinfo_t*)((char *)inf + inf->nextsurf)):(inf=NULL)), surfnum++)
{
skin = GL_ChooseSkin(inf, clmodel, surfnum, e);
if (!skin)
continue;
shader = e->forcedshader?e->forcedshader:skin->shader;
if (shader)
{
b = BE_GetTempBatch();
if (!b)
break;
b->buildmeshes = R_GAlias_DrawBatch;
b->ent = e;
b->mesh = NULL;
b->firstmesh = 0;
b->meshes = 1;
b->skin = skin;
b->texture = NULL;
b->shader = shader;
b->lightmap = -1;
b->surf_first = surfnum;
b->flags = 0;
sort = shader->sort;
if (e->flags & Q2RF_ADDITIVE)
{
b->flags |= BEF_FORCEADDITIVE;
if (sort < SHADER_SORT_ADDITIVE)
sort = SHADER_SORT_ADDITIVE;
}
if (e->flags & Q2RF_TRANSLUCENT)
{
b->flags |= BEF_FORCETRANSPARENT;
if (SHADER_SORT_PORTAL < sort && sort < SHADER_SORT_BLEND)
sort = SHADER_SORT_BLEND;
}
if (e->flags & RF_NODEPTHTEST)
{
b->flags |= BEF_FORCENODEPTH;
if (sort < SHADER_SORT_NEAREST)
sort = SHADER_SORT_NEAREST;
}
if (e->flags & RF_NOSHADOW)
b->flags |= BEF_NOSHADOWS;
b->vbo = 0;
b->next = batches[sort];
batches[sort] = b;
}
}
}
#if 0
void R_Sprite_GenerateBatches(entity_t *e, batch_t **batches)
{
galiasinfo_t *inf;
model_t *clmodel;
shader_t *shader;
batch_t *b;
int surfnum;
texnums_t *skin;
if (r_refdef.externalview && e->flags & Q2RF_WEAPONMODEL)
return;
clmodel = e->model;
if (!(e->flags & Q2RF_WEAPONMODEL))
{
if (R_CullEntityBox (e, clmodel->mins, clmodel->maxs))
return;
#ifdef RTLIGHTS
if (BE_LightCullModel(e->origin, clmodel))
return;
}
else
{
if (BE_LightCullModel(r_origin, clmodel))
return;
#endif
}
if (clmodel->tainted)
{
if (!ruleset_allow_modified_eyes.ival && !strcmp(clmodel->name, "progs/eyes.mdl"))
return;
}
inf = RMod_Extradata (clmodel);
if (!e->model || e->forcedshader)
{
//fixme
return;
}
else
{
frame = R_GetSpriteFrame (e);
psprite = e->model->cache.data;
sprtype = psprite->type;
shader = frame->shader;
}
if (shader)
{
b = BE_GetTempBatch();
if (!b)
break;
b->buildmeshes = R_Sprite_DrawBatch;
b->ent = e;
b->mesh = NULL;
b->firstmesh = 0;
b->meshes = 1;
b->skin = frame-;
b->texture = NULL;
b->shader = frame->shader;
b->lightmap = -1;
b->surf_first = surfnum;
b->flags = 0;
b->vbo = 0;
b->next = batches[shader->sort];
batches[shader->sort] = b;
}
}
#endif
//returns the rotated offset of the two points in result
void RotateLightVector(const vec3_t *axis, const vec3_t origin, const vec3_t lightpoint, vec3_t result)
{
vec3_t offs;
offs[0] = lightpoint[0] - origin[0];
offs[1] = lightpoint[1] - origin[1];
offs[2] = lightpoint[2] - origin[2];
result[0] = DotProduct (offs, axis[0]);
result[1] = DotProduct (offs, axis[1]);
result[2] = DotProduct (offs, axis[2]);
}
#if defined(RTLIGHTS) && defined(GLQUAKE)
void GL_LightMesh (mesh_t *mesh, vec3_t lightpos, vec3_t colours, float radius)
{
vec3_t dir;
int i;
float dot, d, f, a;
vecV_t *xyz = mesh->xyz_array;
vec3_t *normals = mesh->normals_array;
vec4_t *out = mesh->colors4f_array;
if (!out)
return; //urm..
