1
0
Fork 0
forked from fte/fteqw
fteqw/engine/common/q1bsp.c
Spoike e3fdbfcdbd qcc: Add unsigned int, long/__int64, unsigned long/__uint64, and double datatypes to qcc+qcvm.
qcc: Add '#pragma opcode enable|disable NAME' to override whether an individual opcode may be generated for the current target.
nq: Fix smartjump on nq servers.
nq: Fix enemyskin/teamskin stuff on nq servers.
Rework show_fps cvar to give a greater focus on the timings graph.
Update updates menu to allow enabling/disabling specific sources.
Add separate settings for ef_red+ef_green+ef_blue.



git-svn-id: https://svn.code.sf.net/p/fteqw/code/trunk@5768 fc73d0e0-1445-4013-8a0c-d673dee63da5
2020-09-29 07:09:01 +00:00

2921 lines
72 KiB
C

#include "quakedef.h"
#include "pr_common.h"
#include "shader.h"
extern cvar_t r_decal_noperpendicular;
extern cvar_t mod_loadsurfenvmaps;
/*
Decal functions
*/
#define MAXFRAGMENTVERTS 360
int Fragment_ClipPolyToPlane(float *inverts, float *outverts, int incount, float *plane, float planedist)
{
#define C (sizeof(vecV_t)/sizeof(vec_t))
float dotv[MAXFRAGMENTVERTS+1];
char keep[MAXFRAGMENTVERTS+1];
#define KEEP_KILL 0
#define KEEP_KEEP 1
#define KEEP_BORDER 2
int i;
int outcount = 0;
int clippedcount = 0;
float d, *p1, *p2, *out;
#define FRAG_EPSILON (1.0/32) //0.5
for (i = 0; i < incount; i++)
{
dotv[i] = DotProduct((inverts+i*C), plane) - planedist;
if (dotv[i]<-FRAG_EPSILON)
{
keep[i] = KEEP_KILL;
clippedcount++;
}
else if (dotv[i] > FRAG_EPSILON)
keep[i] = KEEP_KEEP;
else
keep[i] = KEEP_BORDER;
}
dotv[i] = dotv[0];
keep[i] = keep[0];
if (clippedcount == incount)
return 0; //all were clipped
if (clippedcount == 0)
{ //none were clipped
for (i = 0; i < incount; i++)
VectorCopy((inverts+i*C), (outverts+i*C));
return incount;
}
for (i = 0; i < incount; i++)
{
p1 = inverts+i*C;
if (keep[i] == KEEP_BORDER)
{
out = outverts+outcount++*C;
VectorCopy(p1, out);
continue;
}
if (keep[i] == KEEP_KEEP)
{
out = outverts+outcount++*C;
VectorCopy(p1, out);
}
if (keep[i+1] == KEEP_BORDER || keep[i] == keep[i+1])
continue;
p2 = inverts+((i+1)%incount)*C;
d = dotv[i] - dotv[i+1];
if (d)
d = dotv[i] / d;
out = outverts+outcount++*C;
VectorInterpolate(p1, d, p2, out);
}
return outcount;
}
//the plane itself must be a vec4_t, but can have other data packed between
size_t Fragment_ClipPlaneToBrush(vecV_t *points, size_t maxpoints, void *planes, size_t planestride, size_t numplanes, vec4_t face)
{
int p, a;
vec4_t verts[MAXFRAGMENTVERTS];
vec4_t verts2[MAXFRAGMENTVERTS];
vec4_t *cverts;
int flip;
// vec3_t d1, d2, n;
size_t numverts;
//generate some huge quad/poly aligned with the plane
vec3_t tmp;
vec3_t right, forward;
double t;
float *plane;
// if (face[2] != 1)
// return 0;
t = fabs(face[2]);
if (t > fabs(face[0]) && t > fabs(face[1]))
VectorSet(tmp, 1, 0, 0);
else
VectorSet(tmp, 0, 0, 1);
CrossProduct(face, tmp, right);
VectorNormalize(right);
CrossProduct(face, right, forward);
VectorNormalize(forward);
VectorScale(face, face[3], verts[0]);
VectorMA(verts[0], 32767, right, verts[0]);
VectorMA(verts[0], 32767, forward, verts[0]);
VectorScale(face, face[3], verts[1]);
VectorMA(verts[1], 32767, right, verts[1]);
VectorMA(verts[1], -32767, forward, verts[1]);
VectorScale(face, face[3], verts[2]);
VectorMA(verts[2], -32767, right, verts[2]);
VectorMA(verts[2], -32767, forward, verts[2]);
VectorScale(face, face[3], verts[3]);
VectorMA(verts[3], -32767, right, verts[3]);
VectorMA(verts[3], 32767, forward, verts[3]);
numverts = 4;
//clip the quad to the various other planes
flip = 0;
for (p = 0; p < numplanes; p++)
{
plane = (float*)((qbyte*)planes + p*planestride);
if (plane != face)
{
vec3_t norm;
flip^=1;
VectorNegate(plane, norm);
if (flip)
numverts = Fragment_ClipPolyToPlane((float*)verts, (float*)verts2, numverts, norm, -plane[3]);
else
numverts = Fragment_ClipPolyToPlane((float*)verts2, (float*)verts, numverts, norm, -plane[3]);
if (numverts < 3) //totally clipped.
return 0;
}
}
if (numverts > maxpoints)
return 0;
if (flip)
cverts = verts2;
else
cverts = verts;
for (p = 0; p < numverts; p++)
{
for (a = 0; a < 3; a++)
{
float f = cverts[p][a];
int rounded = floor(f + 0.5);
//if its within 1/1000th of a qu, just round it.
if (fabs(f - rounded) < 0.001)
points[p][a] = rounded;
else
points[p][a] = f;
}
}
return numverts;
}
#ifndef SERVERONLY
#define MAXFRAGMENTTRIS 256
vec3_t decalfragmentverts[MAXFRAGMENTTRIS*3];
struct fragmentdecal_s
{
vec3_t center;
vec3_t normal;
// vec3_t tangent1;
// vec3_t tangent2;
vec3_t planenorm[6];
float planedist[6];
int numplanes;
vec_t radius;
//will only appear on surfaces with the matching surfaceflag
unsigned int surfflagmask;
unsigned int surfflagmatch;
void (*callback)(void *ctx, vec3_t *fte_restrict points, size_t numpoints, shader_t *shader);
void *ctx;
};
//#define SHOWCLIPS
//#define FRAGMENTASTRIANGLES //works, but produces more fragments.
#ifdef FRAGMENTASTRIANGLES
//if the triangle is clipped away, go recursive if there are tris left.
static void Fragment_ClipTriToPlane(int trinum, float *plane, float planedist, fragmentdecal_t *dec)
{
float *point[3];
float dotv[3];
vec3_t impact1, impact2;
float t;
int i, i2, i3;
int clippedverts = 0;
for (i = 0; i < 3; i++)
{
point[i] = decalfragmentverts[trinum*3+i];
dotv[i] = DotProduct(point[i], plane)-planedist;
clippedverts += dotv[i] < 0;
}
//if they're all clipped away, scrap the tri
switch (clippedverts)
{
case 0:
return; //plane does not clip the triangle.
case 1: //split into 3, disregard the clipped vert
for (i = 0; i < 3; i++)
{
if (dotv[i] < 0)
{ //This is the vertex that's getting clipped.
if (dotv[i] > -DIST_EPSILON)
return; //it's only over the line by a tiny ammount.
i2 = (i+1)%3;
i3 = (i+2)%3;
if (dotv[i2] < DIST_EPSILON)
return;
if (dotv[i3] < DIST_EPSILON)
return;
//work out where the two lines impact the plane
t = (dotv[i]) / (dotv[i]-dotv[i2]);
VectorInterpolate(point[i], t, point[i2], impact1);
t = (dotv[i]) / (dotv[i]-dotv[i3]);
VectorInterpolate(point[i], t, point[i3], impact2);
#ifdef SHOWCLIPS
if (dec->numtris != MAXFRAGMENTTRIS)
{
VectorCopy(impact2, decalfragmentverts[dec->numtris*3+0]);
VectorCopy(decalfragmentverts[trinum*3+i], decalfragmentverts[dec->numtris*3+1]);
VectorCopy(impact1, decalfragmentverts[dec->numtris*3+2]);
dec->numtris++;
}
#endif
//shrink the tri, putting the impact into the killed vertex.
VectorCopy(impact2, point[i]);
if (dec->numtris == MAXFRAGMENTTRIS)
return; //:(
//build the second tri
VectorCopy(impact1, decalfragmentverts[dec->numtris*3+0]);
VectorCopy(decalfragmentverts[trinum*3+i2], decalfragmentverts[dec->numtris*3+1]);
VectorCopy(impact2, decalfragmentverts[dec->numtris*3+2]);
dec->numtris++;
return;
}
}
Sys_Error("Fragment_ClipTriToPlane: Clipped vertex not founc\n");
return; //can't handle it
case 2: //split into 3, disregarding both the clipped.
for (i = 0; i < 3; i++)
{
if (!(dotv[i] < 0))
{ //This is the vertex that's staying.
if (dotv[i] < DIST_EPSILON)
break; //only just inside
i2 = (i+1)%3;
i3 = (i+2)%3;
//work out where the two lines impact the plane
t = (dotv[i]) / (dotv[i]-dotv[i2]);
VectorInterpolate(point[i], t, point[i2], impact1);
t = (dotv[i]) / (dotv[i]-dotv[i3]);
VectorInterpolate(point[i], t, point[i3], impact2);
//shrink the tri, putting the impact into the killed vertex.
#ifdef SHOWCLIPS
if (dec->numtris != MAXFRAGMENTTRIS)
{
VectorCopy(impact1, decalfragmentverts[dec->numtris*3+0]);
VectorCopy(point[i2], decalfragmentverts[dec->numtris*3+1]);
VectorCopy(point[i3], decalfragmentverts[dec->numtris*3+2]);
dec->numtris++;
}
if (dec->numtris != MAXFRAGMENTTRIS)
{
VectorCopy(impact1, decalfragmentverts[dec->numtris*3+0]);
VectorCopy(point[i3], decalfragmentverts[dec->numtris*3+1]);
VectorCopy(impact2, decalfragmentverts[dec->numtris*3+2]);
dec->numtris++;
}
#endif
VectorCopy(impact1, point[i2]);
VectorCopy(impact2, point[i3]);
return;
}
}
case 3://scrap it
//fill the verts with the verts of the last and go recursive (due to the nature of Fragment_ClipTriangle, which doesn't actually know if we clip them away)
#ifndef SHOWCLIPS
dec->numtris--;
VectorCopy(decalfragmentverts[dec->numtris*3+0], decalfragmentverts[trinum*3+0]);
VectorCopy(decalfragmentverts[dec->numtris*3+1], decalfragmentverts[trinum*3+1]);
VectorCopy(decalfragmentverts[dec->numtris*3+2], decalfragmentverts[trinum*3+2]);
if (trinum < dec->numtris)
Fragment_ClipTriToPlane(trinum, plane, planedist, dec);
#endif
return;
}
}
static void Fragment_ClipTriangle(fragmentdecal_t *dec, float *a, float *b, float *c)
{
//emit the triangle, and clip it's fragments.
int start, i;
int p;
if (dec->numtris == MAXFRAGMENTTRIS)
return; //:(
start = dec->numtris;
VectorCopy(a, decalfragmentverts[dec->numtris*3+0]);
VectorCopy(b, decalfragmentverts[dec->numtris*3+1]);
VectorCopy(c, decalfragmentverts[dec->numtris*3+2]);
dec->numtris++;
//clip all the fragments to all of the planes.
