#include "quakedef.h" #ifndef SERVERONLY #include "glquake.h" #endif #include "com_mesh.h" #include "com_bih.h" #define MAX_Q3MAP_INDICES 0x8000000 //just a sanity limit #define MAX_Q3MAP_VERTEXES 0x800000 //just a sanity limit //#define MAX_CM_PATCH_VERTS (4096) //#define MAX_CM_FACES (MAX_Q2MAP_FACES) #ifdef FTE_TARGET_WEB #define MAX_CM_PATCHES (0x1000) //fixme #else #define MAX_CM_PATCHES (0x10000) //fixme #endif //#define MAX_CM_LEAFFACES (MAX_Q2MAP_LEAFFACES) #define MAX_CM_AREAS MAX_Q2MAP_AREAS //#define Q3SURF_NODAMAGE 0x00000001 //#define Q3SURF_SLICK 0x00000002 //#define Q3SURF_SKY 0x00000004 //#define Q3SURF_LADDER 0x00000008 //#define Q3SURF_NOIMPACT 0x00000010 //#define Q3SURF_NOMARKS 0x00000020 //#define Q3SURF_FLESH 0x00000040 #define Q3SURF_NODRAW 0x00000080 // don't generate a drawsurface at all //#define Q3SURF_HINT 0x00000100 #define Q3SURF_SKIP 0x00000200 // completely ignore, allowing non-closed brushes //#define Q3SURF_NOLIGHTMAP 0x00000400 //#define Q3SURF_POINTLIGHT 0x00000800 //#define Q3SURF_METALSTEPS 0x00001000 //#define Q3SURF_NOSTEPS 0x00002000 #define Q3SURF_NONSOLID 0x00004000 // don't collide against curves with this set //#define Q3SURF_LIGHTFILTER 0x00008000 //#define Q3SURF_ALPHASHADOW 0x00010000 //#define Q3SURF_NODLIGHT 0x00020000 //#define Q3SURF_DUST 0x00040000 cvar_t q3bsp_surf_meshcollision_flag = CVARD("q3bsp_surf_meshcollision_flag", "0x80000000", "The surfaceparm flag(s) that enables q3bsp trisoup collision"); cvar_t q3bsp_surf_meshcollision_force = CVARD("q3bsp_surf_meshcollision_force", "0", "Force mesh-based collisions on all q3bsp trisoup surfaces."); cvar_t q3bsp_mergeq3lightmaps = CVARD("q3bsp_mergelightmaps", "1", "Specifies whether to merge lightmaps into atlases in order to boost performance. Unfortunately this breaks tcgen on lightmap passes - if you care, set this to 0."); cvar_t q3bsp_ignorestyles = CVARD("q3bsp_ignorestyles", "0", "Ignores multiple lightstyles in Raven's q3bsp variant(and derivatives) for better batch/rendering performance."); cvar_t q3bsp_bihtraces = CVARFD("_q3bsp_bihtraces", /*FIXME: generate BIH leafs more carefully*/"0", CVAR_RENDERERLATCH, "Uses runtime-generated bih collision culling for faster traces."); #if Q3SURF_NODRAW != TI_NODRAW #error "nodraw isn't constant" #endif extern cvar_t r_shadow_bumpscale_basetexture; //these are in model.c (or gl_model.c) qboolean Mod_LoadVertexes (model_t *loadmodel, qbyte *mod_base, lump_t *l); qboolean Mod_LoadVertexNormals (model_t *loadmodel, bspx_header_t *bspx, qbyte *mod_base, lump_t *l); qboolean Mod_LoadEdges (model_t *loadmodel, qbyte *mod_base, lump_t *l, qboolean lm); qboolean Mod_LoadMarksurfaces (model_t *loadmodel, qbyte *mod_base, lump_t *l, qboolean lm); qboolean Mod_LoadSurfedges (model_t *loadmodel, qbyte *mod_base, lump_t *l); void Mod_LoadEntities (model_t *loadmodel, qbyte *mod_base, lump_t *l); extern void BuildLightMapGammaTable (float g, float c); #if defined(Q2BSPS) || defined(Q3BSPS) static qboolean CM_NativeTrace(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 contents, trace_t *trace); static unsigned int CM_NativeContents(struct model_s *model, int hulloverride, const framestate_t *framestate, const vec3_t axis[3], const vec3_t p, const vec3_t mins, const vec3_t maxs); static unsigned int Q2BSP_PointContents(model_t *mod, const vec3_t axis[3], const vec3_t p); static int CM_PointCluster (model_t *mod, const vec3_t p, int *area); struct cminfo_s; #endif float RadiusFromBounds (const vec3_t mins, const vec3_t maxs) { int i; vec3_t corner; for (i=0 ; i<3 ; i++) { corner[i] = fabs(mins[i]) > fabs(maxs[i]) ? fabs(mins[i]) : fabs(maxs[i]); } return Length (corner); } void CalcSurfaceExtents (model_t *mod, msurface_t *s) { float mins[2], maxs[2], val; int i,j, e; mvertex_t *v; mtexinfo_t *tex; int bmins[2], bmaxs[2]; int idx; mins[0] = mins[1] = 999999; maxs[0] = maxs[1] = -999999; tex = s->texinfo; for (i=0 ; inumedges ; i++) { e = mod->surfedges[s->firstedge+i]; idx = e < 0; if (idx) e = -e; if (e < 0 || e >= mod->numedges) v = &mod->vertexes[0]; else v = &mod->vertexes[mod->edges[e].v[idx]]; for (j=0 ; j<2 ; j++) { //doubles should replicate win32/x87 compiler 80-bit precision better. We have to hope that win64 compilers use the same precision. val = DotProduct_Double(v->position, tex->vecs[j]) + tex->vecs[j][3]; if (val < mins[j]) mins[j] = val; if (val > maxs[j]) maxs[j] = val; } } for (i=0 ; i<2 ; i++) { bmins[i] = floor(mins[i]/(1<lmshift)); bmaxs[i] = ceil(maxs[i]/(1<lmshift)); s->texturemins[i] = bmins[i] << s->lmshift; s->extents[i] = (bmaxs[i] - bmins[i]); // if ( !(tex->flags & TEX_SPECIAL) && s->extents[i] > 17 ) //vanilla used 16(+1), glquake used 17(+1). FTE uses 255(+1), but we omit lightmapping instead of crashing if its larger than our limit, so we omit the check here. different engines use different limits here, many of them make no sense. // Con_Printf ("Bad surface extents (texture %s, more than %i lightmap samples)\n", s->texinfo->texture->name, s->extents[i]); s->extents[i] <<= s->lmshift; } } void AddPointToBounds (const vec3_t v, vec3_t mins, vec3_t maxs) { int i; vec_t val; for (i=0 ; i<3 ; i++) { val = v[i]; if (val < mins[i]) mins[i] = val; if (val > maxs[i]) maxs[i] = val; } } void ClearBounds (vec3_t mins, vec3_t maxs) { mins[0] = mins[1] = mins[2] = FLT_MAX; maxs[0] = maxs[1] = maxs[2] = -FLT_MAX; } void Mod_SortShaders(model_t *mod) { //surely this isn't still needed? texture_t *textemp; int i, j; //sort loadmodel->textures for (i = 0; i < mod->numtextures; i++) { for (j = i+1; j < mod->numtextures; j++) { if ((mod->textures[i]->shader && mod->textures[j]->shader) && (mod->textures[j]->shader->sort < mod->textures[i]->shader->sort)) { textemp = mod->textures[j]; mod->textures[j] = mod->textures[i]; mod->textures[i] = textemp; } } } } #if defined(Q2BSPS) || defined(Q3BSPS) qbyte *ReadPCXPalette(qbyte *buf, int len, qbyte *out); #ifdef SERVERONLY #define Host_Error SV_Error #endif extern qbyte *mod_base; #define capsuledist(dist,plane,mins,maxs) \ case shape_iscapsule: \ dist = DotProduct(trace_up, plane->normal); \ dist = dist*(trace_capsulesize[(dist<0)?1:2]) - trace_capsulesize[0]; \ dist = plane->dist - dist; \ break; unsigned char d_q28to24table[1024]; /* typedef struct q2csurface_s { char name[16]; int flags; int value; } q2csurface_t; */ typedef struct { char shader[64]; int brushNum; int visibleSide; // the brush side that ray tests need to clip against (-1 == none) } dfog_t; #ifdef Q2BSPS typedef struct { int numareaportals; int firstareaportal; } q2carea_t; #endif #ifdef Q3BSPS typedef struct { int numareaportals[MAX_CM_AREAS]; } q3carea_t; #endif typedef struct { int floodnum; // if two areas have equal floodnums, they are connected int floodvalid; // flags the area as having been visited (sequence numbers matching prv->floodvalid) } careaflood_t; typedef struct { int facetype; int numverts; int firstvert; int shadernum; union { struct { unsigned short cp[2]; unsigned short fixedres[2]; } patch; struct { int firstindex; int numindicies; } soup; }; } q3cface_t; typedef struct cmodel_s { vec3_t mins, maxs; vec3_t origin; // for sounds or lights mnode_t *headnode; mleaf_t *headleaf; int numsurfaces; int firstsurface; int firstbrush; //q3 submodels are considered small enough that you will never need to walk any sort of tree. int num_brushes;//the brushes are checked instead. //these things are generated at load time. int firstpatch; int num_patches; int firstcmesh; int num_cmeshes; } cmodel_t; /*used to trace*/ static int checkcount; typedef struct cminfo_s { int numbrushsides; q2cbrushside_t *brushsides; q2mapsurface_t *surfaces; int numleafbrushes; q2cbrush_t **leafbrushes; int numcmodels; cmodel_t *cmodels; int numbrushes; q2cbrush_t *brushes; int numvisibility; q2dvis_t *q2vis; q3dvis_t *q3pvs; q3dvis_t *q3phs; qbyte *phscalced; int numareas; int floodvalid; careaflood_t areaflood[MAX_CM_AREAS]; #ifdef Q3BSPS //q3's areas are simple bidirectional area1/area2 pairs. refcounted (so two areas can have two doors/openings) q3carea_t q3areas[MAX_CM_AREAS]; #endif #ifdef Q2BSPS //q2's areas have a list of portals that open into other areas. q2carea_t *q2areas; //indexes into q2areaportals for flooding size_t numq2areaportals; q2dareaportal_t *q2areaportals; //and this is the state that is actually changed. booleans. qbyte q2portalopen[MAX_Q2MAP_AREAPORTALS]; //memset will work if it's a qbyte, really it should be a qboolean #endif //list of mesh surfaces within the leaf q3cmesh_t cmeshes[MAX_CM_PATCHES]; int numcmeshes; int *leafcmeshes; int numleafcmeshes; int maxleafcmeshes; //FIXME: remove the below //(deprecated) patch collisions q3cpatch_t patches[MAX_CM_PATCHES]; int numpatches; int *leafpatches; int numleafpatches; int maxleafpatches; //FIXME: remove the above qboolean mapisq3; #ifdef Q3BSPS //this is for loading stuff. it used to be globals, but we have threads now. and multiple q3bsps at the same time is a problem. int numvertexes; vecV_t *verts; //3points vec2_t *vertstmexcoords; vec2_t *vertlstmexcoords[MAXRLIGHTMAPS]; vec4_t *colors4f_array[MAXRLIGHTMAPS]; vec3_t *normals_array; //vec3_t *map_svector_array; //vec3_t *map_tvector_array; index_t *surfindexes; //int map_numsurfindexes; q3cface_t *faces; int numfaces; #endif // struct bihnode_s *bihnodes; } cminfo_t; static q2mapsurface_t nullsurface; cvar_t map_noareas = CVAR("map_noareas", "0"); //1 for lack of mod support. cvar_t map_noCurves = CVARF("map_noCurves", "0", CVAR_CHEAT); cvar_t map_autoopenportals = CVARD("map_autoopenportals", "0", "When set to 1, force-opens all area portals. Normally these start closed and are opened by doors when they move, but this requires the gamecode to signal this."); //1 for lack of mod support. cvar_t r_subdivisions = CVAR("r_subdivisions", "2"); static int CM_NumInlineModels (model_t *model); static cmodel_t *CM_InlineModel (model_t *model, char *name); void CM_InitBoxHull (void); static void CM_FinalizeBrush(q2cbrush_t *brush); static void FloodAreaConnections (cminfo_t *prv); qboolean BoundsIntersect (const vec3_t mins1, const vec3_t maxs1, const vec3_t mins2, const vec3_t maxs2) { return (mins1[0] <= maxs2[0] && mins1[1] <= maxs2[1] && mins1[2] <= maxs2[2] && maxs1[0] >= mins2[0] && maxs1[1] >= mins2[1] && maxs1[2] >= mins2[2]); } #ifdef Q3BSPS static int PlaneTypeForNormal ( vec3_t normal ) { vec_t ax, ay, az; // NOTE: should these have an epsilon around 1.0? if ( normal[0] >= 1.0) return PLANE_X; if ( normal[1] >= 1.0 ) return PLANE_Y; if ( normal[2] >= 1.0 ) return PLANE_Z; ax = fabs( normal[0] ); ay = fabs( normal[1] ); az = fabs( normal[2] ); if ( ax >= ay && ax >= az ) return PLANE_ANYX; if ( ay >= ax && ay >= az ) return PLANE_ANYY; return PLANE_ANYZ; } void CategorizePlane ( mplane_t *plane ) { int i; plane->signbits = 0; plane->type = PLANE_ANYZ; for (i = 0; i < 3; i++) { if (plane->normal[i] < 0) plane->signbits |= 1<normal[i] == 1.0f) plane->type = i; } plane->type = PlaneTypeForNormal(plane->normal); } static void PlaneFromPoints ( vec3_t verts[3], mplane_t *plane ) { vec3_t v1, v2; VectorSubtract( verts[1], verts[0], v1 ); VectorSubtract( verts[2], verts[0], v2 ); CrossProduct( v2, v1, plane->normal ); VectorNormalize( plane->normal ); plane->dist = DotProduct( verts[0], plane->normal ); } /* =============== Patch_FlatnessTest =============== */ static int Patch_FlatnessTest( float maxflat2, const float *point0, const float *point1, const float *point2 ) { float d; int ft0, ft1; vec3_t t, n; vec3_t v1, v2, v3; VectorSubtract( point2, point0, n ); if( !VectorNormalize( n ) ) return 0; VectorSubtract( point1, point0, t ); d = -DotProduct( t, n ); VectorMA( t, d, n, t ); if( DotProduct( t, t ) < maxflat2 ) return 0; VectorAvg( point1, point0, v1 ); VectorAvg( point2, point1, v2 ); VectorAvg( v1, v2, v3 ); ft0 = Patch_FlatnessTest( maxflat2, point0, v1, v3 ); ft1 = Patch_FlatnessTest( maxflat2, v3, v2, point2 ); return 1 + (int)( floor( max( ft0, ft1 ) ) + 0.5f ); } /* =============== Patch_GetFlatness =============== */ static void Patch_GetFlatness( float maxflat, const float *points, int comp, const unsigned short *patch_cp, int *flat ) { int i, p, u, v; float maxflat2 = maxflat * maxflat; flat[0] = flat[1] = 0; for( v = 0; v < patch_cp[1] - 1; v += 2 ) { for( u = 0; u < patch_cp[0] - 1; u += 2 ) { p = v * patch_cp[0] + u; i = Patch_FlatnessTest( maxflat2, &points[p*comp], &points[( p+1 )*comp], &points[( p+2 )*comp] ); flat[0] = max( flat[0], i ); i = Patch_FlatnessTest( maxflat2, &points[( p+patch_cp[0] )*comp], &points[( p+patch_cp[0]+1 )*comp], &points[( p+patch_cp[0]+2 )*comp] ); flat[0] = max( flat[0], i ); i = Patch_FlatnessTest( maxflat2, &points[( p+2*patch_cp[0] )*comp], &points[( p+2*patch_cp[0]+1 )*comp], &points[( p+2*patch_cp[0]+2 )*comp] ); flat[0] = max( flat[0], i ); i = Patch_FlatnessTest( maxflat2, &points[p*comp], &points[( p+patch_cp[0] )*comp], &points[( p+2*patch_cp[0] )*comp] ); flat[1] = max( flat[1], i ); i = Patch_FlatnessTest( maxflat2, &points[( p+1 )*comp], &points[( p+patch_cp[0]+1 )*comp], &points[( p+2*patch_cp[0]+1 )*comp] ); flat[1] = max( flat[1], i ); i = Patch_FlatnessTest( maxflat2, &points[( p+2 )*comp], &points[( p+patch_cp[0]+2 )*comp], &points[( p+2*patch_cp[0]+2 )*comp] ); flat[1] = max( flat[1], i ); } } } /* =============== Patch_Evaluate_QuadricBezier =============== */ static void Patch_Evaluate_QuadricBezier( float t, const vec_t *point0, const vec_t *point1, const vec_t *point2, vec_t *out, int comp ) { int i; vec_t qt = t * t; vec_t dt = 2.0f * t, tt, tt2; tt = 1.0f - dt + qt; tt2 = dt - 2.0f * qt; for( i = 0; i < comp; i++ ) out[i] = point0[i] * tt + point1[i] * tt2 + point2[i] * qt; } /* =============== Patch_Evaluate =============== */ static void Patch_Evaluate( const vec_t *p, const unsigned short *numcp, const int *tess, vec_t *dest, int comp ) { int num_patches[2], num_tess[2]; int index[3], dstpitch, i, u, v, x, y; float s, t, step[2]; vec_t *tvec, *tvec2; const vec_t *pv[3][3]; vec4_t v1, v2, v3; if (!tess[0] || !tess[1]) { //not really a patch for( i = 0; i < comp*numcp[1]*numcp[0]; i++ ) dest[i] = p[i]; return; } num_patches[0] = numcp[0] / 2; num_patches[1] = numcp[1] / 2; dstpitch = ( num_patches[0] * tess[0] + 1 ) * comp; step[0] = 1.0f / (float)tess[0]; step[1] = 1.0f / (float)tess[1]; for( v = 0; v < num_patches[1]; v++ ) { // last patch has one more row if( v < num_patches[1] - 1 ) num_tess[1] = tess[1]; else num_tess[1] = tess[1] + 1; for( u = 0; u < num_patches[0]; u++ ) { // last patch has one more column if( u < num_patches[0] - 1 ) num_tess[0] = tess[0]; else num_tess[0] = tess[0] + 1; index[0] = ( v * numcp[0] + u ) * 2; index[1] = index[0] + numcp[0]; index[2] = index[1] + numcp[0]; // current 3x3 patch control points for( i = 0; i < 3; i++ ) { pv[i][0] = &p[( index[0]+i ) * comp]; pv[i][1] = &p[( index[1]+i ) * comp]; pv[i][2] = &p[( index[2]+i ) * comp]; } tvec = dest + v * tess[1] * dstpitch + u * tess[0] * comp; for( y = 0, t = 0.0f; y < num_tess[1]; y++, t += step[1], tvec += dstpitch ) { Patch_Evaluate_QuadricBezier( t, pv[0][0], pv[0][1], pv[0][2], v1, comp ); Patch_Evaluate_QuadricBezier( t, pv[1][0], pv[1][1], pv[1][2], v2, comp ); Patch_Evaluate_QuadricBezier( t, pv[2][0], pv[2][1], pv[2][2], v3, comp ); for( x = 0, tvec2 = tvec, s = 0.0f; x < num_tess[0]; x++, s += step[0], tvec2 += comp ) Patch_Evaluate_QuadricBezier( s, v1, v2, v3, tvec2, comp ); } } } } #define PLANE_NORMAL_EPSILON 0.00001 #define PLANE_DIST_EPSILON 0.01 static qboolean ComparePlanes( const vec3_t p1normal, vec_t p1dist, const vec3_t p2normal, vec_t p2dist ) { if( fabs( p1normal[0] - p2normal[0] ) < PLANE_NORMAL_EPSILON && fabs( p1normal[1] - p2normal[1] ) < PLANE_NORMAL_EPSILON && fabs( p1normal[2] - p2normal[2] ) < PLANE_NORMAL_EPSILON && fabs( p1dist - p2dist ) < PLANE_DIST_EPSILON ) return true; return false; } static void SnapVector( vec3_t normal ) { int i; for( i = 0; i < 3; i++ ) { if( fabs( normal[i] - 1 ) < PLANE_NORMAL_EPSILON ) { VectorClear( normal ); normal[i] = 1; break; } if( fabs( normal[i] + 1 ) < PLANE_NORMAL_EPSILON ) { VectorClear( normal ); normal[i] = -1; break; } } } #define Q_rint( x ) ( ( x ) < 0 ? ( (int)( ( x )-0.5f ) ) : ( (int)( ( x )+0.5f ) ) ) static void SnapPlane( vec3_t normal, vec_t *dist ) { SnapVector( normal ); if( fabs( *dist - Q_rint( *dist ) ) < PLANE_DIST_EPSILON ) { *dist = Q_rint( *dist ); } } /* =============================================================================== PATCH LOADING =============================================================================== */ #define MAX_FACET_PLANES 32 #define cm_subdivlevel 15 /* * CM_CreateFacetFromPoints */ static int CM_CreateFacetFromPoints(q2cbrush_t *facet, vec3_t *verts, int numverts, q2mapsurface_t *shaderref, mplane_t *brushplanes ) { int i, j; int axis, dir; vec3_t normal; float d, dist; mplane_t mainplane; vec3_t vec, vec2; int numbrushplanes; // set default values for brush facet->numsides = 0; facet->brushside = NULL; facet->contents = shaderref->c.value; // calculate plane for this triangle PlaneFromPoints( verts, &mainplane ); if( ComparePlanes( mainplane.normal, mainplane.dist, vec3_origin, 0 ) ) return 0; // test a quad case if( numverts > 3 ) { d = DotProduct( verts[3], mainplane.normal ) - mainplane.dist; if( d < -0.1 || d > 0.1 ) return 0; if( 0 ) { vec3_t v[3]; mplane_t plane; // try different combinations of planes for( i = 1; i < 4; i++ ) { VectorCopy( verts[i], v[0] ); VectorCopy( verts[( i+1 )%4], v[1] ); VectorCopy( verts[( i+2 )%4], v[2] ); PlaneFromPoints( v, &plane ); if( fabs( DotProduct( mainplane.normal, plane.normal ) ) < 0.9 ) return 0; } } } numbrushplanes = 0; // add front plane SnapPlane( mainplane.normal, &mainplane.dist ); VectorCopy( mainplane.normal, brushplanes[numbrushplanes].normal ); brushplanes[numbrushplanes].dist = mainplane.dist; numbrushplanes++; // calculate mins & maxs ClearBounds( facet->absmins, facet->absmaxs ); for( i = 0; i < numverts; i++ ) AddPointToBounds( verts[i], facet->absmins, facet->absmaxs ); // add the axial planes for( axis = 0; axis < 3; axis++ ) { for( dir = -1; dir <= 1; dir += 2 ) { for( i = 0; i < numbrushplanes; i++ ) { if( brushplanes[i].normal[axis] == dir ) break; } if( i == numbrushplanes ) { VectorClear( normal ); normal[axis] = dir; if( dir == 1 ) dist = facet->absmaxs[axis]; else dist = -facet->absmins[axis]; VectorCopy( normal, brushplanes[numbrushplanes].normal ); brushplanes[numbrushplanes].dist = dist; numbrushplanes++; } } } // add the edge bevels for( i = 0; i < numverts; i++ ) { j = ( i + 1 ) % numverts; // k = ( i + 2 ) % numverts; VectorSubtract( verts[i], verts[j], vec ); if( VectorNormalize( vec ) < 0.5 ) continue; SnapVector( vec ); for( j = 0; j < 3; j++ ) { if( vec[j] == 1 || vec[j] == -1 ) break; // axial } if( j != 3 ) continue; // only test non-axial edges // try the six possible slanted axials from this edge for( axis = 0; axis < 3; axis++ ) { for( dir = -1; dir <= 1; dir += 2 ) { // construct a plane VectorClear( vec2 ); vec2[axis] = dir; CrossProduct( vec, vec2, normal ); if( VectorNormalize( normal ) < 0.5 ) continue; dist = DotProduct( verts[i], normal ); for( j = 0; j < numbrushplanes; j++ ) { // if this plane has already been used, skip it if( ComparePlanes( brushplanes[j].normal, brushplanes[j].dist, normal, dist ) ) break; } if( j != numbrushplanes ) continue; // if all other points are behind this plane, it is a proper edge bevel for( j = 0; j < numverts; j++ ) { if( j != i ) { d = DotProduct( verts[j], normal ) - dist; if( d > 0.1 ) break; // point in front: this plane isn't part of the outer hull } } if( j != numverts ) continue; // add this plane VectorCopy( normal, brushplanes[numbrushplanes].normal ); brushplanes[numbrushplanes].dist = dist; numbrushplanes++; if( numbrushplanes == MAX_FACET_PLANES ) break; } } } return ( facet->numsides = numbrushplanes ); } /* * CM_CreatePatch */ static void CM_CreatePatch(model_t *loadmodel, q3cpatch_t *patch, q2mapsurface_t *shaderref, const vec_t *verts, const unsigned short *patch_cp, const unsigned short *patch_subdiv) { int step[2], size[2], flat[2]; int i, j, k ,u, v; int numsides, totalsides; q2cbrush_t *facets, *facet; vecV_t *points; vec3_t tverts[4]; qbyte *data; mplane_t *brushplanes; float subdivlevel; patch->surface = shaderref; if (patch_subdiv) { //fixed step[0] = patch_subdiv[0]; step[1] = patch_subdiv[1]; } else { // find the degree of subdivision in the u and v directions subdivlevel = cm_subdivlevel;//r_subdivisions.value; if ( subdivlevel < 1 ) subdivlevel = 1; Patch_GetFlatness( subdivlevel, verts, sizeof(vecV_t)/sizeof(vec_t), patch_cp, flat ); step[0] = 1 << flat[0]; step[1] = 1 << flat[1]; } if (!step[0] || !step[1]) { size[0] = patch_cp[0]; size[1] = patch_cp[1]; } else { size[0] = ( patch_cp[0] >> 1 ) * step[0] + 1; size[1] = ( patch_cp[1] >> 1 ) * step[1] + 1; } if( size[0] <= 0 || size[1] <= 0 ) return; data = BZ_Malloc( size[0] * size[1] * sizeof( vecV_t ) + ( size[0]-1 ) * ( size[1]-1 ) * 2 * ( sizeof( q2cbrush_t ) + 32 * sizeof( mplane_t ) ) ); points = ( vecV_t * )data; data += size[0] * size[1] * sizeof( vecV_t ); facets = ( q2cbrush_t * )data; data += ( size[0]-1 ) * ( size[1]-1 ) * 2 * sizeof( q2cbrush_t ); brushplanes = ( mplane_t * )data; data += ( size[0]-1 ) * ( size[1]-1 ) * 2 * MAX_FACET_PLANES * sizeof( mplane_t ); // fill in Patch_Evaluate(verts, patch_cp, step, points[0], sizeof(vecV_t)/sizeof(vec_t)); totalsides = 0; patch->numfacets = 0; patch->facets = NULL; ClearBounds( patch->absmins, patch->absmaxs ); // create a set of facets for( v = 0; v < size[1]-1; v++ ) { for( u = 0; u < size[0]-1; u++ ) { i = v * size[0] + u; VectorCopy( points[i], tverts[0] ); VectorCopy( points[i + size[0]], tverts[1] ); VectorCopy( points[i + size[0] + 1], tverts[2] ); VectorCopy( points[i + 1], tverts[3] ); for( i = 0; i < 4; i++ ) AddPointToBounds( tverts[i], patch->absmins, patch->absmaxs ); // try to create one facet from a quad numsides = CM_CreateFacetFromPoints( &facets[patch->numfacets], tverts, 4, shaderref, brushplanes + totalsides ); if( !numsides ) { // create two facets from triangles VectorCopy( tverts[3], tverts[2] ); numsides = CM_CreateFacetFromPoints( &facets[patch->numfacets], tverts, 3, shaderref, brushplanes + totalsides ); if( numsides ) { totalsides += numsides; patch->numfacets++; } VectorCopy( tverts[2], tverts[0] ); VectorCopy( points[v *size[0] + u + size[0] + 1], tverts[2] ); numsides = CM_CreateFacetFromPoints( &facets[patch->numfacets], tverts, 3, shaderref, brushplanes + totalsides ); } if( numsides ) { totalsides += numsides; patch->numfacets++; } } } if (patch->numfacets) { qbyte *data; data = ZG_Malloc(&loadmodel->memgroup, patch->numfacets * sizeof( q2cbrush_t ) + totalsides * ( sizeof( q2cbrushside_t ) + sizeof( mplane_t ) )); patch->facets = ( q2cbrush_t * )data; data += patch->numfacets * sizeof( q2cbrush_t ); memcpy( patch->facets, facets, patch->numfacets * sizeof( q2cbrush_t ) ); for( i = 0, k = 0, facet = patch->facets; i < patch->numfacets; i++, facet++ ) { mplane_t *planes; q2cbrushside_t *s; facet->brushside = ( q2cbrushside_t * )data; data += facet->numsides * sizeof( q2cbrushside_t ); planes = ( mplane_t * )data; data += facet->numsides * sizeof( mplane_t ); for( j = 0, s = facet->brushside; j < facet->numsides; j++, s++ ) { planes[j] = brushplanes[k++]; s->plane = &planes[j]; SnapPlane( s->plane->normal, &s->plane->dist ); CategorizePlane( s->plane ); s->surface = shaderref; } } for( i = 0; i < 3; i++ ) { // spread the mins / maxs by a pixel patch->absmins[i] -= 1; patch->absmaxs[i] += 1; } } BZ_Free( points ); } //====================================================== static qboolean CM_CreatePatchForFace (model_t *loadmodel, cminfo_t *prv, mleaf_t *leaf, int facenum, int *checkout) { size_t u; q3cface_t *face; q2mapsurface_t *surf; q3cpatch_t *patch; q3cmesh_t *cmesh; face = &prv->faces[facenum]; if (face->numverts <= 0) return true; if (face->shadernum < 0 || face->shadernum >= loadmodel->numtextures) return true; surf = &prv->surfaces[face->shadernum]; if (!surf->c.value) //surface has no contents value, so can't ever block anything. return true; switch(face->facetype) { case MST_TRIANGLE_SOUP: if (!face->soup.numindicies) return true; //only enable mesh collisions if its meant to be enabled. //we haven't parsed any shaders, so we depend upon the stuff that the bsp compiler left lying around. if (!(surf->c.flags & q3bsp_surf_meshcollision_flag.ival) && !q3bsp_surf_meshcollision_force.ival) return true; if (prv->numleafcmeshes >= prv->maxleafcmeshes) { prv->maxleafcmeshes *= 2; prv->maxleafcmeshes += 16; if (prv->numleafcmeshes > prv->maxleafcmeshes) { //detect overflow Con_Printf (CON_ERROR "CM_CreateCMeshesForLeafs: map is insanely huge!\n"); return false; } prv->leafcmeshes = realloc(prv->leafcmeshes, sizeof(*prv->leafcmeshes) * prv->maxleafcmeshes); } // the patch was already built if (checkout[facenum] != -1) { prv->leafcmeshes[prv->numleafcmeshes] = checkout[facenum]; cmesh = &prv->cmeshes[checkout[facenum]]; } else { if (prv->numcmeshes >= MAX_CM_PATCHES) { Con_Printf (CON_ERROR "CM_CreatePatchesForLeafs: map has too many patches\n"); return false; } cmesh = &prv->cmeshes[prv->numcmeshes]; prv->leafcmeshes[prv->numleafcmeshes] = prv->numcmeshes; checkout[facenum] = prv->numcmeshes++; //gcc warns without this cast cmesh->surface = surf; cmesh->numverts = face->numverts; cmesh->numincidies = face->soup.numindicies; cmesh->xyz_array = ZG_Malloc(&loadmodel->memgroup, cmesh->numverts * sizeof(*cmesh->xyz_array) + cmesh->numincidies * sizeof(*cmesh->indicies)); cmesh->indicies = (index_t*)(cmesh->xyz_array + cmesh->numverts); VectorCopy(prv->verts[face->firstvert+0], cmesh->xyz_array[0]); VectorCopy(cmesh->xyz_array[0], cmesh->absmaxs); VectorCopy(cmesh->xyz_array[0], cmesh->absmins); for (u = 1; u < cmesh->numverts; u++) { VectorCopy(prv->verts[face->firstvert+u], cmesh->xyz_array[u]); AddPointToBounds(cmesh->xyz_array[u], cmesh->absmins, cmesh->absmaxs); } for (u = 0; u < cmesh->numincidies; u++) cmesh->indicies[u] = prv->surfindexes[face->soup.firstindex+u]; } leaf->contents |= surf->c.value; leaf->numleafcmeshes++; prv->numleafcmeshes++; break; case MST_PATCH: case MST_PATCH_FIXED: if (face->patch.cp[0] <= 0 || face->patch.cp[1] <= 0) return true; if ( !surf->c.value || (surf->c.flags & Q3SURF_NONSOLID) ) return true; if (prv->numleafpatches >= prv->maxleafpatches) { prv->maxleafpatches *= 2; prv->maxleafpatches += 16; if (prv->numleafpatches > prv->maxleafpatches) { //detect overflow Con_Printf (CON_ERROR "CM_CreatePatchesForLeafs: map is insanely huge!\n"); return false; } prv->leafpatches = realloc(prv->leafpatches, sizeof(*prv->leafpatches) * prv->maxleafpatches); } // the patch was already built if (checkout[facenum] != -1) { prv->leafpatches[prv->numleafpatches] = checkout[facenum]; patch = &prv->patches[checkout[facenum]]; } else { if (prv->numpatches >= MAX_CM_PATCHES) { Con_Printf (CON_ERROR "CM_CreatePatchesForLeafs: map has too many patches\n"); return false; } patch = &prv->patches[prv->numpatches]; prv->leafpatches[prv->numleafpatches] = prv->numpatches; checkout[facenum] = prv->numpatches++; //gcc warns without this cast CM_CreatePatch (loadmodel, patch, surf, (const vec_t *)(prv->verts + face->firstvert), face->patch.cp, (face->facetype==MST_PATCH_FIXED)?face->patch.fixedres:NULL ); } leaf->contents |= patch->surface->c.value; leaf->numleafpatches++; prv->numleafpatches++; break; } return true; } static qboolean CM_CreatePatchesForLeaf (model_t *loadmodel, cminfo_t *prv, mleaf_t *leaf, int *checkout) { int j, k; leaf->numleafpatches = 0; leaf->firstleafpatch = prv->numleafpatches; leaf->numleafcmeshes = 0; leaf->firstleafcmesh = prv->numleafcmeshes; if (leaf->cluster == -1) return true; for (j=0 ; jnummarksurfaces ; j++) { k = leaf->firstmarksurface[j] - loadmodel->surfaces; if (k >= prv->numfaces) { Con_Printf (CON_ERROR "CM_CreatePatchesForLeafs: corrupt map\n"); break; } if (!CM_CreatePatchForFace (loadmodel, prv, leaf, k, checkout)) return false; } return true; } /* ================= CM_CreatePatchesForLeafs ================= */ static qboolean CM_CreatePatchesForLeafs (model_t *loadmodel, cminfo_t *prv) { int i, k; mleaf_t *leaf; int *checkout = alloca(sizeof(int)*prv->numfaces); if (map_noCurves.ival) return true; memset (checkout, -1, sizeof(int)*prv->numfaces); for (i = 0; i < prv->numcmodels; i++) { prv->cmodels[i].firstpatch = prv->numpatches; prv->cmodels[i].firstcmesh = prv->numcmeshes; if (i == 0) { //worldmodel's leafs for (k = 0, leaf = loadmodel->leafs; k < loadmodel->numleafs; k++, leaf++) if (!CM_CreatePatchesForLeaf(loadmodel, prv, leaf, checkout)) return false; } else { //submodel uni-leaf thing. leaf = prv->cmodels[i].headleaf; if (leaf) { if (!CM_CreatePatchesForLeaf(loadmodel, prv, leaf, checkout)) return false; for (k = 0; k < prv->cmodels[i].numsurfaces; k++) CM_CreatePatchForFace(loadmodel, prv, leaf, prv->cmodels[i].firstsurface+k, checkout); } } prv->cmodels[i].num_patches = prv->numpatches-prv->cmodels[i].firstpatch; prv->cmodels[i].num_cmeshes = prv->numcmeshes-prv->cmodels[i].firstcmesh; } return true; } #endif /* =============================================================================== MAP LOADING =============================================================================== */ static void CMod_SetParent (mnode_t *node, mnode_t *parent) { node->parent = parent; if (node->contents != -1) return; CMod_SetParent (node->children[0], node); CMod_SetParent (node->children[1], node); } #ifdef Q2BSPS /* ================= CMod_LoadSubmodels ================= */ static qboolean CModQ2_LoadSubmodels (model_t *loadmodel, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)loadmodel->meshinfo; q2dmodel_t *in; cmodel_t *out; int i, j, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no models\n"); return false; } if (count > SANITY_MAX_Q2MAP_MODELS) { Con_Printf (CON_ERROR "Map has too many models\n"); return false; } out = prv->cmodels = ZG_Malloc(&loadmodel->memgroup, count * sizeof(*prv->cmodels)); prv->numcmodels = count; for (i=0 ; imins[j] = LittleFloat (in->mins[j]) - 1; out->maxs[j] = LittleFloat (in->maxs[j]) + 1; out->origin[j] = LittleFloat (in->origin[j]); } out->headnode = loadmodel->nodes + LittleLong (in->headnode); out->firstsurface = LittleLong (in->firstface); out->numsurfaces = LittleLong (in->numfaces); } AddPointToBounds(prv->cmodels[0].mins, loadmodel->mins, loadmodel->maxs); AddPointToBounds(prv->cmodels[0].maxs, loadmodel->mins, loadmodel->maxs); return true; } /* ================= CMod_LoadSurfaces ================= */ static qboolean CModQ2_LoadSurfaces (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; q2texinfo_t *in; q2mapsurface_t *out; int i, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no surfaces\n"); return false; } // if (count > MAX_Q2MAP_TEXINFO) // Host_Error ("Map has too many surfaces"); mod->numtexinfo = count; out = prv->surfaces = ZG_Malloc(&mod->memgroup, count * sizeof(*prv->surfaces)); for ( i=0 ; ic.name, in->texture, sizeof(out->c.name)); Q_strncpyz (out->rname, in->texture, sizeof(out->rname)); out->c.flags = LittleLong (in->flags); out->c.value = LittleLong (in->value); } return true; } #ifndef SERVERONLY static texture_t *Mod_LoadWall(model_t *loadmodel, char *mapname, char *texname, char *shadername, unsigned int imageflags) { char name[MAX_QPATH]; q2miptex_t replacementwal; texture_t *tex; q2miptex_t *wal; image_t *base; Q_snprintfz (name, sizeof(name), "textures/%s.wal", texname); wal = (void *)FS_LoadMallocFile (name, NULL); if (!wal) { wal = &replacementwal; memset(wal, 0, sizeof(*wal)); Q_strncpyz(wal->name, texname, sizeof(wal->name)); wal->width = 64; wal->height = 64; } else { wal->width = LittleLong(wal->width); wal->height = LittleLong(wal->height); } { int i; for (i = 0; i < MIPLEVELS; i++) wal->offsets[i] = LittleLong(wal->offsets[i]); } wal->flags = LittleLong(wal->flags); wal->contents = LittleLong(wal->contents); wal->value = LittleLong(wal->value); tex = ZG_Malloc(&loadmodel->memgroup, sizeof(texture_t)); tex->vwidth = tex->srcwidth = wal->width; tex->vheight = tex->srcheight = wal->height; if (!tex->vwidth || !tex->vheight || wal == &replacementwal) { imageflags |= IF_LOADNOW; //make sure the size is known BEFORE it returns. if (wal->offsets[0]) base = R_LoadReplacementTexture(wal->name, "bmodels", imageflags, (qbyte *)wal+wal->offsets[0], wal->width, wal->height, TF_SOLID8); else base = R_LoadReplacementTexture(wal->name, "bmodels", imageflags, NULL, 