if (normals)
{
for (i = 0; i < mesh->numvertexes; i++)
{
VectorSubtract(lightpos, xyz[i], dir);
dot = DotProduct(dir, normals[i]);
if (dot > 0)
{
d = DotProduct(dir, dir)/radius;
a = 1/d;
if (a>0)
{
a *= dot/sqrt(d);
f = a*colours[0];
out[i][0] = f;
f = a*colours[1];
out[i][1] = f;
f = a*colours[2];
out[i][2] = f;
}
else
{
out[i][0] = 0;
out[i][1] = 0;
out[i][2] = 0;
}
}
else
{
out[i][0] = 0;
out[i][1] = 0;
out[i][2] = 0;
}
out[i][3] = 1;
}
}
else
{
for (i = 0; i < mesh->numvertexes; i++)
{
VectorSubtract(lightpos, xyz[i], dir);
out[i][0] = colours[0];
out[i][1] = colours[1];
out[i][2] = colours[2];
out[i][3] = 1;
}
}
}
//courtesy of DP
void R_BuildBumpVectors(const float *v0, const float *v1, const float *v2, const float *tc0, const float *tc1, const float *tc2, float *svector3f, float *tvector3f, float *normal3f)
{
float f, tangentcross[3], v10[3], v20[3], tc10[2], tc20[2];
// 79 add/sub/negate/multiply (1 cycle), 1 compare (3 cycle?), total cycles not counting load/store/exchange roughly 82 cycles
// 6 add, 28 subtract, 39 multiply, 1 compare, 50% chance of 6 negates
// 6 multiply, 9 subtract
VectorSubtract(v1, v0, v10);
VectorSubtract(v2, v0, v20);
normal3f[0] = v10[1] * v20[2] - v10[2] * v20[1];
normal3f[1] = v10[2] * v20[0] - v10[0] * v20[2];
normal3f[2] = v10[0] * v20[1] - v10[1] * v20[0];
// 12 multiply, 10 subtract
tc10[1] = tc1[1] - tc0[1];
tc20[1] = tc2[1] - tc0[1];
svector3f[0] = tc10[1] * v20[0] - tc20[1] * v10[0];
svector3f[1] = tc10[1] * v20[1] - tc20[1] * v10[1];
svector3f[2] = tc10[1] * v20[2] - tc20[1] * v10[2];
tc10[0] = tc1[0] - tc0[0];
tc20[0] = tc2[0] - tc0[0];
tvector3f[0] = tc10[0] * v20[0] - tc20[0] * v10[0];
tvector3f[1] = tc10[0] * v20[1] - tc20[0] * v10[1];
tvector3f[2] = tc10[0] * v20[2] - tc20[0] * v10[2];
// 12 multiply, 4 add, 6 subtract
f = DotProduct(svector3f, normal3f);
svector3f[0] -= f * normal3f[0];
svector3f[1] -= f * normal3f[1];
svector3f[2] -= f * normal3f[2];
f = DotProduct(tvector3f, normal3f);
tvector3f[0] -= f * normal3f[0];
tvector3f[1] -= f * normal3f[1];
tvector3f[2] -= f * normal3f[2];
// if texture is mapped the wrong way (counterclockwise), the tangents
// have to be flipped, this is detected by calculating a normal from the
// two tangents, and seeing if it is opposite the surface normal
// 9 multiply, 2 add, 3 subtract, 1 compare, 50% chance of: 6 negates
CrossProduct(tvector3f, svector3f, tangentcross);
if (DotProduct(tangentcross, normal3f) < 0)
{
VectorNegate(svector3f, svector3f);
VectorNegate(tvector3f, tvector3f);
}
}
//courtesy of DP
void R_AliasGenerateTextureVectors(mesh_t *mesh, float *normal3f, float *svector3f, float *tvector3f)
{
int i;
float sdir[3], tdir[3], normal[3], *v;
index_t *e;
float *vertex3f = (float*)mesh->xyz_array;
float *texcoord2f = (float*)mesh->st_array;
// clear the vectors
// if (svector3f)
memset(svector3f, 0, mesh->numvertexes * sizeof(float[3]));
// if (tvector3f)
memset(tvector3f, 0, mesh->numvertexes * sizeof(float[3]));
// if (normal3f)
memset(normal3f, 0, mesh->numvertexes * sizeof(float[3]));
// process each vertex of each triangle and accumulate the results
for (e = mesh->indexes; e < mesh->indexes+mesh->numindexes; e += 3)