//This will produce a quad if the source triangle was big enough.
for (p = 0; p < 6; p++)
{
for (i = start; i < dec->numtris; i++)
Fragment_ClipTriToPlane(i, dec->planenorm[p], dec->plantdist[p], dec);
}
}
#else
void Fragment_ClipPoly(fragmentdecal_t *dec, int numverts, float *inverts, shader_t *surfshader)
{
//emit the triangle, and clip it's fragments.
int p;
float verts[MAXFRAGMENTVERTS*C];
float verts2[MAXFRAGMENTVERTS*C];
float *cverts;
int flip;
vec3_t d1, d2, n;
size_t numtris;
if (numverts > MAXFRAGMENTTRIS)
return;
if (r_decal_noperpendicular.ival)
{
VectorSubtract(inverts+C*1, inverts+C*0, d1);
for (p = 2; ; p++)
{
if (p >= numverts)
return;
VectorSubtract(inverts+C*p, inverts+C*0, d2);
CrossProduct(d1, d2, n);
if (DotProduct(n,n)>.1)
break;
}
VectorNormalizeFast(n);
if (DotProduct(n, dec->normal) < 0.1)
return; //faces too far way from the normal
}
//clip to the first plane specially, so we don't have extra copys
numverts = Fragment_ClipPolyToPlane(inverts, verts, numverts, dec->planenorm[0], dec->planedist[0]);
//clip the triangle to the 6 planes.
flip = 0;
for (p = 1; p < dec->numplanes; p++)
{
flip^=1;
if (flip)
numverts = Fragment_ClipPolyToPlane(verts, verts2, numverts, dec->planenorm[p], dec->planedist[p]);
else
numverts = Fragment_ClipPolyToPlane(verts2, verts, numverts, dec->planenorm[p], dec->planedist[p]);
if (numverts < 3) //totally clipped.
return;
}
if (flip)
cverts = verts2;
else
cverts = verts;
//decompose the resultant polygon into triangles.
numtris = 0;
while(numverts-->2)
{
if (numtris == MAXFRAGMENTTRIS)
{
dec->callback(dec->ctx, decalfragmentverts, numtris, NULL);
numtris = 0;
break;
}
VectorCopy((cverts+C*0), decalfragmentverts[numtris*3+0]);
VectorCopy((cverts+C*(numverts-1)), decalfragmentverts[numtris*3+1]);
VectorCopy((cverts+C*numverts), decalfragmentverts[numtris*3+2]);
numtris++;
}
if (numtris)
dec->callback(dec->ctx, decalfragmentverts, numtris, surfshader);
}
#endif
//this could be inlined, but I'm lazy.
static void Fragment_Mesh (fragmentdecal_t *dec, mesh_t *mesh, mtexinfo_t *texinfo)
{
int i;
vecV_t verts[3];
shader_t *surfshader = texinfo->texture->shader;
if ((surfshader->flags & SHADER_NOMARKS) || !mesh)
return;
if (dec->surfflagmask)
{
if ((texinfo->flags & dec->surfflagmask) != dec->surfflagmatch)
return;
}
/*if its a triangle fan/poly/quad then we can just submit the entire thing without generating extra fragments*/
if (mesh->istrifan)
{
Fragment_ClipPoly(dec, mesh->numvertexes, mesh->xyz_array[0], surfshader);
return;
}
//Fixme: optimise q3 patches (quad strips with bends between each strip)
/*otherwise it goes in and out in weird places*/
for (i = 0; i < mesh->numindexes; i+=3)
{
VectorCopy(mesh->xyz_array[mesh->indexes[i+0]], verts[0]);
VectorCopy(mesh->xyz_array[mesh->indexes[i+1]], verts[1]);
VectorCopy(mesh->xyz_array[mesh->indexes[i+2]], verts[2]);
Fragment_ClipPoly(dec, 3, verts[0], surfshader);
}
}
#ifdef Q1BSPS
static void Q1BSP_ClipDecalToNodes (model_t *mod, fragmentdecal_t *dec, mnode_t *node)
{
mplane_t *splitplane;
float dist;
msurface_t *surf;
int i;
if (node->contents < 0)
return;
splitplane = node->plane;
dist = DotProduct (dec->center, splitplane->normal) - splitplane->dist;
if (dist > dec->radius)
{
Q1BSP_ClipDecalToNodes (mod, dec, node->children[0]);
return;
}
if (dist < -dec->radius)
{
Q1BSP_ClipDecalToNodes (mod, dec, node->children[1]);
return;
}
// mark the polygons
surf = mod->surfaces + node->firstsurface;
if (r_decal_noperpendicular.ival)
{
for (i=0 ; i<node->numsurfaces ; i++, surf++)
{
if (surf->flags & SURF_PLANEBACK)
{
if (-DotProduct(surf->plane->normal, dec->normal) > -0.5)
continue;
}
else
{
if (DotProduct(surf->plane->normal, dec->normal) > -0.5)
continue;
}
Fragment_Mesh(dec, surf->mesh, surf->texinfo);
}
}
else
{
for (i=0 ; i<node->numsurfaces ; i++, surf++)
Fragment_Mesh(dec, surf->mesh, surf->texinfo);
}
Q1BSP_ClipDecalToNodes (mod, dec, node->children[0]);
Q1BSP_ClipDecalToNodes (mod, dec, node->children[1]);
}
#endif
#ifdef RTLIGHTS
extern int sh_shadowframe;
#else
static int sh_shadowframe;
#endif
#ifdef Q3BSPS
static void Q3BSP_ClipDecalToNodes (fragmentdecal_t *dec, mnode_t *node)
{
mplane_t *splitplane;
float dist;
msurface_t **msurf;
msurface_t *surf;
mleaf_t *leaf;
int i;
if (node->contents != -1)
{
leaf = (mleaf_t *)node;
// mark the polygons
msurf = leaf->firstmarksurface;
for (i=0 ; i<leaf->nummarksurfaces ; i++, msurf++)
{
surf = *msurf;
//only check each surface once. it can appear in multiple leafs.
if (surf->shadowframe == sh_shadowframe)
continue;
surf->shadowframe = sh_shadowframe;
Fragment_Mesh(dec, surf->mesh, surf->texinfo);
}
return;
}
splitplane = node->plane;
dist = DotProduct (dec->center, splitplane->normal) - splitplane->dist;
if (dist > dec->radius)
{
Q3BSP_ClipDecalToNodes (dec, node->children[0]);
return;
}
if (dist < -dec->radius)
{
Q3BSP_ClipDecalToNodes (dec, node->children[1]);
return;
}
Q3BSP_ClipDecalToNodes (dec, node->children[0]);
Q3BSP_ClipDecalToNodes (dec, node->children[1]);
}
#endif
void Mod_ClipDecal(struct model_s *mod, vec3_t center, vec3_t normal, vec3_t tangent1, vec3_t tangent2, float size, unsigned int surfflagmask, unsigned int surfflagmatch, void (*callback)(void *ctx, vec3_t *fte_restrict points, size_t numpoints, shader_t *shader), void *ctx)
{ //quad marks a full, independant quad
int p;
float r;
fragmentdecal_t dec;
VectorCopy(center, dec.center);
VectorCopy(normal, dec.normal);
dec.radius = 0;
dec.callback = callback;
dec.ctx = ctx;
dec.surfflagmask = surfflagmask;
dec.surfflagmatch = surfflagmatch;
VectorCopy(tangent1, dec.planenorm[0]);
VectorNegate(tangent1, dec.planenorm[1]);
VectorCopy(tangent2, dec.planenorm[2]);
VectorNegate(tangent2, dec.planenorm[3]);
VectorCopy(dec.normal, dec.planenorm[4]);
VectorNegate(dec.normal, dec.planenorm[5]);
for (p = 0; p < 6; p++)
{
r = sqrt(DotProduct(dec.planenorm[p], dec.planenorm[p]));
VectorScale(dec.planenorm[p], 1/r, dec.planenorm[p]);
r*= size/2;
if (r > dec.radius)
dec.radius = r;
dec.planedist[p] = -(r - DotProduct(dec.center, dec.planenorm[p]));
}
dec.numplanes = 6;
sh_shadowframe++;
if (!mod || mod->loadstate != MLS_LOADED || mod->type != mod_brush)
{
}
#ifdef Q1BSPS
else if (mod->fromgame == fg_quake || mod->fromgame == fg_halflife)
Q1BSP_ClipDecalToNodes(mod, &dec, mod->rootnode);
#endif
#ifdef Q3BSPS
else if (mod->fromgame == fg_quake3)
{
if (mod->submodelof)
{
msurface_t *surf;
for (surf = mod->surfaces+mod->firstmodelsurface, p = 0; p < mod->nummodelsurfaces; p++, surf++)
Fragment_Mesh(&dec, surf->mesh, surf->texinfo);
}
else
Q3BSP_ClipDecalToNodes(&dec, mod->rootnode);
}
#endif
#ifdef TERRAIN
if (mod->terrain)
Terrain_ClipDecal(&dec, center, dec.radius, mod);
#endif
}
#endif
/*
Decal functions
============================================================================
Physics functions (common)
*/
#ifdef Q1BSPS
void Q1BSP_CheckHullNodes(hull_t *hull)
{
int num, c;
mclipnode_t *node;
for (num = hull->firstclipnode; num < hull->lastclipnode; num++)
{
node = hull->clipnodes + num;
for (c = 0; c < 2; c++)
if (node->children[c] >= 0)
if (node->children[c] < hull->firstclipnode || node->children[c] > hull->lastclipnode)
Sys_Error ("Q1BSP_CheckHull: bad node number");
}
}
static int Q1_ModelPointContents (mnode_t *node, const vec3_t p)
{
float d;
mplane_t *plane;
while(node->contents >= 0)
{
plane = node->plane;
if (plane->type < 3)
d = p[plane->type] - plane->dist;
else
d = DotProduct(plane->normal, p) - plane->dist;
node = node->children[d<0];
}
return node->contents;
}
#endif//Q1BSPS
/*
==================
SV_HullPointContents
==================
*/
static int Q1_HullPointContents (hull_t *hull, int num, const vec3_t p)
{
float d;
mclipnode_t *node;
mplane_t *plane;
while (num >= 0)
{
node = hull->clipnodes + num;
plane = hull->planes + node->planenum;
if (plane->type < 3)
d = p[plane->type] - plane->dist;
else
d = DotProduct (plane->normal, p) - plane->dist;
if (d < 0)
num = node->children[1];
else
num = node->children[0];
}
return num;
}
#define DIST_EPSILON (0.03125)
#if 1
static const unsigned int q1toftecontents[] =
{
0,//EMPTY
FTECONTENTS_SOLID,//SOLID
FTECONTENTS_WATER,//WATER
FTECONTENTS_SLIME,//SLIME
FTECONTENTS_LAVA,//LAVA
FTECONTENTS_SKY,//SKY
FTECONTENTS_SOLID,//STRIPPED
FTECONTENTS_PLAYERCLIP,//CLIP
Q2CONTENTS_CURRENT_0,//FLOW_1
Q2CONTENTS_CURRENT_90,//FLOW_2
Q2CONTENTS_CURRENT_180,//FLOW_3
Q2CONTENTS_CURRENT_270,//FLOW_4
Q2CONTENTS_CURRENT_UP,//FLOW_5
Q2CONTENTS_CURRENT_DOWN,//FLOW_6
Q2CONTENTS_WINDOW,//TRANS
FTECONTENTS_LADDER,//LADDER
};
enum
{
rht_solid,
rht_empty,
rht_impact
};
struct rhtctx_s
{
unsigned int checkcontents;
vec3_t start, end;
mclipnode_t *clipnodes;
mplane_t *planes;
};
static int Q1BSP_RecursiveHullTrace (struct rhtctx_s *ctx, int num, float p1f, float p2f, const vec3_t p1, const vec3_t p2, trace_t *trace)
{
mclipnode_t *node;
mplane_t *plane;
float t1, t2;
vec3_t mid;
int side;
float midf;
int rht;
reenter:
if (num < 0)