0, 0, TF_INVALID); } else base = NULL; if (wal == &replacementwal) { if (base) { if (base->status == TEX_LOADED||base->status==TEX_LOADING) { tex->vwidth = base->width; tex->vheight = base->height; } else Con_Printf("Unable to load textures/%s.wal\n", wal->name); } } else { qbyte *out; unsigned int size = (wal->width>>0)*(wal->height>>0) + (wal->width>>1)*(wal->height>>1) + (wal->width>>2)*(wal->height>>2) + (wal->width>>3)*(wal->height>>3); tex->srcdata = out = BZ_Malloc(size); tex->srcfmt = TF_MIP4_8PAL24_T255; tex->palette = host_basepal; memcpy(out, (qbyte *)wal + wal->offsets[0], (wal->width>>0)*(wal->height>>0)); out += (wal->width>>0)*(wal->height>>0); memcpy(out, (qbyte *)wal + wal->offsets[1], (wal->width>>1)*(wal->height>>1)); out += (wal->width>>1)*(wal->height>>1); memcpy(out, (qbyte *)wal + wal->offsets[2], (wal->width>>2)*(wal->height>>2)); out += (wal->width>>2)*(wal->height>>2); memcpy(out, (qbyte *)wal + wal->offsets[3], (wal->width>>3)*(wal->height>>3)); out += (wal->width>>3)*(wal->height>>3); BZ_Free(wal); } return tex; } static qboolean CModQ2_LoadTexInfo (model_t *mod, qbyte *mod_base, lump_t *l, char *mapname) //yes I know these load from the same place { q2texinfo_t *in; mtexinfo_t *out; int i, j, count; char *lwr; char sname[MAX_QPATH]; int texcount; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf ("MOD_LoadBmodel: funny lump size in %s\n", mod->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); mod->textures = ZG_Malloc(&mod->memgroup, sizeof(texture_t *)*count); texcount = 0; mod->texinfo = out; mod->numtexinfo = count; if (in[0].nexttexinfo != -1) { for (i = 1; i < count && in[i].nexttexinfo == in[0].nexttexinfo; i++) ; if (i == count) { Con_Printf("WARNING: invalid texture animations in \"%s\"\n", mod->name); for (i = 0; i < count; i++) in[i].nexttexinfo = -1; } } for ( i=0 ; iflags = LittleLong (in->flags); for (j=0 ; j<4 ; j++) out->vecs[0][j] = LittleFloat (in->vecs[0][j]); for (j=0 ; j<4 ; j++) out->vecs[1][j] = LittleFloat (in->vecs[1][j]); out->vecscale[0] = 1.0/Length (out->vecs[0]); out->vecscale[1] = 1.0/Length (out->vecs[1]); if (out->flags & TI_SKY) Q_snprintfz(sname, sizeof(sname), "sky/%s", in->texture); else Q_snprintfz(sname, sizeof(sname), "%s", in->texture); if (out->flags & (TI_WARP)) Q_strncatz(sname, "#WARP", sizeof(sname)); if (out->flags & TI_FLOWING) Q_strncatz(sname, "#FLOW", sizeof(sname)); if (out->flags & TI_TRANS66) Q_strncatz(sname, "#ALPHA=0.66", sizeof(sname)); else if (out->flags & TI_TRANS33) Q_strncatz(sname, "#ALPHA=0.33", sizeof(sname)); else if (out->flags & (TI_WARP)) Q_strncatz(sname, "#ALPHA=1", sizeof(sname)); if (in->nexttexinfo != -1) //used to ensure non-looping and looping don't conflict and get confused. Q_strncatz(sname, "#ANIMLOOP", sizeof(sname)); //in q2, 'TEX_SPECIAL' is TI_LIGHT, and that conflicts. out->flags &= ~TI_LIGHT; if (out->flags & (TI_SKY|TI_TRANS33|TI_TRANS66|TI_WARP)) out->flags |= TEX_SPECIAL; //compact the textures. for (j=0; j < texcount; j++) { if (!strcmp(sname, mod->textures[j]->name)) { out->texture = mod->textures[j]; break; } } if (j == texcount) //load a new one { for (lwr = in->texture; *lwr; lwr++) { if (*lwr >= 'A' && *lwr <= 'Z') *lwr = *lwr - 'A' + 'a'; } out->texture = Mod_LoadWall (mod, mapname, in->texture, sname, (out->flags&TEX_SPECIAL)?0:IF_NOALPHA); if (!out->texture || !out->texture->srcwidth || !out->texture->srcheight) { out->texture = ZG_Malloc(&mod->memgroup, sizeof(texture_t) + 16*16+8*8+4*4+2*2); Con_Printf (CON_WARNING "Couldn't load \"%s.wal\"\n", in->texture); memcpy(out->texture, r_notexture_mip, sizeof(texture_t) + 16*16+8*8+4*4+2*2); } Q_strncpyz(out->texture->name, sname, sizeof(out->texture->name)); mod->textures[texcount++] = out->texture; } // if (in->nexttexinfo != -1) // { // Con_DPrintf("FIXME: %s should animate to %s\n", in->texture, (in->nexttexinfo+(q2texinfo_t *)(mod_base + l->fileofs))->texture); // } } in = (void *)(mod_base + l->fileofs); out = mod->texinfo; for (i=0 ; i= 0 && in[i].nexttexinfo < count) out[i].texture->anim_next = out[in[i].nexttexinfo].texture; } for (i=0 ; ianim_next) continue; out[i].texture->anim_total = 1; for (tex = out[i].texture->anim_next ; tex && tex != out[i].texture && out[i].texture->anim_total < 100; tex=tex->anim_next) out[i].texture->anim_total++; } mod->numtextures = texcount; Mod_SortShaders(mod); return true; } #endif /* static void CalcSurfaceExtents (msurface_t *s) { float mins[2], maxs[2], val; int i,j, e; mvertex_t *v; mtexinfo_t *tex; int bmins[2], bmaxs[2]; mins[0] = mins[1] = 999999; maxs[0] = maxs[1] = -99999; tex = s->texinfo; for (i=0 ; inumedges ; i++) { e = loadmodel->surfedges[s->firstedge+i]; if (e >= 0) v = &loadmodel->vertexes[loadmodel->edges[e].v[0]]; else v = &loadmodel->vertexes[loadmodel->edges[-e].v[1]]; for (j=0 ; j<2 ; j++) { val = v->position[0] * tex->vecs[j][0] + v->position[1] * tex->vecs[j][1] + v->position[2] * tex->vecs[j][2] + tex->vecs[j][3]; if (val < mins[j]) mins[j] = val; if (val > maxs[j]) maxs[j] = val; } } for (i=0 ; i<2 ; i++) { bmins[i] = floor(mins[i]/16); bmaxs[i] = ceil(maxs[i]/16); s->texturemins[i] = bmins[i] * 16; s->extents[i] = (bmaxs[i] - bmins[i]) * 16; // if ( !(tex->flags & TEX_SPECIAL) && s->extents[i] > 512 )// 256 ) // Sys_Error ("Bad surface extents"); } }*/ /* ================= Mod_LoadFaces ================= */ #ifndef SERVERONLY static qboolean CModQ2_LoadFaces (model_t *mod, qbyte *mod_base, lump_t *l, lump_t *lightlump, qboolean lightofsisdouble, bspx_header_t *bspx) { dsface_t *in; msurface_t *out; int i, count, surfnum; int planenum, side; int ti; int style; unsigned short lmshift, lmscale; char buf[64]; lightmapoverrides_t overrides = {0}; overrides.defaultshift = LMSHIFT_DEFAULT; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf ("MOD_LoadBmodel: funny lump size in %s\n",mod->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&mod->memgroup, (count+6)*sizeof(*out)); //spare for skybox mod->surfaces = out; mod->numsurfaces = count; Mod_LoadLighting(mod, bspx, mod_base, lightlump, lightofsisdouble, &overrides); if (overrides.offsets) lmshift = overrides.defaultshift; else { lmscale = atoi(Mod_ParseWorldspawnKey(mod, "lightmap_scale", buf, sizeof(buf))); if (!lmscale) lmshift = LMSHIFT_DEFAULT; else { for(lmshift = 0; lmscale > 1; lmshift++) lmscale >>= 1; } } for ( surfnum=0 ; surfnumfirstedge = LittleLong(in->firstedge); out->numedges = (unsigned short)LittleShort(in->numedges); out->flags = 0; planenum = (unsigned short)LittleShort(in->planenum); side = (unsigned short)LittleShort(in->side); if (side) out->flags |= SURF_PLANEBACK; out->plane = mod->planes + planenum; ti = (unsigned short)LittleShort (in->texinfo); if (ti < 0 || ti >= mod->numtexinfo) { Con_Printf (CON_ERROR "MOD_LoadBmodel: bad texinfo number\n"); return false; } out->texinfo = mod->texinfo + ti; #ifndef SERVERONLY if (out->texinfo->flags & TI_SKY) { out->flags |= SURF_DRAWSKY; } if (out->texinfo->flags & TI_WARP) { out->flags |= SURF_DRAWTURB|SURF_DRAWTILED; } #endif if (overrides.shifts) out->lmshift = overrides.shifts[surfnum]; else out->lmshift = lmshift; CalcSurfaceExtents (mod, out); if (overrides.extents) { out->extents[0] = overrides.extents[surfnum*2+0]; out->extents[1] = overrides.extents[surfnum*2+1]; } // lighting info if (overrides.styles16) { for (i=0 ; ilightmaps.maxstyle < style) mod->lightmaps.maxstyle = style; out->styles[i] = style; } } else if (overrides.styles8) { for (i=0 ; ilightmaps.maxstyle < style) mod->lightmaps.maxstyle = style; out->styles[i] = style; } } else { for (i=0 ; i<4 ; i++) { style = in->styles[i]; if (style == 0xff) style = INVALID_LIGHTSTYLE; else if (mod->lightmaps.maxstyle < style) mod->lightmaps.maxstyle = style; out->styles[i] = style; } } for ( ; istyles[i] = INVALID_LIGHTSTYLE; if (overrides.offsets) i = overrides.offsets[surfnum]; else i = LittleLong(in->lightofs); if (i == -1) out->samples = NULL; else if (lightofsisdouble) out->samples = mod->lightdata + (i/2); else out->samples = mod->lightdata + i; // set the drawing flags if (out->texinfo->flags & TI_WARP) { out->flags |= SURF_DRAWTURB; for (i=0 ; i<2 ; i++) { out->extents[i] = 16384; out->texturemins[i] = -8192; } } } return true; } #endif /* ================= CMod_LoadNodes ================= */ static qboolean CModQ2_LoadNodes (model_t *mod, qbyte *mod_base, lump_t *l) { q2dnode_t *in; int child; mnode_t *out; int i, j, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map has no nodes\n"); return false; } if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Map has too many nodes\n"); return false; } out = ZG_Malloc(&mod->memgroup, sizeof(mnode_t)*count); mod->nodes = out; mod->numnodes = count; for (i=0 ; iminmaxs[j] = LittleShort (in->mins[j]); out->minmaxs[3+j] = LittleShort (in->maxs[j]); } out->plane = mod->planes + LittleLong(in->planenum); out->firstsurface = (unsigned short)LittleShort (in->firstface); out->numsurfaces = (unsigned short)LittleShort (in->numfaces); out->contents = -1; // differentiate from leafs for (j=0 ; j<2 ; j++) { child = LittleLong (in->children[j]); out->childnum[j] = child; if (child < 0) out->children[j] = (mnode_t *)(mod->leafs + -1-child); else out->children[j] = mod->nodes + child; } } CMod_SetParent (mod->nodes, NULL); // sets nodes and leafs return true; } /* ================= CMod_LoadBrushes ================= */ static qboolean CModQ2_LoadBrushes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; q2dbrush_t *in; q2cbrush_t *out; int i, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > SANITY_MAX_MAP_BRUSHES) { Con_Printf (CON_ERROR "Map has too many brushes"); return false; } prv->brushes = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+1)); out = prv->brushes; prv->numbrushes = count; for (i=0 ; ibrushside = &prv->brushsides[LittleLong(in->firstside)]; out->numsides = LittleLong(in->numsides); out->contents = LittleLong(in->contents); CM_FinalizeBrush(out); } return true; } /* ================= CMod_LoadLeafs ================= */ static qboolean CModQ2_LoadLeafs (model_t *mod, qbyte *mod_base, lump_t *l) { int i, j; mleaf_t *out; q2dleaf_t *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no leafs\n"); return false; } // need to save space for box planes if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Map has too many leafs\n"); return false; } out = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+1)); mod->numclusters = 0; mod->leafs = out; mod->numleafs = count; for ( i=0 ; iminmaxs[j] = LittleShort (in->mins[j]); out->minmaxs[3+j] = LittleShort (in->maxs[j]); } out->contents = LittleLong (in->contents); out->cluster = (unsigned short)LittleShort (in->cluster); if (out->cluster == 0xffff) out->cluster = -1; out->area = (unsigned short)LittleShort (in->area); out->firstleafbrush = (unsigned short)LittleShort (in->firstleafbrush); out->numleafbrushes = (unsigned short)LittleShort (in->numleafbrushes); out->firstmarksurface = mod->marksurfaces + (unsigned short)LittleShort(in->firstleafface); out->nummarksurfaces = (unsigned short)LittleShort(in->numleaffaces); if (out->cluster >= mod->numclusters) mod->numclusters = out->cluster + 1; } out = mod->leafs; mod->pvsbytes = ((mod->numclusters + 31)>>3)&~3; if (out[0].contents != Q2CONTENTS_SOLID) { Con_Printf (CON_ERROR "Map leaf 0 is not CONTENTS_SOLID\n"); return false; } return true; } /* ================= CMod_LoadPlanes ================= */ static qboolean CModQ2_LoadPlanes (model_t *mod, qbyte *mod_base, lump_t *l) { int i, j; mplane_t *out; dplane_t *in; int count; int bits; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no planes\n"); return false; } // need to save space for box planes if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Map has too many planes (%i)\n", count); return false; } mod->planes = out = ZG_Malloc(&mod->memgroup, sizeof(*out) * count); mod->numplanes = count; for ( i=0 ; inormal[j] = LittleFloat (in->normal[j]); if (out->normal[j] < 0) bits |= 1<dist = LittleFloat (in->dist); out->type = LittleLong (in->type); out->signbits = bits; } return true; } /* ================= CMod_LoadLeafBrushes ================= */ static qboolean CModQ2_LoadLeafBrushes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; q2cbrush_t **out; unsigned short *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no planes\n"); return false; } // need to save space for box planes if (count > SANITY_MAX_MAP_LEAFBRUSHES) { Con_Printf (CON_ERROR "Map has too many leafbrushes\n"); return false; } //prv->numbrushes is because of submodels being weird. out = prv->leafbrushes = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+prv->numbrushes)); prv->numleafbrushes = count; for ( i=0 ; ibrushes + (unsigned short)(short)LittleShort (*in); return true; } /* ================= CMod_LoadBrushSides ================= */ static qboolean CModQ2_LoadBrushSides (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; unsigned int i, j; q2cbrushside_t *out; q2dbrushside_t *in; int count; int num; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); // need to save space for box planes if (count > SANITY_MAX_MAP_BRUSHSIDES) { Con_Printf (CON_ERROR "Map has too many brushsides (%i)\n", count); return false; } out = prv->brushsides = ZG_Malloc(&mod->memgroup, sizeof(*out) * count); prv->numbrushsides = count; for ( i=0 ; iplanenum); out->plane = &mod->planes[num]; j = (unsigned short)LittleShort (in->texinfo); if (j >= mod->numtexinfo) out->surface = &nullsurface; else out->surface = &prv->surfaces[j]; } return true; } /* ================= CMod_LoadAreas ================= */ static qboolean CModQ2_LoadAreas (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; q2carea_t *out; q2darea_t *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > MAX_Q2MAP_AREAS) { Con_Printf (CON_ERROR "Map has too many areas\n"); return false; } out = prv->q2areas = ZG_Malloc(&mod->memgroup, sizeof(*out) * count);; prv->numareas = count; for ( i=0 ; inumareaportals = LittleLong (in->numareaportals); out->firstareaportal = LittleLong (in->firstareaportal); } return true; } /* ================= CMod_LoadAreaPortals ================= */ static qboolean CModQ2_LoadAreaPortals (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; q2dareaportal_t *out; q2dareaportal_t *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > MAX_Q2MAP_AREAS) { Con_Printf (CON_ERROR "Map has too many areas\n"); return false; } out = prv->q2areaportals = ZG_Malloc(&mod->memgroup, sizeof(*out) * count); prv->numq2areaportals = count; for ( i=0 ; iportalnum = LittleLong (in->portalnum); out->otherarea = LittleLong (in->otherarea); } return true; } /* ================= CMod_LoadVisibility ================= */ static qboolean CModQ2_LoadVisibility (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; prv->numvisibility = l->filelen; // if (l->filelen > MAX_Q2MAP_VISIBILITY) // { // Con_Printf (CON_ERROR "Map has too large visibility lump\n"); // return false; // } prv->q2vis = ZG_Malloc(&mod->memgroup, l->filelen); memcpy (prv->q2vis, mod_base + l->fileofs, l->filelen); mod->vis = prv->q2vis; prv->q2vis->numclusters = LittleLong (prv->q2vis->numclusters); for (i=0 ; iq2vis->numclusters ; i++) { prv->q2vis->bitofs[i][0] = LittleLong (prv->q2vis->bitofs[i][0]); prv->q2vis->bitofs[i][1] = LittleLong (prv->q2vis->bitofs[i][1]); } mod->numclusters = prv->q2vis->numclusters; mod->pvsbytes = ((mod->numclusters + 31)>>3)&~3; return true; } #endif //q2bsps #ifdef Q3BSPS static qboolean CModQ3_LoadMarksurfaces (model_t *loadmodel, qbyte *mod_base, lump_t *l) { int i, j, count; int *in; msurface_t **out; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "CModQ3_LoadMarksurfaces: funny lump size in %s\n",loadmodel->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&loadmodel->memgroup, count*sizeof(*out)); loadmodel->marksurfaces = out; loadmodel->nummarksurfaces = count; for ( i=0 ; i= loadmodel->numsurfaces) { Con_Printf (CON_ERROR "Mod_ParseMarksurfaces: bad surface number\n"); return false; } out[i] = loadmodel->surfaces + j; } return true; } static qboolean CModQ3_LoadSubmodels (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; q3dmodel_t *in; cmodel_t *out; int i, j, count; q2cbrush_t **leafbrush; mleaf_t *bleaf; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no models\n"); return false; } if (count > SANITY_MAX_Q2MAP_MODELS) { Con_Printf (CON_ERROR "Map has too many models\n"); return false; } out = prv->cmodels = ZG_Malloc(&mod->memgroup, count * sizeof(*prv->cmodels)); prv->numcmodels = count; if (count > 1) bleaf = ZG_Malloc(&mod->memgroup, (count-1) * sizeof(*bleaf)); else bleaf = NULL; prv->mapisq3 = true; for (i=0 ; imins[j] = LittleFloat (in->mins[j]) - 1; out->maxs[j] = LittleFloat (in->maxs[j]) + 1; out->origin[j] = (out->maxs[j] + out->mins[j])/2; } out->firstsurface = LittleLong (in->firstsurface); out->numsurfaces = LittleLong (in->num_surfaces); out->firstbrush = LittleLong(in->firstbrush); out->num_brushes = LittleLong(in->num_brushes); if (!i) { out->headnode = mod->nodes; out->headleaf = NULL; } else { //create a new leaf to hold the brushes and be directly clipped out->headleaf = bleaf; out->headnode = NULL; bleaf->numleafbrushes = LittleLong ( in->num_brushes ); bleaf->firstleafbrush = prv->numleafbrushes; bleaf->contents = 0; leafbrush = &prv->leafbrushes[prv->numleafbrushes]; for ( j = 0; j < bleaf->numleafbrushes; j++, leafbrush++ ) { *leafbrush = prv->brushes + LittleLong ( in->firstbrush ) + j; bleaf->contents |= (*leafbrush)->contents; } prv->numleafbrushes += bleaf->numleafbrushes; bleaf++; } //submodels } AddPointToBounds(prv->cmodels[0].mins, mod->mins, mod->maxs); AddPointToBounds(prv->cmodels[0].maxs, mod->mins, mod->maxs); return true; } static qboolean CModQ3_LoadShaders (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; dq3shader_t *in; q2mapsurface_t *out; int i, count; texture_t *tex; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no shaders\n"); return false; } // else if (count > MAX_Q2MAP_TEXINFO) // Host_Error ("Map has too many shaders"); mod->numtexinfo = count; out = prv->surfaces = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); mod->textures = ZG_Malloc(&mod->memgroup, (sizeof(texture_t*)+sizeof(mtexinfo_t)+sizeof(texture_t))*(count*2+1)); //+1 is 'noshader' for flares. mod->texinfo = (mtexinfo_t*)(mod->textures+(count*2+1)); tex = (texture_t*)(mod->texinfo+(count*2+1)); mod->numtextures = count*2+1; for ( i=0 ; ic.flags = LittleLong ( in->surfflags ); out->c.value = LittleLong ( in->contents ); mod->texinfo[i].texture = tex+i; mod->texinfo[i].flags = prv->surfaces[i].c.flags; Q_strncpyz(mod->texinfo[i].texture->name, in->shadername, sizeof(mod->texinfo[i].texture->name)); mod->textures[i] = mod->texinfo[i].texture; } for ( i=0, in-=count ; itexinfo[i+count].texture = tex+i+count; mod->texinfo[i+count].flags = prv->surfaces[i].c.flags; Q_strncpyz(mod->texinfo[i+count].texture->name, in->shadername, sizeof(mod->texinfo[i+count].texture->name)); mod->textures[i+count] = mod->texinfo[i+count].texture; } //and for flares, which are not supported at this time. mod->texinfo[count*2].texture = tex+count*2; mod->texinfo[i+count].flags = 0; Q_strncpyz(mod->texinfo[count*2].texture->name, "noshader", sizeof(mod->texinfo[count*2].texture->name)); mod->textures[count*2] = mod->texinfo[count*2].texture; return true; } static qboolean CModQ3_LoadVertexes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; q3dvertex_t *in; vecV_t *out; vec3_t *nout; //, *sout, *tout; int i, count, j; vec2_t *lmout, *stout; vec4_t *cout; extern cvar_t gl_overbright; extern qbyte lmgamma[256]; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "CMOD_LoadVertexes: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > MAX_Q3MAP_VERTEXES) { Con_Printf (CON_ERROR "Map has too many vertexes\n"); return false; } BuildLightMapGammaTable(1, 1<<(2-gl_overbright.ival)); out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); stout = ZG_Malloc(&mod->memgroup, count*sizeof(*stout)); lmout = ZG_Malloc(&mod->memgroup, count*sizeof(*lmout)); cout = ZG_Malloc(&mod->memgroup, count*sizeof(*cout)); nout = ZG_Malloc(&mod->memgroup, count*sizeof(*nout)); // sout = ZG_Malloc(&mod->memgroup, count*sizeof(*nout)); // tout = ZG_Malloc(&mod->memgroup, count*sizeof(*nout)); prv->verts = out; prv->vertstmexcoords = stout; for (i = 0; i < MAXRLIGHTMAPS; i++) { prv->vertlstmexcoords[i] = lmout; prv->colors4f_array[i] = cout; } prv->normals_array = nout; // prv->svector_array = sout; // prv->tvector_array = tout; prv->numvertexes = count; for ( i=0 ; ipoint[j] ); nout[i][j] = LittleFloat (in->normal[j]); } for ( j=0 ; j < 2 ; j++) { stout[i][j] = LittleFloat ( ((float *)in->texcoords)[j] ); lmout[i][j] = LittleFloat ( ((float *)in->texcoords)[j+2] ); } cout[i][0] = (lmgamma[in->color[0]]<color[1]]<color[2]]<color[3]/255.0f; } // if (r_lightmap_saturation.value != 1.0f) // SaturateR8G8B8(cout, count*4, r_lightmap_saturation.value); return true; } #ifdef RFBSPS static qboolean CModRBSP_LoadVertexes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; rbspvertex_t *in; vecV_t *out; vec3_t *nout; //, *sout, *tout; int i, count, j; vec2_t *lmout, *stout; vec4_t *cout; int sty; extern qbyte lmgamma[256]; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "CMOD_LoadVertexes: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > MAX_Q3MAP_VERTEXES) { Con_Printf (CON_ERROR "Map has too many vertexes\n"); return false; } out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); stout = ZG_Malloc(&mod->memgroup, count*sizeof(*stout)); lmout = ZG_Malloc(&mod->memgroup, MAXRLIGHTMAPS*count*sizeof(*lmout)); cout = ZG_Malloc(&mod->memgroup, MAXRLIGHTMAPS*count*sizeof(*cout)); nout = ZG_Malloc(&mod->memgroup, count*sizeof(*nout)); // sout = ZG_Malloc(&mod->memgroup, count*sizeof(*sout)); // tout = ZG_Malloc(&mod->memgroup, count*sizeof(*tout)); prv->verts = out; prv->vertstmexcoords = stout; for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { prv->vertlstmexcoords[sty] = lmout + sty*count; prv->colors4f_array[sty] = cout + sty*count; } prv->normals_array = nout; // prv->svector_array = sout; // prv->tvector_array = tout; prv->numvertexes = count; for ( i=0 ; ipoint[j] ); nout[i][j] = LittleFloat (in->normal[j]); } for ( j=0 ; j < 2 ; j++) { stout[i][j] = LittleFloat (in->stcoords[j]); for (sty = 0; sty < min(MAXRLIGHTMAPS, RBSP_STYLESPERSURF); sty++) prv->vertlstmexcoords[sty][i][j] = LittleFloat(in->lmtexcoords[sty][j]); } for (sty = 0; sty < min(MAXRLIGHTMAPS, RBSP_STYLESPERSURF); sty++) { prv->colors4f_array[sty][i][0] = lmgamma[in->color[sty][0]]/255.0f; prv->colors4f_array[sty][i][1] = lmgamma[in->color[sty][1]]/255.0f; prv->colors4f_array[sty][i][2] = lmgamma[in->color[sty][2]]/255.0f; prv->colors4f_array[sty][i][3] = in->color[sty][3]/255.0f; } } return true; } #endif #ifndef SERVERONLY static qboolean CModQ3_LoadIndexes (model_t *loadmodel, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)loadmodel->meshinfo; int i, count; int *in; index_t *out; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n", loadmodel->name); return false; } count = l->filelen / sizeof(*in); if (count < 1 || count >= MAX_Q3MAP_INDICES) { Con_Printf (CON_ERROR "MOD_LoadBmodel: too many indicies in %s: %i\n", loadmodel->name, count); return false; } out = ZG_Malloc(&loadmodel->memgroup, count*sizeof(*out)); prv->surfindexes = out; // prv->numsurfindexes = count; for ( i=0 ; imeshinfo; q3dface_t *in; q3cface_t *out; int i, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Map has too many faces\n"); return false; } out = BZ_Malloc ( count*sizeof(*out) ); prv->faces = out; prv->numfaces = count; for ( i=0 ; ifacetype = LittleLong ( in->facetype ); out->shadernum = LittleLong ( in->shadernum ); out->numverts = LittleLong ( in->num_vertices ); out->firstvert = LittleLong ( in->firstvertex ); if (out->facetype == MST_PATCH || out->facetype == MST_PATCH_FIXED) { unsigned int pw = LittleLong ( in->patchwidth ); unsigned int ph = LittleLong ( in->patchheight ); out->patch.cp[0] = pw&0xffff; out->patch.cp[1] = ph&0xffff; out->patch.fixedres[0] = pw>>16; out->patch.fixedres[1] = ph>>16; } else { out->soup.firstindex = LittleLong(in->firstindex); out->soup.numindicies = LittleLong(in->num_indexes); } } mod->numsurfaces = i; return true; } #ifdef RFBSPS static qboolean CModRBSP_LoadFaces (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; rbspface_t *in; q3cface_t *out; int i, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Map has too many faces\n"); return false; } out = BZ_Malloc ( count*sizeof(*out) ); prv->faces = out; prv->numfaces = count; for ( i=0 ; ifacetype = LittleLong ( in->facetype ); out->shadernum = LittleLong ( in->shadernum ); out->numverts = LittleLong ( in->num_vertices ); out->firstvert = LittleLong ( in->firstvertex ); if (out->facetype == MST_PATCH || out->facetype == MST_PATCH_FIXED) { unsigned int pw = LittleLong ( in->patchwidth ); unsigned int ph = LittleLong ( in->patchheight ); out->patch.cp[0] = pw&0xffff; out->patch.cp[1] = ph&0xffff; out->patch.fixedres[0] = pw>>16; out->patch.fixedres[1] = ph>>16; } else { out->soup.firstindex = LittleLong(in->firstindex); out->soup.numindicies = LittleLong(in->num_indexes); } } mod->numsurfaces = i; return true; } #endif #ifndef SERVERONLY /* ================= Mod_LoadFogs ================= */ static qboolean CModQ3_LoadFogs (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; dfog_t *in; mfog_t *out; q2cbrush_t *brush; q2cbrushside_t *visibleside, *brushsides; int i, j, count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n", mod->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); mod->fogs = out; mod->numfogs = count; for ( i=0 ; ivisibleSide ) == -1 ) { continue; } brush = prv->brushes + LittleLong ( in->brushNum ); brushsides = brush->brushside; visibleside = brushsides + LittleLong ( in->visibleSide ); out->visibleplane = visibleside->plane; Q_strncpyz(out->shadername, in->shader, sizeof(out->shadername)); out->numplanes = brush->numsides; out->planes = ZG_Malloc(&mod->memgroup, out->numplanes*sizeof(cplane_t *)); for ( j = 0; j < out->numplanes; j++ ) { out->planes[j] = brushsides[j].plane; } } return true; } mfog_t *Mod_FogForOrigin(model_t *wmodel, vec3_t org) { int i, j; mfog_t *ret; float dot; if (!wmodel || wmodel->loadstate != MLS_LOADED) return NULL; for ( i=0 , ret=wmodel->fogs ; inumfogs ; i++, ret++) { if (!ret->shader) continue; for (j = 0; j < ret->numplanes; j++) { dot = DotProduct(ret->planes[j]->normal, org); if (dot - ret->planes[j]->dist > 0) break; } if (j == ret->numplanes) { return ret; } } return NULL; } //Convert a patch in to a list of glpolys #define MAX_ARRAY_VERTS 65535 static index_t tempIndexesArray[MAX_ARRAY_VERTS*6]; static void GL_SizePatchFixed(mesh_t *mesh, int patchwidth, int patchheight, int numverts, int firstvert, cminfo_t *prv) { unsigned short patch_cp[2]; int step[2], size[2]; patch_cp[0] = patchwidth&0xffff; patch_cp[1] = patchheight&0xffff; if (patch_cp[0] <= 0 || patch_cp[1] <= 0 ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } // allocate space for mesh step[0] = patchwidth>>16; step[1] = patchheight>>16; if (!step[0] || !step[1]) { size[0] = patch_cp[0]; size[1] = patch_cp[1]; } else { size[0] = (patch_cp[0] / 2) * step[0] + 1; size[1] = (patch_cp[1] / 2) * step[1] + 1; } mesh->numvertexes = size[0] * size[1]; mesh->numindexes = (size[0]-1) * (size[1]-1) * 6; } static void GL_SizePatch(mesh_t *mesh, int patchwidth, int patchheight, int numverts, int firstvert, cminfo_t *prv) { unsigned short patch_cp[2]; int step[2], size[2], flat[2]; float subdivlevel; patch_cp[0] = patchwidth; patch_cp[1] = patchheight; if (patch_cp[0] <= 0 || patch_cp[1] <= 0 ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } subdivlevel = r_subdivisions.value; if ( subdivlevel < 1 ) subdivlevel = 1; // find the degree of subdivision in the u and v directions Patch_GetFlatness ( subdivlevel, prv->verts[firstvert], sizeof(vecV_t)/sizeof(vec_t), patch_cp, flat ); // allocate space for mesh step[0] = (1 << flat[0]); step[1] = (1 << flat[1]); size[0] = (patch_cp[0] / 2) * step[0] + 1; size[1] = (patch_cp[1] / 2) * step[1] + 1; mesh->numvertexes = size[0] * size[1]; mesh->numindexes = (size[0]-1) * (size[1]-1) * 6; } //mesh_t *GL_CreateMeshForPatch ( model_t *mod, q3dface_t *surf ) static void GL_CreateMeshForPatch (model_t *mod, mesh_t *mesh, int patchwidth, int patchheight, int numverts, int firstvert) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int numindexes, step[2], size[2], flat[2], i, u, v, p; unsigned short patch_cp[2]; index_t *indexes; float subdivlevel; int sty; patch_cp[0] = patchwidth; patch_cp[1] = patchheight; if (patch_cp[0] <= 0 || patch_cp[1] <= 0 ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } subdivlevel = r_subdivisions.value; if ( subdivlevel < 1 ) subdivlevel = 1; // find the degree of subdivision in the u and v directions Patch_GetFlatness ( subdivlevel, prv->verts[firstvert], sizeof(vecV_t)/sizeof(vec_t), patch_cp, flat ); // allocate space for mesh step[0] = (1 << flat[0]); step[1] = (1 << flat[1]); size[0] = (patch_cp[0] / 2) * step[0] + 1; size[1] = (patch_cp[1] / 2) * step[1] + 1; numverts = size[0] * size[1]; if ( numverts < 0 || numverts > MAX_ARRAY_VERTS ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } if (mesh->numvertexes != numverts) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } // fill in Patch_Evaluate ( prv->verts[firstvert], patch_cp, step, mesh->xyz_array[0], sizeof(vecV_t)/sizeof(vec_t)); for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { if (mesh->colors4f_array[sty]) Patch_Evaluate ( prv->colors4f_array[sty][firstvert], patch_cp, step, mesh->colors4f_array[sty][0], 4 ); } Patch_Evaluate ( prv->normals_array[firstvert], patch_cp, step, mesh->normals_array[0], 3 ); Patch_Evaluate ( prv->vertstmexcoords[firstvert], patch_cp, step, mesh->st_array[0], 2 ); for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { if (mesh->lmst_array[sty]) Patch_Evaluate ( prv->vertlstmexcoords[sty][firstvert], patch_cp, step, mesh->lmst_array[sty][0], 2 ); } // compute new indexes avoiding adding invalid triangles numindexes = 0; indexes = tempIndexesArray; for (v = 0, i = 0; v < size[1]-1; v++) { for (u = 0; u < size[0]-1; u++, i += 6) { indexes[0] = p = v * size[0] + u; indexes[1] = p + size[0]; indexes[2] = p + 1; // if ( !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[1]]) && // !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[2]]) && // !VectorEquals(mesh->xyz_array[indexes[1]], mesh->xyz_array[indexes[2]]) ) { indexes += 3; numindexes += 3; } indexes[0] = p + 1; indexes[1] = p + size[0]; indexes[2] = p + size[0] + 1; // if ( !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[1]]) && // !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[2]]) && // !VectorEquals(mesh->xyz_array[indexes[1]], mesh->xyz_array[indexes[2]]) ) { indexes += 3; numindexes += 3; } } } // allocate and fill index table mesh->numindexes = numindexes; memcpy (mesh->indexes, tempIndexesArray, numindexes * sizeof(index_t) ); } static void GL_CreateMeshForPatchFixed (model_t *mod, mesh_t *mesh, int patchwidth, int patchheight, int numverts, int firstvert) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int numindexes, step[2], size[2], i, u, v, p; unsigned short patch_cp[2]; index_t *indexes; float subdivlevel; int sty; patch_cp[0] = patchwidth&0xffff; patch_cp[1] = patchheight&0xffff; if (patch_cp[0] <= 0 || patch_cp[1] <= 0 ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } subdivlevel = r_subdivisions.value; if ( subdivlevel < 1 ) subdivlevel = 1; // allocate space for mesh step[0] = patchwidth>>16; step[1] = patchheight>>16; if (!step[0] || !step[1]) { //explicit CPs only. size[0] = patch_cp[0]; size[1] = patch_cp[1]; } else { size[0] = (patch_cp[0] / 2) * step[0] + 1; size[1] = (patch_cp[1] / 2) * step[1] + 1; } numverts = size[0] * size[1]; if ( numverts < 0 || numverts > MAX_ARRAY_VERTS ) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } if (mesh->numvertexes != numverts) { mesh->numindexes = 0; mesh->numvertexes = 0; return; } // fill in Patch_Evaluate ( prv->verts[firstvert], patch_cp, step, mesh->xyz_array[0], sizeof(vecV_t)/sizeof(vec_t)); for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { if (mesh->colors4f_array[sty]) Patch_Evaluate ( prv->colors4f_array[sty][firstvert], patch_cp, step, mesh->colors4f_array[sty][0], 4 ); } Patch_Evaluate ( prv->normals_array[firstvert], patch_cp, step, mesh->normals_array[0], 3 ); Patch_Evaluate ( prv->vertstmexcoords[firstvert], patch_cp, step, mesh->st_array[0], 2 ); for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { if (mesh->lmst_array[sty]) Patch_Evaluate ( prv->vertlstmexcoords[sty][firstvert], patch_cp, step, mesh->lmst_array[sty][0], 2 ); } // compute new indexes avoiding adding invalid triangles numindexes = 0; indexes = tempIndexesArray; for (v = 0, i = 0; v < size[1]-1; v++) { for (u = 0; u < size[0]-1; u++, i += 6) { indexes[0] = p = v * size[0] + u; indexes[1] = p + size[0]; indexes[2] = p + 1; // if ( !