{
R_BuildBumpVectors(vertex3f + e[0] * 3, vertex3f + e[1] * 3, vertex3f + e[2] * 3, texcoord2f + e[0] * 2, texcoord2f + e[1] * 2, texcoord2f + e[2] * 2, sdir, tdir, normal);
// if (!areaweighting)
// {
// VectorNormalize(sdir);
// VectorNormalize(tdir);
// VectorNormalize(normal);
// }
// if (svector3f)
for (i = 0;i < 3;i++)
VectorAdd(svector3f + e[i]*3, sdir, svector3f + e[i]*3);
// if (tvector3f)
for (i = 0;i < 3;i++)
VectorAdd(tvector3f + e[i]*3, tdir, tvector3f + e[i]*3);
// if (normal3f)
for (i = 0;i < 3;i++)
VectorAdd(normal3f + e[i]*3, normal, normal3f + e[i]*3);
}
// now we could divide the vectors by the number of averaged values on
// each vertex... but instead normalize them
// 4 assignments, 1 divide, 1 sqrt, 2 adds, 6 multiplies
if (svector3f)
for (i = 0, v = svector3f;i < mesh->numvertexes;i++, v += 3)
VectorNormalize(v);
// 4 assignments, 1 divide, 1 sqrt, 2 adds, 6 multiplies
if (tvector3f)
for (i = 0, v = tvector3f;i < mesh->numvertexes;i++, v += 3)
VectorNormalize(v);
// 4 assignments, 1 divide, 1 sqrt, 2 adds, 6 multiplies
if (normal3f)
for (i = 0, v = normal3f;i < mesh->numvertexes;i++, v += 3)
VectorNormalize(v);
}
void R_AliasGenerateVertexLightDirs(mesh_t *mesh, vec3_t lightdir, vec3_t *results, vec3_t *normal3f, vec3_t *svector3f, vec3_t *tvector3f)
{
int i;
R_AliasGenerateTextureVectors(mesh, (float*)normal3f, (float*)svector3f, (float*)tvector3f);
for (i = 0; i < mesh->numvertexes; i++)
{
results[i][0] = -DotProduct(lightdir, tvector3f[i]);
results[i][1] = -DotProduct(lightdir, svector3f[i]);
results[i][2] = -DotProduct(lightdir, normal3f[i]);
}
}
//FIXME: Be less agressive.
//This function will have to be called twice (for geforce cards), with the same data, so do the building once and rendering twice.
void R_DrawGAliasShadowVolume(entity_t *e, vec3_t lightpos, float radius)
{
model_t *clmodel = e->model;
galiasinfo_t *inf;
mesh_t mesh;
vec3_t lightorg;
int surfnum = 0;
if (clmodel->engineflags & (MDLF_FLAME | MDLF_BOLT))
return;
if (r_noaliasshadows.ival)
return;
// if (e->shaderRGBAf[3] < 0.5)
// return;
RotateLightVector((void *)e->axis, e->origin, lightpos, lightorg);
if (Length(lightorg) > radius + clmodel->radius)
return;
BE_SelectEntity(e);
inf = RMod_Extradata (clmodel);
while(inf)
{
if (inf->ofs_trineighbours)
{
Alias_GAliasBuildMesh(&mesh, inf, surfnum, e, false);
R_CalcFacing(&mesh, lightorg);
R_ProjectShadowVolume(&mesh, lightorg);
R_DrawShadowVolume(&mesh);
}
if (inf->nextsurf)
inf = (galiasinfo_t*)((char *)inf + inf->nextsurf);
else
inf = NULL;
surfnum++;
}
}
#endif
#if 0
static int R_FindTriangleWithEdge ( index_t *indexes, int numtris, index_t start, index_t 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;
}
#endif
#if 0
static void R_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);
}
}
#endif
#if 0
void GL_GenerateNormals(float *orgs, float *normals, int *indicies, int numtris, int numverts)
{
vec3_t d1, d2;
vec3_t norm;
int t, i, v1, v2, v3;
int tricounts[MD2MAX_VERTS];
vec3_t combined[MD2MAX_VERTS];
int triremap[MD2MAX_VERTS];
if (numverts > MD2MAX_VERTS)
return; //not an issue, you just loose the normals.