{
unsigned int c = q1toftecontents[-1-num];
/*hit a leaf*/
if (c & ctx->checkcontents)
{
trace->contents = c;
if (trace->allsolid)
trace->startsolid = true;
return rht_solid;
}
else
{
trace->allsolid = false;
if (c & FTECONTENTS_FLUID)
trace->inwater = true;
else
trace->inopen = true;
return rht_empty;
}
}
/*its a node*/
/*get the node info*/
node = ctx->clipnodes + num;
plane = ctx->planes + node->planenum;
if (plane->type < 3)
{
t1 = p1[plane->type] - plane->dist;
t2 = p2[plane->type] - plane->dist;
}
else
{
t1 = DotProduct (plane->normal, p1) - plane->dist;
t2 = DotProduct (plane->normal, p2) - plane->dist;
}
/*if its completely on one side, resume on that side*/
if (t1 >= 0 && t2 >= 0)
{
//return Q1BSP_RecursiveHullTrace (hull, node->children[0], p1f, p2f, p1, p2, trace);
num = node->children[0];
goto reenter;
}
if (t1 < 0 && t2 < 0)
{
//return Q1BSP_RecursiveHullTrace (hull, node->children[1], p1f, p2f, p1, p2, trace);
num = node->children[1];
goto reenter;
}
if (plane->type < 3)
{
t1 = ctx->start[plane->type] - plane->dist;
t2 = ctx->end[plane->type] - plane->dist;
}
else
{
t1 = DotProduct (plane->normal, ctx->start) - plane->dist;
t2 = DotProduct (plane->normal, ctx->end) - plane->dist;
}
side = t1 < 0;
midf = t1 / (t1 - t2);
if (midf < p1f) midf = p1f;
if (midf > p2f) midf = p2f;
VectorInterpolate(ctx->start, midf, ctx->end, mid);
//check the near side
rht = Q1BSP_RecursiveHullTrace(ctx, node->children[side], p1f, midf, p1, mid, trace);
if (rht != rht_empty && !trace->allsolid)
return rht;
//check the far side
rht = Q1BSP_RecursiveHullTrace(ctx, node->children[side^1], midf, p2f, mid, p2, trace);
if (rht != rht_solid)
return rht;
if (side)
{
/*we impacted the back of the node, so flip the plane*/
trace->plane.dist = -plane->dist;
VectorNegate(plane->normal, trace->plane.normal);
midf = (t1 + DIST_EPSILON) / (t1 - t2);
}
else
{
/*we impacted the front of the node*/
trace->plane.dist = plane->dist;
VectorCopy(plane->normal, trace->plane.normal);
midf = (t1 - DIST_EPSILON) / (t1 - t2);
}
t1 = DotProduct (trace->plane.normal, ctx->start) - trace->plane.dist;
t2 = DotProduct (trace->plane.normal, ctx->end) - trace->plane.dist;
midf = (t1 - DIST_EPSILON) / (t1 - t2);
midf = bound(0, midf, 1);
trace->fraction = midf;
VectorCopy (mid, trace->endpos);
VectorInterpolate(ctx->start, midf, ctx->end, trace->endpos);
return rht_impact;
}
qboolean Q1BSP_RecursiveHullCheck (hull_t *hull, int num, const vec3_t p1, const vec3_t p2, unsigned int hitcontents, trace_t *trace)
{
if (VectorEquals(p1, p2))
{
/*points cannot cross planes, so do it faster*/
int q1 = Q1_HullPointContents(hull, num, p1);
unsigned int c = q1toftecontents[-1-q1];
trace->contents = c;
if (c & hitcontents)
trace->startsolid = true;
else if (c & FTECONTENTS_FLUID)
{
trace->allsolid = false;
trace->inwater = true;
}
else
{
trace->allsolid = false;
trace->inopen = true;
}
return true;
}
else
{
struct rhtctx_s ctx;
VectorCopy(p1, ctx.start);
VectorCopy(p2, ctx.end);
ctx.clipnodes = hull->clipnodes;
ctx.planes = hull->planes;
ctx.checkcontents = hitcontents;
return Q1BSP_RecursiveHullTrace(&ctx, num, 0, 1, p1, p2, trace) != rht_impact;
}
}
#else
qboolean Q1BSP_RecursiveHullCheck (hull_t *hull, int num, float p1f, float p2f, vec3_t p1, vec3_t p2, trace_t *trace)
{
mclipnode_t *node;
mplane_t *plane;
float t1, t2;
float frac;
int i;
vec3_t mid;
int side;
float midf;
// check for empty
if (num < 0)
{
if (num != Q1CONTENTS_SOLID)
{
trace->allsolid = false;
if (num == Q1CONTENTS_EMPTY)
trace->inopen = true;
else
trace->inwater = true;
}
else
trace->startsolid = true;
return true; // empty
}
//
// find the point distances
//
node = hull->clipnodes + num;
plane = hull->planes + node->planenum;
if (plane->type < 3)
{
t1 = p1[plane->type] - plane->dist;
t2 = p2[plane->type] - plane->dist;
}
else
{
t1 = DotProduct (plane->normal, p1) - plane->dist;
t2 = DotProduct (plane->normal, p2) - plane->dist;
}
#if 1
if (t1 >= 0 && t2 >= 0)
return Q1BSP_RecursiveHullCheck (hull, node->children[0], p1f, p2f, p1, p2, trace);
if (t1 < 0 && t2 < 0)
return Q1BSP_RecursiveHullCheck (hull, node->children[1], p1f, p2f, p1, p2, trace);
#else
if ( (t1 >= DIST_EPSILON && t2 >= DIST_EPSILON) || (t2 > t1 && t1 >= 0) )
return Q1BSP_RecursiveHullCheck (hull, node->children[0], p1f, p2f, p1, p2, trace);
if ( (t1 <= -DIST_EPSILON && t2 <= -DIST_EPSILON) || (t2 < t1 && t1 <= 0) )
return Q1BSP_RecursiveHullCheck (hull, node->children[1], p1f, p2f, p1, p2, trace);
#endif
// put the crosspoint DIST_EPSILON pixels on the near side
if (t1 < 0)
frac = (t1 + DIST_EPSILON)/(t1-t2);
else
frac = (t1 - DIST_EPSILON)/(t1-t2);
if (frac < 0)
frac = 0;
if (frac > 1)
frac = 1;
midf = p1f + (p2f - p1f)*frac;
for (i=0 ; i<3 ; i++)
mid[i] = p1[i] + frac*(p2[i] - p1[i]);
side = (t1 < 0);
// move up to the node
if (!Q1BSP_RecursiveHullCheck (hull, node->children[side], p1f, midf, p1, mid, trace) )
return false;
#ifdef PARANOID
if (Q1BSP_RecursiveHullCheck (sv_hullmodel, mid, node->children[side])
== Q1CONTENTS_SOLID)
{
Con_Printf ("mid PointInHullSolid\n");
return false;
}
#endif
if (Q1_HullPointContents (hull, node->children[side^1], mid)
!= Q1CONTENTS_SOLID)
// go past the node
return Q1BSP_RecursiveHullCheck (hull, node->children[side^1], midf, p2f, mid, p2, trace);
if (trace->allsolid)
return false; // never got out of the solid area
//==================
// the other side of the node is solid, this is the impact point
//==================
if (!side)
{
VectorCopy (plane->normal, trace->plane.normal);
trace->plane.dist = plane->dist;
}
else
{
VectorNegate (plane->normal, trace->plane.normal);
trace->plane.dist = -plane->dist;
}
while (Q1_HullPointContents (hull, hull->firstclipnode, mid)
== Q1CONTENTS_SOLID)
{ // shouldn't really happen, but does occasionally
if (!(frac < 10000000) && !(frac > -10000000))
{
trace->fraction = 0;
VectorClear (trace->endpos);
Con_Printf ("nan in traceline\n");
return false;
}
frac -= 0.1;
if (frac < 0)
{
trace->fraction = midf;
VectorCopy (mid, trace->endpos);
Con_DPrintf ("backup past 0\n");
return false;
}
midf = p1f + (p2f - p1f)*frac;
for (i=0 ; i<3 ; i++)
mid[i] = p1[i] + frac*(p2[i] - p1[i]);
}
trace->fraction = midf;
VectorCopy (mid, trace->endpos);
return false;
}
#endif
#ifdef Q1BSPS
/*
the bsp tree we're walking through is the renderable hull
we need to trace a box through the world.
by its very nature, this will reach more nodes than we really want, and as we can follow a node sideways, the underlying bsp structure is no longer 100% reliable (meaning we cross planes that are entirely to one side, and follow its children too)
so all contents and solidity must come from the brushes and ONLY the brushes.
*/
struct traceinfo_s
{
unsigned int solidcontents;
trace_t trace;
qboolean capsule;
float radius;
/*set even for sphere traces (used for bbox tests)*/
vec3_t mins;
vec3_t maxs;
vec3_t start;
vec3_t end;
vec3_t up;
vec3_t capsulesize;
vec3_t extents;
};
static void Q1BSP_ClipToBrushes(struct traceinfo_s *traceinfo, mbrush_t *brush)
{
struct mbrushplane_s *plane;
struct mbrushplane_s *enterplane;
int i, j;
vec3_t ofs;
qboolean startout, endout;
float d1,d2,dist,enterdist=0;
float f, enterfrac, exitfrac;
for (; brush; brush = brush->next)
{
/*ignore if its not solid to us*/
if (!(traceinfo->solidcontents & brush->contents))
continue;
startout = false;
endout = false;
enterplane= NULL;
enterfrac = -1;
exitfrac = 10;
for (i = brush->numplanes, plane = brush->planes; i; i--, plane++)
{
/*calculate the distance based upon the shape of the object we're tracing for*/
if (traceinfo->capsule)
{
dist = DotProduct(traceinfo->up, plane->normal);
dist = dist*(traceinfo->capsulesize[(dist<0)?1:2]) - traceinfo->capsulesize[0];
dist = plane->dist - dist;
//dist = plane->dist + traceinfo->radius;
}
else
{
for (j=0 ; j<3 ; j++)
{
if (plane->normal[j] < 0)
ofs[j] = traceinfo->maxs[j];
else
ofs[j] = traceinfo->mins[j];
}
dist = DotProduct (ofs, plane->normal);
dist = plane->dist - dist;
}
d1 = DotProduct (traceinfo->start, plane->normal) - dist;
d2 = DotProduct (traceinfo->end, plane->normal) - dist;
if (d1 >= 0)
startout = true;
if (d2 > 0)
endout = true;
//if we're fully outside any plane, then we cannot possibly enter the brush, skip to the next one
if (d1 > 0 && d2 >= 0)
goto nextbrush;
//if we're fully inside the plane, then whatever is happening is not relevent for this plane
if (d1 < 0 && d2 <= 0)
continue;
f = d1 / (d1-d2);
if (d1 > d2)
{
//entered the brush. favour the furthest fraction to avoid extended edges (yay for convex shapes)
if (enterfrac < f)
{
enterfrac = f;
enterplane = plane;
enterdist = dist;
}
}
else
{
//left the brush, favour the nearest plane (smallest frac)
if (exitfrac > f)
{
exitfrac = f;
}
}
}
if (!startout)
{
traceinfo->trace.startsolid = true;
if (!endout)
traceinfo->trace.allsolid = true;
traceinfo->trace.contents |= brush->contents;
return;
}
if (enterfrac != -1 && enterfrac < exitfrac)
{
//impact!