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[1]]) && // !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[2]]) && // !VectorEquals(mesh->xyz_array[indexes[1]], mesh->xyz_array[indexes[2]]) ) { indexes += 3; numindexes += 3; } indexes[0] = p + 1; indexes[1] = p + size[0]; indexes[2] = p + size[0] + 1; // if ( !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[1]]) && // !VectorEquals(mesh->xyz_array[indexes[0]], mesh->xyz_array[indexes[2]]) && // !VectorEquals(mesh->xyz_array[indexes[1]], mesh->xyz_array[indexes[2]]) ) { indexes += 3; numindexes += 3; } } } // allocate and fill index table mesh->numindexes = numindexes; memcpy (mesh->indexes, tempIndexesArray, numindexes * sizeof(index_t) ); } #ifdef RFBSPS static void CModRBSP_BuildSurfMesh(model_t *mod, msurface_t *out, builddata_t *bd) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; rbspface_t *in = (rbspface_t*)(bd+1); int idx = (out - mod->surfaces) - mod->firstmodelsurface; int sty; in += idx; if (LittleLong(in->facetype) == MST_PATCH) { GL_CreateMeshForPatch(mod, out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex)); } else if (LittleLong(in->facetype) == MST_PATCH_FIXED) { GL_CreateMeshForPatchFixed(mod, out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex)); } else if (LittleLong(in->facetype) == MST_PLANAR || LittleLong(in->facetype) == MST_TRIANGLE_SOUP) { unsigned int fv = LittleLong(in->firstvertex), i; for (i = 0; i < out->mesh->numvertexes; i++) { VectorCopy(prv->verts[fv + i], out->mesh->xyz_array[i]); Vector2Copy(prv->vertstmexcoords[fv + i], out->mesh->st_array[i]); for (sty = 0; sty < MAXRLIGHTMAPS; sty++) { Vector2Copy(prv->vertlstmexcoords[sty][fv + i], out->mesh->lmst_array[sty][i]); Vector4Copy(prv->colors4f_array[sty][fv + i], out->mesh->colors4f_array[sty][i]); } VectorCopy(prv->normals_array[fv + i], out->mesh->normals_array[i]); } fv = LittleLong(in->firstindex); for (i = 0; i < out->mesh->numindexes; i++) { out->mesh->indexes[i] = prv->surfindexes[fv + i]; } } else { /* //flare int r, g, b; extern index_t r_quad_indexes[6]; static vec2_t st[4] = {{0,0},{0,1},{1,1},{1,0}}; mesh = out->mesh = (mesh_t *)Hunk_Alloc(sizeof(mesh_t)); mesh->xyz_array = (vecV_t *)Hunk_Alloc(sizeof(vecV_t)*4); mesh->colors4b_array = (byte_vec4_t *)Hunk_Alloc(sizeof(byte_vec4_t)*4); mesh->numvertexes = 4; mesh->indexes = r_quad_indexes; mesh->st_array = st; mesh->numindexes = 6; VectorCopy (in->lightmap_origin, mesh->xyz_array[0]); VectorCopy (in->lightmap_origin, mesh->xyz_array[1]); VectorCopy (in->lightmap_origin, mesh->xyz_array[2]); VectorCopy (in->lightmap_origin, mesh->xyz_array[3]); r = LittleFloat(in->lightmap_vecs[0][0]) * 255.0f; r = bound (0, r, 255); g = LittleFloat(in->lightmap_vecs[0][1]) * 255.0f; g = bound (0, g, 255); b = LittleFloat(in->lightmap_vecs[0][2]) * 255.0f; b = bound (0, b, 255); mesh->colors4b_array[0][0] = r; mesh->colors4b_array[0][1] = g; mesh->colors4b_array[0][2] = b; mesh->colors4b_array[0][3] = 255; Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[1]); Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[2]); Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[3]); */ } Mod_AccumulateMeshTextureVectors(out->mesh); Mod_NormaliseTextureVectors(out->mesh->normals_array, out->mesh->snormals_array, out->mesh->tnormals_array, out->mesh->numvertexes, false); } #endif static void CModQ3_BuildSurfMesh(model_t *mod, msurface_t *out, builddata_t *bd) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int idx = (out - mod->surfaces) - mod->firstmodelsurface; q3dface_t *in = (q3dface_t*)(bd+1) + idx; int facetype = LittleLong(in->facetype); if (facetype == MST_PATCH) { GL_CreateMeshForPatch(mod, out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex)); } else if (facetype == MST_PATCH_FIXED) { GL_CreateMeshForPatchFixed(mod, out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex)); } else if (facetype == MST_PLANAR || facetype == MST_TRIANGLE_SOUP) { unsigned int fv = LittleLong(in->firstvertex), fi = LittleLong(in->firstindex), i; for (i = 0; i < out->mesh->numvertexes; i++) { VectorCopy(prv->verts[fv + i], out->mesh->xyz_array[i]); Vector2Copy(prv->vertstmexcoords[fv + i], out->mesh->st_array[i]); Vector2Copy(prv->vertlstmexcoords[0][fv + i], out->mesh->lmst_array[0][i]); Vector4Copy(prv->colors4f_array[0][fv + i], out->mesh->colors4f_array[0][i]); VectorCopy(prv->normals_array[fv + i], out->mesh->normals_array[i]); } for (i = 0; i < out->mesh->numindexes; i++) { out->mesh->indexes[i] = prv->surfindexes[fi + i]; } } else { /* //flare int r, g, b; extern index_t r_quad_indexes[6]; static vec2_t st[4] = {{0,0},{0,1},{1,1},{1,0}}; mesh = out->mesh = (mesh_t *)Hunk_Alloc(sizeof(mesh_t)); mesh->xyz_array = (vecV_t *)Hunk_Alloc(sizeof(vecV_t)*4); mesh->colors4b_array = (byte_vec4_t *)Hunk_Alloc(sizeof(byte_vec4_t)*4); mesh->numvertexes = 4; mesh->indexes = r_quad_indexes; mesh->st_array = st; mesh->numindexes = 6; VectorCopy (in->lightmap_origin, mesh->xyz_array[0]); VectorCopy (in->lightmap_origin, mesh->xyz_array[1]); VectorCopy (in->lightmap_origin, mesh->xyz_array[2]); VectorCopy (in->lightmap_origin, mesh->xyz_array[3]); r = LittleFloat(in->lightmap_vecs[0][0]) * 255.0f; r = bound (0, r, 255); g = LittleFloat(in->lightmap_vecs[0][1]) * 255.0f; g = bound (0, g, 255); b = LittleFloat(in->lightmap_vecs[0][2]) * 255.0f; b = bound (0, b, 255); mesh->colors4b_array[0][0] = r; mesh->colors4b_array[0][1] = g; mesh->colors4b_array[0][2] = b; mesh->colors4b_array[0][3] = 255; Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[1]); Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[2]); Vector4Copy(mesh->colors4b_array[0], mesh->colors4b_array[3]); */ } Mod_AccumulateMeshTextureVectors(out->mesh); Mod_NormaliseTextureVectors(out->mesh->normals_array, out->mesh->snormals_array, out->mesh->tnormals_array, out->mesh->numvertexes, false); } static qboolean CModQ3_LoadRFaces (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; extern cvar_t r_vertexlight; q3dface_t *in; msurface_t *out; mplane_t *pl; int facetype; int count; int surfnum; int fv; int sty; mesh_t *mesh; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",mod->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); pl = ZG_Malloc(&mod->memgroup, count*sizeof(*pl));//create a new array of planes for speed. mesh = ZG_Malloc(&mod->memgroup, count*sizeof(*mesh)); mod->surfaces = out; mod->numsurfaces = count; mod->lightmaps.first = 0; for (surfnum = 0; surfnum < count; surfnum++, out++, in++, pl++) { out->plane = pl; facetype = LittleLong(in->facetype); out->texinfo = mod->texinfo + LittleLong(in->shadernum); out->lightmaptexturenums[0] = LittleLong(in->lightmapnum); if (facetype == MST_FLARE) out->texinfo = mod->texinfo + mod->numtexinfo*2; else if (out->lightmaptexturenums[0] < 0 /*|| facetype == MST_TRIANGLE_SOUP*/ || r_vertexlight.value) out->texinfo += mod->numtexinfo; //various surfaces use a different version of the same shader (with all the lightmaps collapsed) out->light_s[0] = LittleLong(in->lightmap_offs[0]); out->light_t[0] = LittleLong(in->lightmap_offs[1]); out->styles[0] = INVALID_LIGHTSTYLE; out->vlstyles[0] = INVALID_VLIGHTSTYLE; for (sty = 1; sty < MAXRLIGHTMAPS; sty++) { out->styles[sty] = INVALID_LIGHTSTYLE; out->vlstyles[sty] = INVALID_VLIGHTSTYLE; out->lightmaptexturenums[sty] = -1; } for (; sty < MAXCPULIGHTMAPS; sty++) out->styles[sty] = INVALID_LIGHTSTYLE; out->lmshift = LMSHIFT_DEFAULT; //fixme: determine texturemins from lightmap_origin out->extents[0] = (LittleLong(in->lightmap_width)-1)<lmshift; out->extents[1] = (LittleLong(in->lightmap_height)-1)<lmshift; out->samples=NULL; if (mod->lightmaps.count < out->lightmaptexturenums[0]+1) mod->lightmaps.count = out->lightmaptexturenums[0]+1; fv = LittleLong(in->firstvertex); { vec3_t v[3]; VectorCopy(prv->verts[fv+0], v[0]); VectorCopy(prv->verts[fv+1], v[1]); VectorCopy(prv->verts[fv+2], v[2]); PlaneFromPoints(v, pl); CategorizePlane(pl); } if (prv->surfaces[LittleLong(in->shadernum)].c.value == 0 || prv->surfaces[LittleLong(in->shadernum)].c.value & Q3CONTENTS_TRANSLUCENT) //q3dm10's thingie is 0 out->flags |= SURF_DRAWALPHA; if (mod->texinfo[LittleLong(in->shadernum)].flags & TI_SKY) out->flags |= SURF_DRAWSKY; if (LittleLong(in->fognum) == -1 || !mod->numfogs) out->fog = NULL; else out->fog = mod->fogs + LittleLong(in->fognum); if (prv->surfaces[LittleLong(in->shadernum)].c.flags & (Q3SURF_NODRAW | Q3SURF_SKIP)) { out->mesh = &mesh[surfnum]; out->mesh->numindexes = 0; out->mesh->numvertexes = 0; } else if (facetype == MST_PATCH) { out->mesh = &mesh[surfnum]; GL_SizePatch(out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex), prv); } else if (facetype == MST_PATCH_FIXED) { out->mesh = &mesh[surfnum]; GL_SizePatchFixed(out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex), prv); } else if (facetype == MST_PLANAR || facetype == MST_TRIANGLE_SOUP) { out->mesh = &mesh[surfnum]; out->mesh->numindexes = LittleLong(in->num_indexes); out->mesh->numvertexes = LittleLong(in->num_vertices); /* Mod_AccumulateMeshTextureVectors(out->mesh); */ } else { out->mesh = &mesh[surfnum]; out->mesh->numindexes = 6; out->mesh->numvertexes = 4; } } Mod_SortShaders(mod); return true; } #ifdef RFBSPS static qboolean CModRBSP_LoadRFaces (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; extern cvar_t r_vertexlight; rbspface_t *in; msurface_t *out; mplane_t *pl; int facetype; int count; int surfnum; int fv; int j; mesh_t *mesh; int maxstyle = q3bsp_ignorestyles.ival?1:min(MAXRLIGHTMAPS, RBSP_STYLESPERSURF); in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",mod->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&mod->memgroup, count*sizeof(*out)); pl = ZG_Malloc(&mod->memgroup, count*sizeof(*pl));//create a new array of planes for speed. mesh = ZG_Malloc(&mod->memgroup, count*sizeof(*mesh)); mod->surfaces = out; mod->numsurfaces = count; for (surfnum = 0; surfnum < count; surfnum++, out++, in++, pl++) { out->plane = pl; facetype = LittleLong(in->facetype); out->texinfo = mod->texinfo + LittleLong(in->shadernum); for (j = 0; j < maxstyle; j++) { out->lightmaptexturenums[j] = LittleLong(in->lightmapnum[j]); out->light_s[j] = LittleLong(in->lightmap_offs[0][j]); out->light_t[j] = LittleLong(in->lightmap_offs[1][j]); out->styles[j] = (in->lm_styles[j]!