memset(triremap, 0, numverts*sizeof(triremap[0]));
v2=0;
for (i = 0; i < numverts; i++) //weld points
{
for (v1 = 0; v1 < v2; v1++)
{
if (orgs[i*3+0] == combined[v1][0] &&
orgs[i*3+1] == combined[v1][1] &&
orgs[i*3+2] == combined[v1][2])
{
triremap[i] = v1;
break;
}
}
if (v1 == v2)
{
combined[v1][0] = orgs[i*3+0];
combined[v1][1] = orgs[i*3+1];
combined[v1][2] = orgs[i*3+2];
v2++;
triremap[i] = v1;
}
}
memset(tricounts, 0, v2*sizeof(tricounts[0]));
memset(combined, 0, v2*sizeof(*combined));
for (t = 0; t < numtris; t++)
{
v1 = triremap[indicies[t*3]];
v2 = triremap[indicies[t*3+1]];
v3 = triremap[indicies[t*3+2]];
VectorSubtract((orgs+v2*3), (orgs+v1*3), d1);
VectorSubtract((orgs+v3*3), (orgs+v1*3), d2);
CrossProduct(d1, d2, norm);
VectorNormalize(norm);
VectorAdd(norm, combined[v1], combined[v1]);
VectorAdd(norm, combined[v2], combined[v2]);
VectorAdd(norm, combined[v3], combined[v3]);
tricounts[v1]++;
tricounts[v2]++;
tricounts[v3]++;
}
for (i = 0; i < numverts; i++)
{
if (tricounts[triremap[i]])
{
VectorScale(combined[triremap[i]], 1.0f/tricounts[triremap[i]], normals+i*3);
}
}
}
#endif
#endif
qboolean BE_ShouldDraw(entity_t *e)
{
if (!r_refdef.externalview && (e->externalmodelview & (1<<r_refdef.currentplayernum)))
return false;
if (!Cam_DrawPlayer(r_refdef.currentplayernum, e->keynum-1))
return false;
return true;
}
#ifdef Q3CLIENT
//q3 lightning gun
static void R_DB_LightningBeam(batch_t *batch)
{
entity_t *e = batch->ent;
vec3_t v;
vec3_t dir, cr;
float scale = e->scale;
float length;
static vecV_t points[4];
static vec2_t texcoords[4] = {{0, 0}, {0, 1}, {1, 1}, {1, 0}};
static index_t indexarray[6] = {0, 1, 2, 0, 2, 3};
static vec4_t colors[4];
static mesh_t mesh;
static mesh_t *meshptr = &mesh;
if (batch->ent == &r_worldentity)
{
mesh.numindexes = 0;
mesh.numindexes = 0;
return;
}
scale *= -10;
if (!scale)
scale = 10;
VectorSubtract(e->origin, e->oldorigin, dir);
length = Length(dir);
//this seems to be about right.
texcoords[2][0] = length/128;
texcoords[3][0] = length/128;
VectorSubtract(r_refdef.vieworg, e->origin, v);
CrossProduct(v, dir, cr);
VectorNormalize(cr);
VectorMA(e->origin, -scale/2, cr, points[0]);
VectorMA(e->origin, scale/2, cr, points[1]);
VectorSubtract(r_refdef.vieworg, e->oldorigin, v);
CrossProduct(v, dir, cr);
VectorNormalize(cr);
VectorMA(e->oldorigin, scale/2, cr, points[2]);
VectorMA(e->oldorigin, -scale/2, cr, points[3]);
/*this is actually meant to be 4 separate quads at 45 degrees from each other*/
Vector4Copy(e->shaderRGBAf, colors[0]);
Vector4Copy(e->shaderRGBAf, colors[1]);
Vector4Copy(e->shaderRGBAf, colors[2]);
Vector4Copy(e->shaderRGBAf, colors[3]);
batch->ent = &r_worldentity;
batch->mesh = &meshptr;
memset(&mesh, 0, sizeof(mesh));
mesh.vbofirstelement = 0;
mesh.vbofirstvert = 0;
mesh.xyz_array = points;
mesh.indexes = indexarray;
mesh.numindexes = sizeof(indexarray)/sizeof(indexarray[0]);
mesh.colors4f_array = (vec4_t*)colors;
mesh.lmst_array = NULL;
mesh.normals_array = NULL;
mesh.numvertexes = 4;
mesh.st_array = texcoords;
}
//q3 railgun beam
static void R_DB_RailgunBeam(batch_t *batch)
{
entity_t *e = batch->ent;
vec3_t v;
vec3_t dir, cr;
float scale = e->scale;
float length;
static mesh_t mesh;
static mesh_t *meshptr = &mesh;
static vecV_t points[4];
static vec2_t texcoords[4] = {{0, 0}, {0, 1}, {1, 1}, {1, 0}};
static index_t indexarray[6] = {0, 1, 2, 0, 2, 3};
static vec4_t colors[4];
if (batch->ent == &r_worldentity)
{
mesh.numindexes = 0;
mesh.numindexes = 0;
return;
}
if (!e->forcedshader)
return;
if (!scale)
scale = 10;
VectorSubtract(e->origin, e->oldorigin, dir);
length = Length(dir);
//this seems to be about right.