if (enterfrac < traceinfo->trace.fraction)
{
traceinfo->trace.fraction = enterfrac;
traceinfo->trace.plane.dist = enterdist;
VectorCopy(enterplane->normal, traceinfo->trace.plane.normal);
traceinfo->trace.contents = brush->contents;
}
}
nextbrush:
;
}
}
static void Q1BSP_InsertBrush(mnode_t *node, mbrush_t *brush, vec3_t bmins, vec3_t bmaxs)
{
vec3_t nearp, farp;
float nd, fd;
int i;
while(1)
{
if (node->contents < 0) /*leaf, so no smaller node to put it in (I'd be surprised if it got this far)*/
{
brush->next = node->brushes;
node->brushes = brush;
return;
}
for (i = 0; i < 3; i++)
{
if (node->plane->normal[i] > 0)
{
nearp[i] = bmins[i];
farp[i] = bmaxs[i];
}
else
{
nearp[i] = bmaxs[i];
farp[i] = bmins[i];
}
}
nd = DotProduct(node->plane->normal, nearp) - node->plane->dist;
fd = DotProduct(node->plane->normal, farp) - node->plane->dist;
/*if its fully on either side, continue walking*/
if (nd < 0 && fd < 0)
node = node->children[1];
else if (nd > 0 && fd > 0)
node = node->children[0];
else
{
/*plane crosses bbox, so insert here*/
brush->next = node->brushes;
node->brushes = brush;
return;
}
}
}
static void Q1BSP_RecursiveBrushCheck (struct traceinfo_s *traceinfo, mnode_t *node, float p1f, float p2f, const vec3_t p1, const vec3_t p2)
{
mplane_t *plane;
float t1, t2;
float frac;
int i;
vec3_t mid;
int side;
float midf;
float offset;
if (node->brushes)
{
Q1BSP_ClipToBrushes(traceinfo, node->brushes);
}
if (traceinfo->trace.fraction < p1f)
{
//already hit something closer than this node
return;
}
if (node->contents < 0)
{
//we're in a leaf
return;
}
//
// find the point distances
//
plane = node->plane;
if (plane->type < 3)
{
t1 = p1[plane->type] - plane->dist;
t2 = p2[plane->type] - plane->dist;
if (plane->normal[plane->type] < 0)
offset = -traceinfo->mins[plane->type];
else
offset = traceinfo->maxs[plane->type];
}
else
{
t1 = DotProduct (plane->normal, p1) - plane->dist;
t2 = DotProduct (plane->normal, p2) - plane->dist;
offset = 0;
for (i = 0; i < 3; i++)
{
if (plane->normal[i] < 0)
offset += plane->normal[i] * -traceinfo->mins[i];
else
offset += plane->normal[i] * traceinfo->maxs[i];
}
}
/*if we're fully on one side of the trace, go only down that side*/
if (t1 >= offset && t2 >= offset)
{
Q1BSP_RecursiveBrushCheck (traceinfo, node->children[0], p1f, p2f, p1, p2);
return;
}
if (t1 < -offset && t2 < -offset)
{
Q1BSP_RecursiveBrushCheck (traceinfo, node->children[1], p1f, p2f, p1, p2);
return;
}
// put the crosspoint DIST_EPSILON pixels on the near side
if (t1 == t2)
{
side = 0;
frac = 0;
}
else if (t1 < 0)
{
frac = (t1 + DIST_EPSILON)/(t1-t2);
side = 1;
}
else
{
frac = (t1 - DIST_EPSILON)/(t1-t2);
side = 0;
}
if (frac < 0)
frac = 0;
if (frac > 1)
frac = 1;
midf = p1f + (p2f - p1f)*frac;
for (i=0 ; i<3 ; i++)
mid[i] = p1[i] + frac*(p2[i] - p1[i]);
// move up to the node
Q1BSP_RecursiveBrushCheck (traceinfo, node->children[side], p1f, midf, p1, mid);
// go past the node
Q1BSP_RecursiveBrushCheck (traceinfo, node->children[side^1], midf, p2f, mid, p2);
}
#endif //Q1BSPS
static unsigned int Q1BSP_TranslateContents(int contents)
{
switch(contents)
{
case Q1CONTENTS_EMPTY:
return FTECONTENTS_EMPTY;
case Q1CONTENTS_SOLID:
return FTECONTENTS_SOLID;
case Q1CONTENTS_WATER:
return FTECONTENTS_WATER;
case Q1CONTENTS_SLIME:
return FTECONTENTS_SLIME;
case Q1CONTENTS_LAVA:
return FTECONTENTS_LAVA;
case Q1CONTENTS_SKY:
return FTECONTENTS_SKY|FTECONTENTS_PLAYERCLIP|FTECONTENTS_MONSTERCLIP;
case Q1CONTENTS_LADDER:
return FTECONTENTS_LADDER;
case Q1CONTENTS_CLIP:
return FTECONTENTS_PLAYERCLIP|FTECONTENTS_MONSTERCLIP;
case Q1CONTENTS_TRANS:
return FTECONTENTS_SOLID;
//q2 is better than nothing, right?
case Q1CONTENTS_FLOW_1:
return Q2CONTENTS_CURRENT_0;
case Q1CONTENTS_FLOW_2:
return Q2CONTENTS_CURRENT_90;
case Q1CONTENTS_FLOW_3:
return Q2CONTENTS_CURRENT_180;
case Q1CONTENTS_FLOW_4:
return Q2CONTENTS_CURRENT_270;
case Q1CONTENTS_FLOW_5:
return Q2CONTENTS_CURRENT_UP;
case Q1CONTENTS_FLOW_6:
return Q2CONTENTS_CURRENT_DOWN;
default:
Con_Printf("Q1BSP_TranslateContents: Unknown contents type - %i", contents);
return FTECONTENTS_SOLID;
}
}
int Q1BSP_HullPointContents(hull_t *hull, const vec3_t p)
{
return Q1BSP_TranslateContents(Q1_HullPointContents(hull, hull->firstclipnode, p));
}
#ifdef Q1BSPS
unsigned int Q1BSP_PointContents(model_t *model, const vec3_t axis[3], const vec3_t point)
{
int contents;
if (axis)
{
vec3_t transformed;
transformed[0] = DotProduct(point, axis[0]);
transformed[1] = DotProduct(point, axis[1]);
transformed[2] = DotProduct(point, axis[2]);
return Q1BSP_PointContents(model, NULL, transformed);
}
else
{
if (!model->firstmodelsurface)
{
contents = Q1BSP_TranslateContents(Q1_ModelPointContents(model->nodes, point));
}
else
contents = Q1BSP_HullPointContents(&model->hulls[0], point);
}
#ifdef TERRAIN
if (model->terrain)
contents |= Heightmap_PointContents(model, NULL, point);
#endif
return contents;
}
void Q1BSP_LoadBrushes(model_t *model, bspx_header_t *bspx, void *mod_base)
{
struct {
unsigned int ver;
unsigned int modelnum;
unsigned int numbrushes;
unsigned int numplanes;
} *permodel;
struct {
float mins[3];
float maxs[3];
signed short contents;
unsigned short numplanes;
} *perbrush;
/*
Note to implementors:
a pointy brush with angles pointier than 90 degrees will extend further than any adjacent brush, thus creating invisible walls with larger expansions.
the engine inserts 6 axial planes acording to the bbox, thus the qbsp need not write any axial planes
note that doing it this way probably isn't good if you want to query textures...
*/
struct {
vec3_t normal;
float dist;
} *perplane;
static vec3_t axis[3] = {{1, 0, 0}, {0, 1, 0}, {0, 0, 1}};
int br, pl, remainingplanes;
mbrush_t *brush;
mnode_t *rootnode;
unsigned int lumpsizeremaining;
model->engineflags &= ~MDLF_HASBRUSHES;
permodel = BSPX_FindLump(bspx, mod_base, "BRUSHLIST", &lumpsizeremaining);
if (!permodel)
return;
while (lumpsizeremaining)
{
if (lumpsizeremaining < sizeof(*permodel))
return;
permodel->ver = LittleLong(permodel->ver);
permodel->modelnum = LittleLong(permodel->modelnum);
permodel->numbrushes = LittleLong(permodel->numbrushes);
permodel->numplanes = LittleLong(permodel->numplanes);
if (permodel->ver != 1 || lumpsizeremaining < sizeof(*permodel) + permodel->numbrushes*sizeof(*perbrush) + permodel->numplanes*sizeof(*perplane))
return;
//find the correct rootnode for the submodel (submodels are not set up yet).
rootnode = model->nodes;
if (permodel->modelnum > model->numsubmodels)
return;
rootnode += model->submodels[permodel->modelnum].headnode[0];
brush = ZG_Malloc(&model->memgroup, (sizeof(*brush) - sizeof(brush->planes[0]))*permodel->numbrushes + sizeof(brush->planes[0])*(permodel->numbrushes*6+permodel->numplanes));
remainingplanes = permodel->numplanes;
perbrush = (void*)(permodel+1);
for (br = 0; br < permodel->numbrushes; br++)
{
/*byteswap it all in place*/
perbrush->mins[0] = LittleFloat(perbrush->mins[0]);
perbrush->mins[1] = LittleFloat(perbrush->mins[1]);
perbrush->mins[2] = LittleFloat(perbrush->mins[2]);
perbrush->maxs[0] = LittleFloat(perbrush->maxs[0]);
perbrush->maxs[1] = LittleFloat(perbrush->maxs[1]);
perbrush->maxs[2] = LittleFloat(perbrush->maxs[2]);
perbrush->contents = LittleShort(perbrush->contents);
perbrush->numplanes = LittleShort(perbrush->numplanes);
/*make sure planes don't overflow*/
if (perbrush->numplanes > remainingplanes)
return;
remainingplanes-=perbrush->numplanes;
/*set up the mbrush from the file*/
brush->contents = Q1BSP_TranslateContents(perbrush->contents);
brush->numplanes = perbrush->numplanes;
for (pl = 0, perplane = (void*)(perbrush+1); pl < perbrush->numplanes; pl++, perplane++)
{
brush->planes[pl].normal[0] = LittleFloat(perplane->normal[0]);
brush->planes[pl].normal[1] = LittleFloat(perplane->normal[1]);
brush->planes[pl].normal[2] = LittleFloat(perplane->normal[2]);
brush->planes[pl].dist = LittleFloat(perplane->dist);
}
/*and add axial planes acording to the brush's bbox*/
for (pl = 0; pl < 3; pl++)
{
VectorCopy(axis[pl], brush->planes[brush->numplanes].normal);
brush->planes[brush->numplanes].dist = perbrush->maxs[pl];
brush->numplanes++;
}
for (pl = 0; pl < 3; pl++)
{
VectorNegate(axis[pl], brush->planes[brush->numplanes].normal);
brush->planes[brush->numplanes].dist = -perbrush->mins[pl];
brush->numplanes++;
}
/*link it in to the bsp tree*/
Q1BSP_InsertBrush(rootnode, brush, perbrush->mins, perbrush->maxs);
/*set up for the next brush*/
brush = (void*)&brush->planes[brush->numplanes];
perbrush = (void*)perplane;
}
/*move on to the next model*/
lumpsizeremaining -= sizeof(*permodel) + permodel->numbrushes*sizeof(*perbrush) + permodel->numplanes*sizeof(*perplane);
permodel = (void*)((char*)permodel + sizeof(*permodel) + permodel->numbrushes*sizeof(*perbrush) + permodel->numplanes*sizeof(*perplane));
}
/*parsing was successful! flag it as okay*/
model->engineflags |= MDLF_HASBRUSHES;
}
hull_t *Q1BSP_ChooseHull(model_t *model, int forcehullnum, const vec3_t mins, const vec3_t maxs, vec3_t offset)
{
hull_t *hull;
vec3_t size;
VectorSubtract (maxs, mins, size);
if (forcehullnum >= 1 && forcehullnum <= MAX_MAP_HULLSM && model->hulls[forcehullnum-1].available)
hull = &model->hulls[forcehullnum-1];
else
{
#ifdef HEXEN2
if (model->hulls[5].available)
{ //choose based on hexen2 sizes.
if (size[0] < 3) // Point
hull = &model->hulls[0];
else if (size[0] <= 8.1 && model->hulls[4].available)
hull = &model->hulls[4]; //Pentacles
else if (size[0] <= 32.1 && size[2] <= 28.1) // Half Player
hull = &model->hulls[3];
else if (size[0] <= 32.1) // Full Player
hull = &model->hulls[1];
else // Golumn
hull = &model->hulls[5];
}
else
#endif
{
if (size[0] < 3 || !model->hulls[1].available)
hull = &model->hulls[0];
else if (size[0] <= 32.1 || !model->hulls[2].available)
{
if (size[2] < 54.1 && model->hulls[3].available)
hull = &model->hulls[3]; // 32x32x36 (half-life's crouch)
else if (model->hulls[1].available)
hull = &model->hulls[1];
else
hull = &model->hulls[0];
}
else
hull = &model->hulls[2];
}
}
VectorSubtract (hull->clip_mins, mins, offset);
return hull;
}
qboolean Q1BSP_Trace(model_t *model, int forcehullnum, const framestate_t *framestate, const vec3_t axis[3], const vec3_t start, const vec3_t end, const vec3_t mins, const vec3_t maxs, qboolean capsule, unsigned int hitcontentsmask, trace_t *trace)
{
hull_t *hull;
vec3_t start_l, end_l;
vec3_t offset;
if ((model->engineflags & MDLF_HASBRUSHES))// && (size[0] || size[1] || size[2]))
{
struct traceinfo_s traceinfo;
memset (&traceinfo.trace, 0, sizeof(trace_t));
traceinfo.trace.fraction = 1;
traceinfo.trace.allsolid = false;
VectorCopy(mins, traceinfo.mins);
VectorCopy(maxs, traceinfo.maxs);
if (axis)
{
traceinfo.start[0] = DotProduct(start, axis[0]);
traceinfo.start[1] = DotProduct(start, axis[1]);
traceinfo.start[2] = DotProduct(start, axis[2]);
traceinfo.end[0] = DotProduct(end, axis[0]);
traceinfo.end[1] = DotProduct(end, axis[1]);
traceinfo.end[2] = DotProduct(end, axis[2]);
}
else
{
VectorCopy(start, traceinfo.start);
VectorCopy(end, traceinfo.end);
}
traceinfo.capsule = capsule;
if (traceinfo.capsule)
{
float ext;
traceinfo.capsulesize[0] = ((maxs[0]-mins[0]) + (maxs[1]-mins[1]))/4.0;
traceinfo.capsulesize[1] = maxs[2];
traceinfo.capsulesize[2] = mins[2];
ext = (traceinfo.capsulesize[1] > -traceinfo.capsulesize[2])?traceinfo.capsulesize[1]:-traceinfo.capsulesize[2];
traceinfo.capsulesize[1] -= traceinfo.capsulesize[0];
traceinfo.capsulesize[2] += traceinfo.capsulesize[0];
traceinfo.extents[0] = ext+1;
traceinfo.extents[1] = ext+1;
traceinfo.extents[2] = ext+1;
VectorSet(traceinfo.up, 0, 0, 1);
}
/* traceinfo.sphere = true;
traceinfo.radius = 48;
traceinfo.mins[0] = -traceinfo.radius;
traceinfo.mins[1] = -traceinfo.radius;
traceinfo.mins[2] = -traceinfo.radius;
traceinfo.maxs[0] = traceinfo.radius;
traceinfo.maxs[1] = traceinfo.radius;
traceinfo.maxs[2] = traceinfo.radius;
*/
traceinfo.solidcontents = hitcontentsmask;
Q1BSP_RecursiveBrushCheck(&traceinfo, model->rootnode, 0, 1, traceinfo.start, traceinfo.end);
memcpy(trace, &traceinfo.trace, sizeof(trace_t));
if (trace->fraction < 1)
{
float d1 = DotProduct(start, trace->plane.normal) - trace->plane.dist;
float d2 = DotProduct(end, trace->plane.normal) - trace->plane.dist;
float f = (d1 - DIST_EPSILON) / (d1 - d2);
if (f < 0)
f = 0;
trace->fraction = f;
if (axis)
{
vec3_t iaxis[3];
vec3_t norm;
Matrix3x3_RM_Invert_Simple((const 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]);
}
}
VectorInterpolate(start, trace->fraction, end, trace->endpos);
return trace->fraction != 1;
}
memset (trace, 0, sizeof(trace_t));
trace->fraction = 1;
trace->allsolid = true;
hull = Q1BSP_ChooseHull(model, forcehullnum, mins, maxs, offset);
//fix for hexen2 monsters half-walking into walls.
// if (forent.flags & FL_MONSTER)
// {
// offset[0] = 0;
// offset[1] = 0;
// }
if (axis)
{
vec3_t tmp;
VectorSubtract(start, offset, tmp);
start_l[0] = DotProduct(tmp, axis[0]);
start_l[1] = DotProduct(tmp, axis[1]);
start_l[2] = DotProduct(tmp, axis[2]);
VectorSubtract(end, offset, tmp);
end_l[0] = DotProduct(tmp, axis[0]);
end_l[1] = DotProduct(tmp, axis[1]);
end_l[2] = DotProduct(tmp, axis[2]);
Q1BSP_RecursiveHullCheck(hull, hull->firstclipnode, start_l, end_l, hitcontentsmask, trace);
if (trace->fraction == 1)
{
VectorCopy (end, trace->endpos);
}
else
{
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]);
/*just interpolate it, its easier than inverse matrix rotations*/
VectorInterpolate(start, trace->fraction, end, trace->endpos);
}
}
else
{
VectorSubtract(start, offset, start_l);
VectorSubtract(end, offset, end_l);
Q1BSP_RecursiveHullCheck(hull, hull->firstclipnode, start_l, end_l, hitcontentsmask, trace);
if (trace->fraction == 1)
{
VectorCopy (end, trace->endpos);
}
else
{
VectorAdd (trace->endpos, offset, trace->endpos);
}
}
#ifdef TERRAIN
if (model->terrain && trace->fraction)
{
trace_t hmt;
Heightmap_Trace(model, forcehullnum, framestate, axis, start, end, mins, maxs, capsule, hitcontentsmask, &hmt);
if (hmt.fraction < trace->fraction)
*trace = hmt;
}
#endif
return trace->fraction != 1;
}
/*
Physics functions (common)
============================================================================
Rendering functions (Client only)
*/
#ifndef SERVERONLY
extern int r_dlightframecount;
//goes through the nodes marking the surfaces near the dynamic light as lit.
void Q1BSP_MarkLights (dlight_t *light, dlightbitmask_t bit, mnode_t *node)
{
mplane_t *splitplane;
float dist;
msurface_t *surf;
int i;
float l, maxdist;
int j, s, t;
vec3_t impact;
if (node->contents < 0)
return;
splitplane = node->plane;
if (splitplane->type < 3)
dist = light->origin[splitplane->type] - splitplane->dist;
else
dist = DotProduct (light->origin, splitplane->normal) - splitplane->dist;
if (dist > light->radius)
{
Q1BSP_MarkLights (light, bit, node->children[0]);
return;
}
if (dist < -light->radius)
{
Q1BSP_MarkLights (light, bit, node->children[1]);
return;
}
maxdist = light->radius*light->radius;
// mark the polygons
surf = currentmodel->surfaces + node->firstsurface;
for (i=0 ; i<node->numsurfaces ; i++, surf++)
{
//Yeah, you can blame LordHavoc for this alternate code here.
for (j=0 ; j<3 ; j++)
impact[j] = light->origin[j] - surf->plane->normal[j]*dist;
// clamp center of light to corner and check brightness
l = DotProduct (impact, surf->texinfo->vecs[0]) + surf->texinfo->vecs[0][3] - surf->texturemins[0];
s = l+0.5;if (s < 0) s = 0;else if (s > surf->extents[0]) s = surf->extents[0];
s = (l - s)*surf->texinfo->vecscale[0];
l = DotProduct (impact, surf->texinfo->vecs[1]) + surf->texinfo->vecs[1][3] - surf->texturemins[1];
t = l+0.5;if (t < 0) t = 0;else if (t > surf->extents[1]) t = surf->extents[1];
t = (l - t)*surf->texinfo->vecscale[1];
// compare to minimum light
if ((s*s+t*t+dist*dist) < maxdist)
{
if (surf->dlightframe != r_dlightframecount)
{
surf->dlightbits = bit;
surf->dlightframe = r_dlightframecount;
}
else
surf->dlightbits |= bit;
}
}
Q1BSP_MarkLights (light, bit, node->children[0]);
Q1BSP_MarkLights (light, bit, node->children[1]);
}
#endif
/*
Rendering functions (Client only)
==============================================================================
Server only functions
*/
#ifndef CLIENTONLY
static qbyte *Q1BSP_ClusterPVS (model_t *model, int cluster, pvsbuffer_t *buffer, pvsmerge_t merge);
//does the recursive work of Q1BSP_FatPVS
static void SV_Q1BSP_AddToFatPVS (model_t *mod, const vec3_t org, mnode_t *node, pvsbuffer_t *pvsbuffer)
{
mplane_t *plane;
float d;
while (1)
{
// if this is a leaf, accumulate the pvs bits
if (node->contents < 0)
{
if (node->contents != Q1CONTENTS_SOLID)
{
Q1BSP_ClusterPVS(mod, ((mleaf_t *)node - mod->leafs)-1, pvsbuffer, PVM_MERGE);
}
return;
}
plane = node->plane;
d = DotProduct (org, plane->normal) - plane->dist;
if (d > 8)
node = node->children[0];
else if (d < -8)
node = node->children[1];
else
{ // go down both
SV_Q1BSP_AddToFatPVS (mod, org, node->children[0], pvsbuffer);
node = node->children[1];
}
}
}
/*
=============
Q1BSP_FatPVS
Calculates a PVS that is the inclusive or of all leafs within 8 pixels of the
given point.