texcoords[2][0] = length/128;
texcoords[3][0] = length/128;
VectorSubtract(r_refdef.vieworg, e->origin, v);
CrossProduct(v, dir, cr);
VectorNormalize(cr);
VectorMA(e->origin, -scale/2, cr, points[0]);
VectorMA(e->origin, scale/2, cr, points[1]);
VectorSubtract(r_refdef.vieworg, e->oldorigin, v);
CrossProduct(v, dir, cr);
VectorNormalize(cr);
VectorMA(e->oldorigin, scale/2, cr, points[2]);
VectorMA(e->oldorigin, -scale/2, cr, points[3]);
Vector4Copy(e->shaderRGBAf, colors[0]);
Vector4Copy(e->shaderRGBAf, colors[1]);
Vector4Copy(e->shaderRGBAf, colors[2]);
Vector4Copy(e->shaderRGBAf, colors[3]);
batch->ent = &r_worldentity;
batch->mesh = &meshptr;
memset(&mesh, 0, sizeof(mesh));
mesh.vbofirstelement = 0;
mesh.vbofirstvert = 0;
mesh.xyz_array = points;
mesh.indexes = indexarray;
mesh.numindexes = sizeof(indexarray)/sizeof(indexarray[0]);
mesh.colors4f_array = (vec4_t*)colors;
mesh.lmst_array = NULL;
mesh.normals_array = NULL;
mesh.numvertexes = 4;
mesh.st_array = texcoords;
}
#endif
static void R_DB_Sprite(batch_t *batch)
{
entity_t *e = batch->ent;
vec3_t point;
mspriteframe_t *frame, genframe;
vec3_t forward, right, up;
msprite_t *psprite;
vec3_t sprorigin;
unsigned int sprtype;
static vec2_t texcoords[4]={{0, 1},{0,0},{1,0},{1,1}};
static index_t indexes[6] = {0, 1, 2, 0, 2, 3};
static vecV_t vertcoords[4];
static avec4_t colours[4];
static mesh_t mesh;
static mesh_t *meshptr = &mesh;
if (batch->ent == &r_worldentity)
{
mesh.numindexes = 0;
mesh.numindexes = 0;
return;
}
if (e->flags & Q2RF_WEAPONMODEL && r_refdef.currentplayernum >= 0)
{
sprorigin[0] = cl.viewent[r_refdef.currentplayernum].origin[0];
sprorigin[1] = cl.viewent[r_refdef.currentplayernum].origin[1];
sprorigin[2] = cl.viewent[r_refdef.currentplayernum].origin[2];
VectorMA(sprorigin, e->origin[0], cl.viewent[r_refdef.currentplayernum].axis[0], sprorigin);
VectorMA(sprorigin, e->origin[1], cl.viewent[r_refdef.currentplayernum].axis[1], sprorigin);
VectorMA(sprorigin, e->origin[2], cl.viewent[r_refdef.currentplayernum].axis[2], sprorigin);
VectorMA(sprorigin, 12, vpn, sprorigin);
batch->flags |= BEF_FORCENODEPTH;
}
else
VectorCopy(e->origin, sprorigin);
if (!e->model || e->forcedshader)
{
genframe.shader = e->forcedshader;
genframe.up = genframe.left = -1;
genframe.down = genframe.right = 1;
sprtype = SPR_VP_PARALLEL;
frame = &genframe;
}
else
{
// don't even bother culling, because it's just a single
// polygon without a surface cache
frame = R_GetSpriteFrame (e);
psprite = e->model->cache.data;
sprtype = psprite->type;
}
if (!frame->shader)
return;
switch(sprtype)
{
case SPR_ORIENTED:
// bullet marks on walls
AngleVectors (e->angles, forward, right, up);
break;
case SPR_FACING_UPRIGHT:
up[0] = 0;up[1] = 0;up[2]=1;
right[0] = sprorigin[1] - r_origin[1];
right[1] = -(sprorigin[0] - r_origin[0]);
right[2] = 0;
VectorNormalize (right);
break;