=============
*/
static unsigned int Q1BSP_FatPVS (model_t *mod, const vec3_t org, pvsbuffer_t *pvsbuffer, qboolean add)
{
if (pvsbuffer->buffersize < mod->pvsbytes)
pvsbuffer->buffer = BZ_Realloc(pvsbuffer->buffer, pvsbuffer->buffersize=mod->pvsbytes);
if (!add)
Q_memset (pvsbuffer->buffer, 0, mod->pvsbytes);
SV_Q1BSP_AddToFatPVS (mod, org, mod->nodes, pvsbuffer);
return mod->pvsbytes;
}
#endif
static qboolean Q1BSP_EdictInFatPVS(model_t *mod, const struct pvscache_s *ent, const qbyte *pvs, const int *areas)
{
int i;
//if (areas)areas[0] is the area count... but q1bsp has no areas so we ignore it entirely.
if (ent->num_leafs < 0)
return true; //it's in too many leafs for us to cope with. Just trivially accept it.
for (i=0 ; i < ent->num_leafs ; i++)
if (pvs[ent->leafnums[i] >> 3] & (1 << (ent->leafnums[i]&7) ))
return true; //we might be able to see this one.
return false; //none of this ents leafs were visible, so neither is the ent.
}
/*
===============
SV_FindTouchedLeafs
Links the edict to the right leafs so we can get it's potential visability.
===============
*/
static void Q1BSP_RFindTouchedLeafs (model_t *wm, struct pvscache_s *ent, mnode_t *node, const float *mins, const float *maxs)
{
mplane_t *splitplane;
mleaf_t *leaf;
int sides;
int leafnum;
// add an efrag if the node is a leaf
if (node->contents < 0)
{
//ignore solid leafs. this should include leaf 0 (which has no pvs info)
if (node->contents == Q1CONTENTS_SOLID)
return;
if ((unsigned)ent->num_leafs >= MAX_ENT_LEAFS)
{
ent->num_leafs = -1; //too many. mark it as such so we can trivially accept huge mega-big brush models.
return;
}
leaf = (mleaf_t *)node;
leafnum = leaf - wm->leafs - 1;
ent->leafnums[ent->num_leafs] = leafnum;
ent->num_leafs++;
return;
}
// NODE_MIXED
splitplane = node->plane;
sides = BOX_ON_PLANE_SIDE(mins, maxs, splitplane);
// recurse down the contacted sides
if (sides & 1)
Q1BSP_RFindTouchedLeafs (wm, ent, node->children[0], mins, maxs);
if (sides & 2)
Q1BSP_RFindTouchedLeafs (wm, ent, node->children[1], mins, maxs);
}
static void Q1BSP_FindTouchedLeafs(model_t *mod, struct pvscache_s *ent, const float *mins, const float *maxs)
{
ent->num_leafs = 0;
if (mins && maxs)
Q1BSP_RFindTouchedLeafs (mod, ent, mod->nodes, mins, maxs);
}
/*
Server only functions
==============================================================================
PVS type stuff
*/
/*
===================
Mod_DecompressVis
===================
*/
static qbyte *Q1BSP_DecompressVis (qbyte *in, model_t *model, qbyte *decompressed, unsigned int buffersize, qboolean merge)
{
int c;
qbyte *out;
int row;
row = (model->numclusters+7)>>3;
out = decompressed;
if (buffersize < row)
row = buffersize;
if (!in)
{ // no vis info, so make all visible
while (row)
{
*out++ = 0xff;
row--;
}
return decompressed;
}
if (merge)
{
do
{
if (*in)
{
*out++ |= *in++;
continue;
}
out += in[1];
in += 2;
} while (out - decompressed < row);
}
else
{
do
{
if (*in)
{
*out++ = *in++;
continue;
}
c = in[1];
in += 2;
if ((out - decompressed) + c > row)
{
c = row - (out - decompressed);
Con_DPrintf ("warning: Vis decompression overrun\n");
}
while (c)
{
*out++ = 0;
c--;
}
} while (out - decompressed < row);
}
return decompressed;
}
static pvsbuffer_t mod_novis;
static pvsbuffer_t mod_tempvis;
void Q1BSP_Shutdown(void)
{
Z_Free(mod_novis.buffer);
memset(&mod_novis, 0, sizeof(mod_novis));
Z_Free(mod_tempvis.buffer);
memset(&mod_tempvis, 0, sizeof(mod_tempvis));
}
//pvs is 1-based. clusters are 0-based. otherwise, q1bsp has a 1:1 mapping.
static qbyte *Q1BSP_ClusterPVS (model_t *model, int cluster, pvsbuffer_t *buffer, pvsmerge_t merge)
{
if (cluster == -1)
{
if (merge == PVM_FAST)
{
if (mod_novis.buffersize < model->pvsbytes)
{
mod_novis.buffer = BZ_Realloc(mod_novis.buffer, mod_novis.buffersize=model->pvsbytes);
memset(mod_novis.buffer, 0xff, mod_novis.buffersize);
}
return mod_novis.buffer;
}
if (buffer->buffersize < model->pvsbytes)
buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=model->pvsbytes);
memset(buffer->buffer, 0xff, model->pvsbytes);
return buffer->buffer;
}
if (merge == PVM_FAST && model->pvs)
return model->pvs + cluster * model->pvsbytes;
cluster++;
if (!buffer)
buffer = &mod_tempvis;
if (buffer->buffersize < model->pvsbytes)
buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=model->pvsbytes);
return Q1BSP_DecompressVis (model->leafs[cluster].compressed_vis, model, buffer->buffer, buffer->buffersize, merge==PVM_MERGE);
}
/*static qbyte *Q1BSP_ClusterPHS (model_t *model, int cluster, pvsbuffer_t *buffer, pvsmerge_t merge)
{
if (cluster == -1 || !model->phs)
{
if (merge == PVM_FAST)
{
if (mod_novis.buffersize < model->pvsbytes)
{
mod_novis.buffer = BZ_Realloc(mod_novis.buffer, mod_novis.buffersize=model->pvsbytes);
memset(mod_novis.buffer, 0xff, mod_novis.buffersize);
}
return mod_novis.buffer;
}
if (buffer->buffersize < model->pvsbytes)
buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=model->pvsbytes);
memset(buffer->buffer, 0xff, model->pvsbytes);
return buffer->buffer;
}
if (merge == PVM_FAST)
return model->pvs + cluster * model->pvsbytes;
if (!buffer)
buffer = &mod_tempvis;
if (buffer->buffersize < model->pvsbytes)
buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=model->pvsbytes);
memcpy(buffer->buffer, model->pvs + cluster * model->pvsbytes, model->pvsbytes);
return buffer->buffer;
}*/
//returns the leaf number, which is used as a bit index into the pvs.
static int Q1BSP_LeafnumForPoint (model_t *model, vec3_t p)
{
mnode_t *node;
float d;
mplane_t *plane;
if (!model)
{
Sys_Error ("Mod_PointInLeaf: bad model");
}
if (!model->nodes)
return 0;
node = model->nodes;
while (1)
{
if (node->contents < 0)
return (mleaf_t *)node - model->leafs;
plane = node->plane;
d = DotProduct (p,plane->normal) - plane->dist;
if (d > 0)
node = node->children[0];
else
node = node->children[1];
}
return 0; // never reached
}
mleaf_t *Q1BSP_LeafForPoint (model_t *model, vec3_t p)
{
return model->leafs + Q1BSP_LeafnumForPoint(model, p);
}
static void Q1BSP_ClustersInSphere_Union(mleaf_t *firstleaf, const vec3_t center, float radius, mnode_t *node, qbyte *out, qbyte *unionwith)
{ //this is really for rtlights.
float t1, t2;
mplane_t *plane;
while (1)
{
if (node->contents < 0)
{ //leaf! mark/merge it.
size_t c = (mleaf_t *)node - firstleaf;
if (c == -1)
return;
if (unionwith)
out[c>>3] |= (1<<(c&7)) & unionwith[c>>3];
else
out[c>>3] |= (1<<(c&7));
return;
}
plane = node->plane;
if (plane->type < 3)
t1 = center[plane->type] - plane->dist;
else
t1 = DotProduct (plane->normal, center) - plane->dist;
t2 = t1 - radius;
t1 = t1 + radius;
//if the sphere is fully to one side, only walk that side.
if (t1 > 0 && t2 > 0)
{
node = node->children[0];
continue;
}
if (t1 < 0 && t2 < 0)
{
node = node->children[1];
continue;
}
//both sides are within the sphere
Q1BSP_ClustersInSphere_Union(firstleaf, center, radius, node->children[0], out, unionwith);
node = node->children[1];
continue;
}
}
static qbyte *Q1BSP_ClustersInSphere(model_t *mod, const vec3_t center, float radius, pvsbuffer_t *fte_restrict pvsbuffer, const qbyte *unionwith)
{
if (!mod)
Sys_Error ("Mod_PointInLeaf: bad model");
if (!mod->nodes)
return NULL;
if (pvsbuffer->buffersize < mod->pvsbytes)
pvsbuffer->buffer = BZ_Realloc(pvsbuffer->buffer, pvsbuffer->buffersize=mod->pvsbytes);
Q_memset (pvsbuffer->buffer, 0, mod->pvsbytes);
Q1BSP_ClustersInSphere_Union(mod->leafs+1, center, radius, mod->nodes, pvsbuffer->buffer, NULL);//unionwith);
return pvsbuffer->buffer;
}
//returns the leaf number, which is used as a direct bit index into the pvs.
//-1 for invalid
static int Q1BSP_ClusterForPoint (model_t *model, const vec3_t p, int *area)
{
mnode_t *node;
float d;
mplane_t *plane;
if (!model)
{
Sys_Error ("Mod_PointInLeaf: bad model");
}
if (area)
*area = 0; //no areas with q1bsp.
if (!model->nodes)
return -1;
node = model->nodes;
while (1)
{
if (node->contents < 0)
return ((mleaf_t *)node - model->leafs) - 1;
plane = node->plane;
d = DotProduct (p,plane->normal) - plane->dist;
if (d > 0)
node = node->children[0];
else
node = node->children[1];
}
return -1; // never reached
}
/*
PVS type stuff
==============================================================================
Init stuff
*/
void Q1BSP_Init(void)
{
}
//sets up the functions a server needs.
//fills in bspfuncs_t
void Q1BSP_SetModelFuncs(model_t *mod)
{
#ifndef CLIENTONLY
mod->funcs.FatPVS = Q1BSP_FatPVS;
#endif
mod->funcs.EdictInFatPVS = Q1BSP_EdictInFatPVS;
mod->funcs.FindTouchedLeafs = Q1BSP_FindTouchedLeafs;
mod->funcs.LightPointValues = NULL;
mod->funcs.StainNode = NULL;
mod->funcs.MarkLights = NULL;
mod->funcs.ClustersInSphere = Q1BSP_ClustersInSphere;
mod->funcs.ClusterForPoint = Q1BSP_ClusterForPoint;
mod->funcs.ClusterPVS = Q1BSP_ClusterPVS;
// mod->funcs.ClusterPHS = Q1BSP_ClusterPHS;
mod->funcs.NativeTrace = Q1BSP_Trace;
mod->funcs.PointContents = Q1BSP_PointContents;
}
#endif
/*
Init stuff
==============================================================================
BSPX Stuff
*/
typedef struct {
char lumpname[24]; // up to 23 chars, zero-padded
int fileofs; // from file start
int filelen;
} bspx_lump_t;
struct bspx_header_s {
char id[4]; // 'BSPX'
int numlumps;
bspx_lump_t lumps[1];
};
//supported lumps:
//RGBLIGHTING (.lit)
//LIGHTING_E5BGR9 (hdr lit)
//LIGHTINGDIR (.lux)
//LMSHIFT (lightmap scaling)
//LMOFFSET (lightmap scaling)
//LMSTYLE (lightmap scaling)
//VERTEXNORMALS (smooth specular)
//BRUSHLIST (no hull size issues)
//ENVMAP (cubemaps)
//SURFENVMAP (cubemaps)
void *BSPX_FindLump(bspx_header_t *bspxheader, void *mod_base, char *lumpname, int *lumpsize)
{
int i;
*lumpsize = 0;
if (!bspxheader)
return NULL;
for (i = 0; i < bspxheader->numlumps; i++)
{
if (!strncmp(bspxheader->lumps[i].lumpname, lumpname, 24))
{
*lumpsize = bspxheader->lumps[i].filelen;
return (char*)mod_base + bspxheader->lumps[i].fileofs;
}
}
return NULL;
}
bspx_header_t *BSPX_Setup(model_t *mod, char *filebase, unsigned int filelen, lump_t *lumps, int numlumps)
{
int i;
int offs = 0;
bspx_header_t *h;
for (i = 0; i < numlumps; i++, lumps++)
{
if (offs < lumps->fileofs + lumps->filelen)
offs = lumps->fileofs + lumps->filelen;
}
offs = (offs + 3) & ~3;
if (offs + sizeof(*h) > filelen)
return NULL; /*no space for it*/
h = (bspx_header_t*)(filebase + offs);
i = LittleLong(h->numlumps);
/*verify the header*/
if (*(int*)h->id != (('B'<<0)|('S'<<8)|('P'<<16)|('X'<<24)) ||
i < 0 ||
offs + sizeof(*h) + sizeof(h->lumps[0])*(i-1) > filelen)
return NULL;
h->numlumps = i;
while(i-->0)
{
h->lumps[i].fileofs = LittleLong(h->lumps[i].fileofs);
h->lumps[i].filelen = LittleLong(h->lumps[i].filelen);
if (h->lumps[i].fileofs + h->lumps[i].filelen > filelen)
return NULL;
}
return h;
}
#ifdef SERVERONLY
void BSPX_LoadEnvmaps(model_t *mod, bspx_header_t *bspx, void *mod_base)
{
}
#else
/*
void *SCR_ScreenShot_Capture(int fbwidth, int fbheight, int *stride, enum uploadfmt *fmt);
void BSPX_RenderEnvmaps(model_t *mod)
{
int c, i;
void *buffer;
int stride, cubesize;
uploadfmt_t fmt;
char filename[MAX_QPATH];
char olddrawviewmodel[64]; //hack, so we can set r_drawviewmodel to 0 so that it doesn't appear in screenshots even if the csqc is generating new data.
vec3_t oldangles;
const struct
{
vec3_t angle;
const char *postfix;
qboolean verticalflip;
qboolean horizontalflip;
} sides[] =
{
{{0, 0, 90}, "_px", true},
{{0, 180, -90}, "_nx", true},
{{0, 90, 0}, "_py", true}, //upside down
{{0, 270, 0}, "_ny", false, true},
{{-90, 0, 90}, "_pz", true},
{{90, 0, 90}, "_nz", true},
};
char base[MAX_QPATH];
COM_FileBase(cl.worldmodel->name, base, sizeof(base));
r_refdef.stereomethod = STEREO_OFF;
Q_strncpyz(olddrawviewmodel, r_drawviewmodel.string, sizeof(olddrawviewmodel));
Cvar_Set(&r_drawviewmodel, "0");
VectorCopy(cl.playerview->viewangles, oldangles);
for (c = 0; c < mod->numenvmaps; c++)
{
cubesize = mod->envmaps[c].cubesize;
if (cubesize < 1)
cubesize = 32;
VectorCopy(mod->envmaps[c].origin, r_refdef.vieworg);
for (i = 0; i < 6; i++)
{
Q_snprintfz(filename, sizeof(filename), "%s/%i_%i_%i%s.tga", base, (int)mod->envmaps[c].origin[0], (int)mod->envmaps[c].origin[1], (int)mod->envmaps[c].origin[2], sides[i].postfix);
VectorCopy(sides[i].angle, cl.playerview->simangles);
VectorCopy(cl.playerview->simangles, cl.playerview->viewangles);
buffer = SCR_ScreenShot_Capture(cubesize, cubesize, &stride, &fmt);
if (buffer)
{
char sysname[1024];
if (sides[i].horizontalflip)
{
int y, x, p;
int pxsize;
char *bad = buffer;
char *in = buffer, *out;
switch(fmt)
{
case TF_RGBA32:
case TF_BGRA32:
case TF_RGBX32:
case TF_BGRX32:
pxsize = 4;
break;
case TF_RGB24:
case TF_BGR24:
pxsize = 3;
break;
case PTI_RGBA16F:
pxsize = 8;
break;
case PTI_RGBA32F:
pxsize = 16;
break;
default: //erk!
pxsize = 1;
break;
}
buffer = out = BZ_Malloc(cubesize*cubesize*pxsize);
for (y = 0; y < cubesize; y++, in += abs(stride), out += cubesize*pxsize)
{
for (x = 0; x < cubesize*pxsize; x+=pxsize)
{
for (p = 0; p < pxsize; p++)
out[x+p] = in[(cubesize-1)*pxsize-x+p];
}
}
BZ_Free(bad);
if (stride < 0)
stride = -cubesize*pxsize;
else
stride = cubesize*pxsize;
}
if (sides[i].verticalflip)
stride = -stride;
if (SCR_ScreenShot(filename, FS_GAMEONLY, &buffer, 1, stride, cubesize, cubesize, fmt))
{
FS_NativePath(filename, FS_GAMEONLY, sysname, sizeof(sysname));
Con_Printf ("Wrote %s\n", sysname);
}
else
{
FS_NativePath(filename, FS_GAMEONLY, sysname, sizeof(sysname));
Con_Printf ("Failed to write %s\n", sysname);
}
BZ_Free(buffer);
}
}
}
Cvar_Set(&r_drawviewmodel, olddrawviewmodel);
VectorCopy(oldangles, cl.playerview->viewangles);
}
*/
void BSPX_LoadEnvmaps(model_t *mod, bspx_header_t *bspx, void *mod_base)
{
unsigned int *envidx, idx;
int i;
char base[MAX_QPATH];
char imagename[MAX_QPATH];
menvmap_t *out;
int count;
denvmap_t *in = BSPX_FindLump(bspx, mod_base, "ENVMAP", &count);
mod->envmaps = NULL;
mod->numenvmaps = 0;
if (!mod_loadsurfenvmaps.ival)
return;
if (count%sizeof(*in))
return; //erk
count /= sizeof(*in);
if (!count)
return;
out = ZG_Malloc(&mod->memgroup, sizeof(*out)*count);
mod->envmaps = out;
mod->numenvmaps = count;
COM_FileBase(mod->name, base, sizeof(base));
for (i = 0; i < count; i++)
{
out[i].origin[0] = LittleFloat(in[i].origin[0]);
out[i].origin[1] = LittleFloat(in[i].origin[1]);
out[i].origin[2] = LittleFloat(in[i].origin[2]);
out[i].cubesize = LittleLong(in[i].cubesize);
Q_snprintfz(imagename, sizeof(imagename), "textures/env/%s_%i_%i_%i", base, (int)mod->envmaps[i].origin[0], (int)mod->envmaps[i].origin[1], (int)mod->envmaps[i].origin[2]);
out[i].image = Image_GetTexture(imagename, NULL, IF_TEXTYPE_CUBE|IF_NOREPLACE, NULL, NULL, out[i].cubesize, out[i].cubesize, PTI_INVALID);
}
//now update surface lists.
envidx = BSPX_FindLump(bspx, mod_base, "SURFENVMAP", &i);
if (i/sizeof(*envidx) == mod->numsurfaces)
{
for (i = 0; i < mod->numsurfaces; i++)
{
idx = LittleLong(envidx[i]);
if (idx < (unsigned int)count)
mod->surfaces[i].envmap = out[idx].image;
}
}
}
struct bspxrw
{
const char *fname;
char *origfile;
qofs_t origsize;
int lumpofs;
fromgame_t fg;
size_t corelumps;
size_t totallumps;
struct
{
char lumpname[24]; // up to 23 chars, zero-padded
void *data; // from file start
qofs_t filelen;
} *lumps;
};
void Mod_BSPXRW_Free(struct bspxrw *ctx)
{
FS_FreeFile(ctx->origfile);
Z_Free(ctx->lumps);
ctx->corelumps = ctx->totallumps = 0;
ctx->origfile = NULL;
}
void Mod_BSPXRW_Write(struct bspxrw *ctx)
{
#if 1
vfsfile_t *f = FS_OpenVFS(ctx->fname, "wb", FS_GAMEONLY);
if (f)
{
qofs_t bspxofs;
size_t i, j;
int pad, paddata = 0;
int nxlumps = ctx->totallumps-ctx->corelumps;
lump_t *lumps = alloca(sizeof(*lumps)*ctx->corelumps);
bspx_lump_t *xlumps = alloca(sizeof(*xlumps)*(ctx->totallumps-ctx->corelumps));
//bsp header info
VFS_WRITE(f, ctx->origfile, ctx->lumpofs);
VFS_WRITE(f, lumps, sizeof(lumps[0])*ctx->corelumps); //placeholder
//orig lumps
for (i = 0; i < ctx->corelumps; i++)
{
lumps[i].fileofs = VFS_TELL(f);
lumps[i].filelen = ctx->lumps[i].filelen;
VFS_WRITE(f, ctx->lumps[i].data, ctx->lumps[i].filelen);
//ALL lumps must be 4-aligned, so pad if needed.
pad = ((ctx->lumps[i].filelen+3)&~3)-ctx->lumps[i].filelen;
VFS_WRITE(f, &paddata, pad);
}
//bspx header
VFS_WRITE(f, "BSPX", 4);
VFS_WRITE(f, &nxlumps, sizeof(nxlumps));
bspxofs = VFS_TELL(f);
VFS_WRITE(f, xlumps, sizeof(xlumps[0])*(ctx->totallumps-ctx->corelumps)); //placeholder
//bspx data
for (i = 0; i < nxlumps; i++)
{
j = ctx->corelumps+i;
xlumps[i].fileofs = VFS_TELL(f);
xlumps[i].filelen = ctx->lumps[j].filelen;
memcpy(xlumps[i].lumpname, ctx->lumps[j].lumpname, sizeof(xlumps[i].lumpname));
VFS_WRITE(f, ctx->lumps[j].data, ctx->lumps[j].filelen);
//ALL lumps must be 4-aligned, so pad if needed.
pad = ((ctx->lumps[j].filelen+3)&~3)-ctx->lumps[j].filelen;
VFS_WRITE(f, &paddata, pad);
}
//now rewrite both sets of offsets.