case SPR_VP_PARALLEL_UPRIGHT:
up[0] = 0;up[1] = 0;up[2]=1;
VectorCopy (vright, right);
break;
default:
case SPR_VP_PARALLEL:
//normal sprite
VectorCopy(vup, up);
VectorCopy(vright, right);
break;
}
up[0]*=e->scale;
up[1]*=e->scale;
up[2]*=e->scale;
right[0]*=e->scale;
right[1]*=e->scale;
right[2]*=e->scale;
if (e->shaderRGBAf[0] > 1)
e->shaderRGBAf[0] = 1;
if (e->shaderRGBAf[1] > 1)
e->shaderRGBAf[1] = 1;
if (e->shaderRGBAf[2] > 1)
e->shaderRGBAf[2] = 1;
Vector4Copy(e->shaderRGBAf, colours[0]);
Vector4Copy(e->shaderRGBAf, colours[1]);
Vector4Copy(e->shaderRGBAf, colours[2]);
Vector4Copy(e->shaderRGBAf, colours[3]);
VectorMA (sprorigin, frame->down, up, point);
VectorMA (point, frame->left, right, vertcoords[0]);
VectorMA (sprorigin, frame->up, up, point);
VectorMA (point, frame->left, right, vertcoords[1]);
VectorMA (sprorigin, frame->up, up, point);
VectorMA (point, frame->right, right, vertcoords[2]);
VectorMA (sprorigin, frame->down, up, point);
VectorMA (point, frame->right, right, vertcoords[3]);
batch->ent = &r_worldentity;
batch->mesh = &meshptr;
memset(&mesh, 0, sizeof(mesh));
mesh.vbofirstelement = 0;
mesh.vbofirstvert = 0;
mesh.xyz_array = vertcoords;
mesh.indexes = indexes;
mesh.numindexes = sizeof(indexes)/sizeof(indexes[0]);
mesh.colors4f_array = colours;
mesh.lmst_array = NULL;
mesh.normals_array = NULL;
mesh.numvertexes = 4;
mesh.st_array = texcoords;
mesh.istrifan = true;
}
static void R_Sprite_GenerateBatch(entity_t *e, batch_t **batches, void (*drawfunc)(batch_t *batch))
{
extern cvar_t gl_blendsprites;
shader_t *shader = NULL;
batch_t *b;
shadersort_t sort;
if (!e->model || e->model->type != mod_sprite || e->forcedshader)
{
shader = e->forcedshader;
if (!shader)
shader = R_RegisterShader("q2beam",
"{\n"
"{\n"
"map $whiteimage\n"
"rgbgen vertex\n"
"alphagen vertex\n"
"blendfunc blend\n"
"}\n"
"}\n"
);
}
else
{
// don't even bother culling, because it's just a single
// polygon without a surface cache
shader = R_GetSpriteFrame(e)->shader;
}
if (!shader)
return;
b = BE_GetTempBatch();
if (!b)
return;
b->flags = 0;
sort = shader->sort;
if (e->flags & Q2RF_ADDITIVE)
{
b->flags |= BEF_FORCEADDITIVE;
if (sort < SHADER_SORT_ADDITIVE)
sort = SHADER_SORT_ADDITIVE;
}
if (e->flags & Q2RF_TRANSLUCENT || (gl_blendsprites.ival && drawfunc == R_DB_Sprite))
{
b->flags |= BEF_FORCETRANSPARENT;
if (SHADER_SORT_PORTAL < sort && sort < SHADER_SORT_BLEND)
sort = SHADER_SORT_BLEND;
}
if (e->flags & RF_NODEPTHTEST)
{
b->flags |= BEF_FORCENODEPTH;
if (sort < SHADER_SORT_BANNER)
sort = SHADER_SORT_BANNER;
}
b->buildmeshes = drawfunc;
b->ent = e;
b->mesh = NULL;
b->firstmesh = 0;
b->meshes = 1;
b->skin = &shader->defaulttextures;
b->texture = NULL;
b->shader = shader;
b->lightmap = -1;
b->surf_first = 0;
b->flags |= BEF_NODLIGHT|BEF_NOSHADOWS;
b->vbo = 0;
b->next = batches[sort];
batches[sort] = b;
}