VFS_SEEK(f, ctx->lumpofs);
VFS_WRITE(f, lumps, sizeof(lumps[0])*ctx->corelumps);
VFS_SEEK(f, bspxofs);
VFS_WRITE(f, xlumps, sizeof(xlumps[0])*(ctx->totallumps-ctx->corelumps));
VFS_CLOSE(f);
}
#endif
Mod_BSPXRW_Free(ctx);
}
void Mod_BSPXRW_SetLump(struct bspxrw *ctx, const char *lumpname, void *data, size_t datasize)
{
int i;
for (i = 0; i < ctx->totallumps; i++)
{
if (!strcmp(ctx->lumps[i].lumpname, lumpname))
{ //replace the existing lump
ctx->lumps[i].data = data;
ctx->lumps[i].filelen = datasize;
return;
}
}
Z_ReallocElements((void**)&ctx->lumps, &ctx->totallumps, ctx->totallumps+1, sizeof(*ctx->lumps));
Q_strncpyz(ctx->lumps[i].lumpname, lumpname, sizeof(ctx->lumps[i].lumpname));
ctx->lumps[i].data = data;
ctx->lumps[i].filelen = datasize;
}
qboolean Mod_BSPXRW_Read(struct bspxrw *ctx, const char *fname)
{
int i;
lump_t *l;
const char **corelumpnames = NULL;
bspx_header_t *bspxheader;
#ifdef Q3BSPS
static const char *q3corelumpnames[Q3LUMPS_TOTAL] = {"entities","shaders","planes","nodes","leafs","leafsurfs","leafbrushes","submodels","brushes","brushsides","verts","indexes","fogs","surfaces","lightmaps","lightgrid","visibility"
#ifdef RFBSPS
,"lightgrididx"
#endif
};
#endif
ctx->fname = fname;
ctx->origfile = FS_MallocFile(ctx->fname, FS_GAME, &ctx->origsize);
if (!ctx->origfile)
return false;
ctx->lumps = 0;
ctx->totallumps = 0;
i = LittleLong(*(int*)ctx->origfile);
switch(i)
{
case 29:
case 30:
ctx->fg = ((i==30)?fg_halflife:fg_quake);
ctx->lumpofs = 4;
ctx->corelumps = 0;
break;
case IDBSPHEADER:
i = LittleLong(*(int*)(ctx->origfile+4));
ctx->lumpofs = 8;
switch(i)
{
#ifdef Q2BSPS
case BSPVERSION_Q2:
// case BSPVERSION_Q2W:
ctx->fg = fg_quake2;
ctx->corelumps = Q2HEADER_LUMPS;
break;
#endif
#ifdef Q3BSPS
case BSPVERSION_Q3:
case BSPVERSION_RTCW:
ctx->fg = fg_quake3;
ctx->corelumps = 17;
corelumpnames = q3corelumpnames;
break;
#endif
default:
Mod_BSPXRW_Free(ctx);
return false;
}
break;
#ifdef RFBSPS
case ('R'<<0)+('B'<<8)+('S'<<16)+('P'<<24):
case ('F'<<0)+('B'<<8)+('S'<<16)+('P'<<24):
i = LittleLong(*(int*)(ctx->origfile+4));
ctx->lumpofs = 8;
switch(i)
{
case BSPVERSION_RBSP:
ctx->fg = fg_quake3;
ctx->corelumps = 18;
corelumpnames = q3corelumpnames;
break;
default:
Mod_BSPXRW_Free(ctx);
return false;
}
break;
#endif
default:
Mod_BSPXRW_Free(ctx);
return false;
}
l = (lump_t*)(ctx->origfile+ctx->lumpofs);
for (i = 0; i < ctx->corelumps; i++)
{
Z_ReallocElements((void**)&ctx->lumps, &ctx->totallumps, ctx->totallumps+1, sizeof(*ctx->lumps));
ctx->lumps[ctx->totallumps-1].data = ctx->origfile+l[i].fileofs;
ctx->lumps[ctx->totallumps-1].filelen = l[i].filelen;
if (corelumpnames)
Q_snprintfz(ctx->lumps[ctx->totallumps-1].lumpname, sizeof(ctx->lumps[0].lumpname), "%s", corelumpnames[i]);
else
Q_snprintfz(ctx->lumps[ctx->totallumps-1].lumpname, sizeof(ctx->lumps[0].lumpname), "lump%u", i);
}
bspxheader = BSPX_Setup(NULL, ctx->origfile, ctx->origsize, l, ctx->corelumps);
if (bspxheader)
{
for (i = 0; i < bspxheader->numlumps; i++)
{
Z_ReallocElements((void**)&ctx->lumps, &ctx->totallumps, ctx->totallumps+1, sizeof(*ctx->lumps));
ctx->lumps[ctx->totallumps-1].data = ctx->origfile+bspxheader->lumps[i].fileofs;
ctx->lumps[ctx->totallumps-1].filelen = bspxheader->lumps[i].filelen;
memcpy(ctx->lumps[ctx->totallumps-1].lumpname, bspxheader->lumps[i].lumpname, sizeof(ctx->lumps[0].lumpname));
}
}
return true;
}
unsigned int Mod_NearestCubeForSurf(msurface_t *surf, denvmap_t *envmap, size_t nenvmap)
{ //this is slow, yes.
size_t n, v;
unsigned int best = ~0;
float bestdist = FLT_MAX, dist;
vec3_t diff, mins, maxs, mid;
if (surf->mesh && surf->mesh->numvertexes)
{
VectorCopy(surf->mesh->xyz_array[0], mins);
VectorCopy(surf->mesh->xyz_array[0], maxs);
for (v = 1; v < surf->mesh->numvertexes; v++)
AddPointToBounds(surf->mesh->xyz_array[v], mins, maxs);
VectorAvg(mins, maxs, mid);
for (n = 0; n < nenvmap; n++)
{
VectorSubtract(envmap[n].origin, mid, diff);
#if 0
//axial distance
dist = fabs(diff[0]) + fabs(diff[1]) + fabs(diff[2]);
#else
//radial distance (squared)
dist = DotProduct(diff,diff);
#endif
if (bestdist > dist)
{
best = n;
bestdist = dist;
}
}
}
return best;
}
int QDECL envmapsort(const void *av, const void *bv)
{ //sorts cubemaps in order of size, to make texturearrays easier, if ever. The loader can then just make runs.
const denvmap_t *a=av, *b=bv;
if (a->cubesize == b->cubesize)
return 0;
if (a->cubesize > b->cubesize)
return 1;
return -1;
}
void Mod_FindCubemaps_f(void)
{
struct bspxrw bspctx;
if (Mod_BSPXRW_Read(&bspctx, cl.worldmodel->name))
{
const char *entlump = Mod_GetEntitiesString(cl.worldmodel), *lmp;
int nest;
char key[1024];
char value[1024];
qboolean isenvmap;
float size;
vec3_t origin;
denvmap_t *envmap = NULL; //*nenvmap
size_t nenvmap = 0;
unsigned int *envmapidx = NULL; //*numsurfaces
size_t nenvmapidx = 0, i;
//find targetnames, and store their origins so that we can deal with spotlights.
for (lmp = entlump; ;)
{
lmp = COM_Parse(lmp);
if (com_token[0] != '{')
break;
isenvmap = false;
size = 128;
VectorClear(origin);
nest = 1;
while (1)
{
lmp = COM_ParseOut(lmp, key, sizeof(key));
if (!lmp)
break; // error
if (key[0] == '{')
{
nest++;
continue;
}
if (key[0] == '}')
{
nest--;
if (!nest)
break; // end of entity
continue;
}
if (nest!=1)
continue;
if (key[0] == '_')
memmove(key, key+1, strlen(key));
while (key[strlen(key)-1] == ' ') // remove trailing spaces
key[strlen(key)-1] = 0;
lmp = COM_ParseOut(lmp, value, sizeof(value));
if (!lmp)
break; // error
// now that we have the key pair worked out...
if (!strcmp("classname", key) && !strcmp(value, "env_cubemap"))
isenvmap = true;
else if (!strcmp("origin", key))
sscanf(value, "%f %f %f", &origin[0], &origin[1], &origin[2]);
else if (!strcmp("size", key))
sscanf(value, "%f", &size);
}
if (isenvmap)
{
int e = nenvmap;
if (ZF_ReallocElements((void**)&envmap, &nenvmap, nenvmap+1, sizeof(*envmap)))
{
VectorCopy(origin, envmap[e].origin);
envmap[e].cubesize = size;
}
}
}
if (nenvmap)
{
qsort(envmap, nenvmap, sizeof(*envmap), envmapsort); //sort them by size
if (ZF_ReallocElements((void**)&envmapidx, &nenvmapidx, cl.worldmodel->numsurfaces, sizeof(*envmapidx)))
{
for(i = 0; i < cl.worldmodel->numsurfaces; i++)
envmapidx[i] = Mod_NearestCubeForSurf(cl.worldmodel->surfaces+i, envmap, nenvmap);
}
Mod_BSPXRW_SetLump(&bspctx, "ENVMAP", envmap, nenvmap*sizeof(*envmap));
Mod_BSPXRW_SetLump(&bspctx, "SURFENVMAP", envmapidx, cl.worldmodel->numsurfaces*sizeof(*envmapidx));
Mod_BSPXRW_Write(&bspctx);
}
else
{
Con_Printf("No cubemaps found on map\n");
Mod_BSPXRW_Free(&bspctx);
}
Z_Free(envmapidx);
Z_Free(envmap);
}
}
void Mod_Realign_f(void)
{
struct bspxrw bspctx;
if (Mod_BSPXRW_Read(&bspctx, cl.worldmodel->name))
Mod_BSPXRW_Write(&bspctx);
}
void Mod_BSPX_List_f(void)
{
int i;
struct bspxrw ctx;
char *fname = Cmd_Argv(1);
if (!*fname && cl.worldmodel)
fname = cl.worldmodel->name;
if (Mod_BSPXRW_Read(&ctx, fname))
{
for (i = 0; i < ctx.corelumps; i++)
{
Con_Printf("%s: %u\n", ctx.lumps[i].lumpname, (unsigned int)ctx.lumps[i].filelen);
}
for ( ; i < ctx.totallumps; i++)
{
Con_Printf("%s: %u\n", ctx.lumps[i].lumpname, (unsigned int)ctx.lumps[i].filelen);
}
Mod_BSPXRW_Free(&ctx);
}
}
void Mod_BSPX_Strip_f(void)
{
int i;
struct bspxrw ctx;
qboolean found = false;
if (Cmd_Argc() != 3)
Con_Printf("%s FILENAME NAME: removes an extended lump from the named bsp file\n", Cmd_Argv(0));
else if (Mod_BSPXRW_Read(&ctx, Cmd_Argv(1)))
{
for (i = ctx.corelumps; i < ctx.totallumps;)
{
if (!Q_strcasecmp(ctx.lumps[i].lumpname, Cmd_Argv(2)))
{
found = true;
memmove(&ctx.lumps[i], &ctx.lumps[i+1], sizeof(ctx.lumps[0])*(ctx.totallumps-(i+1)));
ctx.totallumps--;
}
else
i++;
}
if (found)
Mod_BSPXRW_Write(&ctx);
else
Mod_BSPXRW_Free(&ctx);
}
}
image_t *Mod_CubemapForOrigin(model_t *wmodel, vec3_t org)
{
int i;
menvmap_t *e;
float bestdist = FLT_MAX, dist;
image_t *ret = NULL;
vec3_t move;
if (!wmodel || wmodel->loadstate != MLS_LOADED)
return NULL;
for ( i=0 , e=wmodel->envmaps ; i<wmodel->numenvmaps ; i++, e++)
{
VectorSubtract(org, e->origin, move);
dist = DotProduct(move,move);
if (bestdist > dist)
{
bestdist = dist;
ret = e->image;
}
}
return ret;
}
#endif