static void R_DB_Poly(batch_t *batch)
{
static mesh_t mesh;
static mesh_t *meshptr = &mesh;
unsigned int i = batch->surf_first;
batch->mesh = &meshptr;
mesh.xyz_array = cl_strisvertv + cl_stris[i].firstvert;
mesh.st_array = cl_strisvertt + cl_stris[i].firstvert;
mesh.colors4f_array = cl_strisvertc + cl_stris[i].firstvert;
mesh.indexes = cl_strisidx + cl_stris[i].firstidx;
mesh.numindexes = cl_stris[i].numidx;
mesh.numvertexes = cl_stris[i].numvert;
}
void BE_GenPolyBatches(batch_t **batches)
{
shader_t *shader = NULL;
batch_t *b;
unsigned int i;
for (i = 0; i < cl_numstris; i++)
{
b = BE_GetTempBatch();
if (!b)
return;
shader = cl_stris[i].shader;
b->buildmeshes = R_DB_Poly;
b->ent = &r_worldentity;
b->mesh = NULL;
b->firstmesh = 0;
b->meshes = 1;
b->skin = &shader->defaulttextures;
b->texture = NULL;
b->shader = shader;
b->lightmap = -1;
b->surf_first = i;
b->flags = BEF_NODLIGHT|BEF_NOSHADOWS;
b->vbo = 0;
b->next = batches[shader->sort];
batches[shader->sort] = b;
}
}
void BE_GenModelBatches(batch_t **batches)
{
int i;
entity_t *ent;
/*clear the batch list*/
for (i = 0; i < SHADER_SORT_COUNT; i++)
batches[i] = NULL;
if (!r_drawentities.ival)
return;
Alias_FlushCache();
#if defined(TERRAIN)
if (cl.worldmodel && cl.worldmodel->type == mod_heightmap)
GL_DrawHeightmapModel(batches, &r_worldentity);
#endif
// draw sprites seperately, because of alpha blending
for (i=0 ; i<cl_numvisedicts ; i++)
{
ent = &cl_visedicts[i];
if (!BE_ShouldDraw(ent))
continue;
switch(ent->rtype)
{
case RT_MODEL:
default:
if (!ent->model)
continue;
if (ent->model->needload)
continue;
if (cl.lerpents && (cls.allow_anyparticles || ent->visframe)) //allowed or static
{
if (gl_part_flame.value)
{
if (ent->model->engineflags & MDLF_ENGULPHS)
continue;
}
}
if (ent->model->engineflags & MDLF_NOTREPLACEMENTS)
{
if (ent->model->fromgame != fg_quake || ent->model->type != mod_alias)
if (!ruleset_allow_sensative_texture_replacements.value)
continue;
}
switch(ent->model->type)
{
case mod_brush:
if (r_drawentities.ival == 2)
continue;
Surf_GenBrushBatches(batches, ent);
break;
case mod_alias:
if (r_drawentities.ival == 3)
continue;
R_GAlias_GenerateBatches(ent, batches);
break;
case mod_sprite:
R_Sprite_GenerateBatch(ent, batches, R_DB_Sprite);
break;
// warning: enumeration value <20>mod_*<2A> not handled in switch
case mod_dummy:
case mod_halflife:
case mod_heightmap:
break;
}
break;
case RT_SPRITE:
R_Sprite_GenerateBatch(ent, batches, R_DB_Sprite);
break;
#ifdef Q3CLIENT
case RT_BEAM:
case RT_RAIL_RINGS:
case RT_LIGHTNING:
R_Sprite_GenerateBatch(ent, batches, R_DB_LightningBeam);
continue;
case RT_RAIL_CORE:
R_Sprite_GenerateBatch(ent, batches, R_DB_RailgunBeam);
continue;
#endif
case RT_POLY:
/*not implemented*/
break;
case RT_PORTALSURFACE:
/*nothing*/
break;
}
}
if (cl_numstris)
BE_GenPolyBatches(batches);
}
#endif // defined(GLQUAKE)