=255)?in->lm_styles[j]:INVALID_LIGHTSTYLE; out->vlstyles[j] = (in->vt_styles[j]!=255)?in->vt_styles[j]:INVALID_VLIGHTSTYLE; if (mod->lightmaps.count < out->lightmaptexturenums[j]+1) mod->lightmaps.count = out->lightmaptexturenums[j]+1; } for (; j < MAXRLIGHTMAPS; j++) { out->lightmaptexturenums[j] = -1; out->light_s[j] = 0; out->light_t[j] = 0; out->styles[j] = INVALID_LIGHTSTYLE; out->vlstyles[j] = INVALID_VLIGHTSTYLE; } for (; j < MAXCPULIGHTMAPS; j++) out->styles[j] = INVALID_LIGHTSTYLE; if (facetype == MST_FLARE) out->texinfo = mod->texinfo + mod->numtexinfo*2; else if (out->lightmaptexturenums[0]<0 || r_vertexlight.value) out->texinfo += mod->numtexinfo; //soup/vertex light uses a different version of the same shader (with all the lightmaps collapsed) out->lmshift = LMSHIFT_DEFAULT; out->extents[0] = (LittleLong(in->lightmap_width)-1)<lmshift; out->extents[1] = (LittleLong(in->lightmap_height)-1)<lmshift; out->samples=NULL; fv = LittleLong(in->firstvertex); { vec3_t v[3]; VectorCopy(prv->verts[fv+0], v[0]); VectorCopy(prv->verts[fv+1], v[1]); VectorCopy(prv->verts[fv+2], v[2]); PlaneFromPoints(v, pl); CategorizePlane(pl); } if (prv->surfaces[in->shadernum].c.value == 0 || prv->surfaces[in->shadernum].c.value & Q3CONTENTS_TRANSLUCENT) //q3dm10's thingie is 0 out->flags |= SURF_DRAWALPHA; if (mod->texinfo[in->shadernum].flags & TI_SKY) out->flags |= SURF_DRAWSKY; if (in->fognum < 0 || in->fognum >= mod->numfogs) out->fog = NULL; else out->fog = mod->fogs + in->fognum; if (prv->surfaces[LittleLong(in->shadernum)].c.flags & (Q3SURF_NODRAW | Q3SURF_SKIP)) { out->mesh = &mesh[surfnum]; out->mesh->numindexes = 0; out->mesh->numvertexes = 0; } else if (facetype == MST_PATCH) { out->mesh = &mesh[surfnum]; GL_SizePatch(out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex), prv); } else if (facetype == MST_PATCH_FIXED) { out->mesh = &mesh[surfnum]; GL_SizePatchFixed(out->mesh, LittleLong(in->patchwidth), LittleLong(in->patchheight), LittleLong(in->num_vertices), LittleLong(in->firstvertex), prv); } else if (facetype == MST_PLANAR || facetype == MST_TRIANGLE_SOUP) { out->mesh = &mesh[surfnum]; out->mesh->numindexes = LittleLong(in->num_indexes); out->mesh->numvertexes = LittleLong(in->num_vertices); /* Mod_AccumulateMeshTextureVectors(out->mesh); */ } else { out->mesh = &mesh[surfnum]; out->mesh->numindexes = 6; out->mesh->numvertexes = 4; } } Mod_SortShaders(mod); return true; } #endif #endif static qboolean CModQ3_LoadNodes (model_t *loadmodel, qbyte *mod_base, lump_t *l) { int i, j, count, p; q3dnode_t *in; mnode_t *out; //dnode_t in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",loadmodel->name); return false; } count = l->filelen / sizeof(*in); out = ZG_Malloc(&loadmodel->memgroup, count*sizeof(*out)); if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Too many nodes on map\n"); return false; } loadmodel->nodes = out; loadmodel->numnodes = count; for ( i=0 ; iminmaxs[j] = LittleLong (in->mins[j]); out->minmaxs[3+j] = LittleLong (in->maxs[j]); } AddPointToBounds(out->minmaxs, loadmodel->mins, loadmodel->maxs); AddPointToBounds(out->minmaxs+3, loadmodel->mins, loadmodel->maxs); p = LittleLong(in->plane); out->plane = loadmodel->planes + p; out->firstsurface = 0;//LittleShort (in->firstface); out->numsurfaces = 0;//LittleShort (in->numfaces); out->contents = -1; for (j=0 ; j<2 ; j++) { p = LittleLong (in->children[j]); out->childnum[j] = p; if (p >= 0) { out->children[j] = loadmodel->nodes + p; } else out->children[j] = (mnode_t *)(loadmodel->leafs + (-1 - p)); } } CMod_SetParent (loadmodel->nodes, NULL); // sets nodes and leafs return true; } static qboolean CModQ3_LoadBrushes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; q3dbrush_t *in; q2cbrush_t *out; int i, count; int shaderref; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count > SANITY_MAX_MAP_BRUSHES) { Con_Printf (CON_ERROR "Map has too many brushes"); return false; } prv->brushes = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+1)); out = prv->brushes; prv->numbrushes = count; for (i=0 ; ishadernum ); out->contents = prv->surfaces[shaderref].c.value; out->brushside = &prv->brushsides[LittleLong ( in->firstside )]; out->numsides = LittleLong ( in->num_sides ); CM_FinalizeBrush(out); } return true; } static qboolean CModQ3_LoadLeafs (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i, j; mleaf_t *out; q3dleaf_t *in; int count; q2cbrush_t **brush; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no leafs\n"); return false; } // need to save space for box planes if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Too many leaves on map"); return false; } out = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+1)); mod->leafs = out; mod->numleafs = count; for ( i=0 ; iminmaxs[0+j] = LittleLong(in->mins[j]); out->minmaxs[3+j] = LittleLong(in->maxs[j]); } out->cluster = LittleLong(in->cluster); out->area = LittleLong(in->area); // out->firstleafface = LittleLong(in->firstleafsurface); // out->numleaffaces = LittleLong(in->num_leafsurfaces); out->contents = 0; out->firstleafbrush = LittleLong(in->firstleafbrush); out->numleafbrushes = LittleLong(in->num_leafbrushes); out->firstmarksurface = mod->marksurfaces + LittleLong(in->firstleafsurface); out->nummarksurfaces = LittleLong(in->num_leafsurfaces); if (out->minmaxs[0] > out->minmaxs[3+0] || out->minmaxs[1] > out->minmaxs[3+1] || out->minmaxs[2] > out->minmaxs[3+2])// || VectorEquals (out->minmaxs, out->minmaxs+3)) { out->nummarksurfaces = 0; } brush = &prv->leafbrushes[out->firstleafbrush]; for (j=0 ; jnumleafbrushes ; j++) { out->contents |= brush[j]->contents; } if (out->area >= prv->numareas) { prv->numareas = out->area + 1; } } return true; } static qboolean CModQ3_LoadPlanes (model_t *loadmodel, qbyte *mod_base, lump_t *l) { int i, j; mplane_t *out; Q3PLANE_t *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",loadmodel->name); return false; } count = l->filelen / sizeof(*in); if (count > SANITY_LIMIT(*out)) { Con_Printf (CON_ERROR "Too many planes on map (%i)\n", count); return false; } loadmodel->planes = out = ZG_Malloc(&loadmodel->memgroup, sizeof(*out) * count); loadmodel->numplanes = count; for ( i=0 ; inormal[j] = LittleFloat (in->n[j]); } out->dist = LittleFloat (in->d); CategorizePlane(out); } return true; } static qboolean CModQ3_LoadLeafBrushes (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; q2cbrush_t **out; int *in; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); if (count < 1) { Con_Printf (CON_ERROR "Map with no leafbrushes\n"); return false; } // need to save space for box planes if (count > SANITY_MAX_MAP_LEAFBRUSHES) { Con_Printf (CON_ERROR "Map has too many leafbrushes\n"); return false; } //prv->numbrushes is because of submodels being weird. out = prv->leafbrushes = ZG_Malloc(&mod->memgroup, sizeof(*out) * (count+prv->numbrushes)); prv->numleafbrushes = count; for ( i=0 ; ibrushes + (unsigned int)LittleLong (*in); return true; } static qboolean CModQ3_LoadBrushSides (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i, j; q2cbrushside_t *out; q3dbrushside_t *in; int count; int num; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); // need to save space for box planes if (count > SANITY_MAX_MAP_BRUSHSIDES) { Con_Printf (CON_ERROR "Map has too many brushsides (%i)\n", count); return false; } out = prv->brushsides = ZG_Malloc(&mod->memgroup, sizeof(*out) * count); prv->numbrushsides = count; for ( i=0 ; iplanenum); out->plane = &mod->planes[num]; j = LittleLong (in->texinfo); if (j >= mod->numtexinfo) { Con_Printf (CON_ERROR "Bad brushside texinfo\n"); return false; } out->surface = &prv->surfaces[j]; } return true; } #ifdef RFBSPS static qboolean CModRBSP_LoadBrushSides (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i, j; q2cbrushside_t *out; rbspbrushside_t *in; int count; int num; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size\n"); return false; } count = l->filelen / sizeof(*in); // need to save space for box planes if (count > SANITY_MAX_MAP_BRUSHSIDES) { Con_Printf (CON_ERROR "Map has too many brushsides (%i)\n", count); return false; } out = prv->brushsides = ZG_Malloc(&mod->memgroup, sizeof(*out) * count); prv->numbrushsides = count; for ( i=0 ; iplanenum); out->plane = &mod->planes[num]; j = LittleLong (in->texinfo); if (j >= mod->numtexinfo) { Con_Printf (CON_ERROR "Bad brushside texinfo\n"); return false; } out->surface = &prv->surfaces[j]; } return true; } #endif static qboolean CModQ3_LoadVisibility (model_t *mod, qbyte *mod_base, lump_t *l) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; unsigned int numclusters; if (l->filelen == 0) { int i; #if 0 //the 'correct' code numclusters = 0; for (i = 0; i < mod->numleafs; i++) if (numclusters < mod->leafs[i].cluster+1) numclusters = mod->leafs[i].cluster+1; numclusters++; #else //but its much faster to merge all leafs into a single pvs cluster. no vis is no vis. numclusters = 8*sizeof(int); for (i = 0; i < mod->numleafs; i++) mod->leafs[i].cluster = !!mod->leafs[i].cluster; #endif prv->q3pvs = ZG_Malloc(&mod->memgroup, sizeof(*prv->q3pvs) + (numclusters+7)/8 * numclusters); memset (prv->q3pvs, 0xff, sizeof(*prv->q3pvs) + (numclusters+7)/8 * numclusters); prv->q3pvs->numclusters = numclusters; prv->numvisibility = 0; prv->q3pvs->rowsize = (prv->q3pvs->numclusters+7)/8; } else { prv->numvisibility = l->filelen; prv->q3pvs = ZG_Malloc(&mod->memgroup, l->filelen); mod->vis = (q2dvis_t *)prv->q3pvs; memcpy (prv->q3pvs, mod_base + l->fileofs, l->filelen); numclusters = prv->q3pvs->numclusters = LittleLong (prv->q3pvs->numclusters); prv->q3pvs->rowsize = LittleLong (prv->q3pvs->rowsize); } mod->numclusters = numclusters; mod->pvsbytes = ((mod->numclusters + 31)>>3)&~3; return true; } #ifndef SERVERONLY static void CModQ3_LoadLighting (model_t *loadmodel, qbyte *mod_base, lump_t *l) { qbyte *in = mod_base + l->fileofs; qbyte *out; unsigned int samples = l->filelen; int m, s, t; int mapstride = loadmodel->lightmaps.width*3; int mapsize = mapstride*loadmodel->lightmaps.height; int maps; int merge; int mergestride; extern cvar_t gl_overbright; float scale = (1<<(2-gl_overbright.ival)); loadmodel->lightmaps.fmt = LM_L8; //round up the samples, in case the last one is partial. maps = ((samples+mapsize-1)&~(mapsize-1)) / mapsize; //q3 maps have built in 4-fold overbright. //if we're not rendering with that, we need to brighten the lightmaps in order to keep the darker parts the same brightness. we loose the 2 upper bits. those bright areas become uniform and indistinct. gl_overbright.flags |= CVAR_RENDERERLATCH; BuildLightMapGammaTable(1, (1<<(2-gl_overbright.ival))); loadmodel->lightmaps.mergew = 0; loadmodel->lightmaps.mergeh = 0; loadmodel->engineflags |= MDLF_NEEDOVERBRIGHT; if (!samples) return; loadmodel->lightmaps.fmt = LM_RGB8; if (loadmodel->lightmaps.deluxemapping) maps /= 2; { int limitw = sh_config.texture2d_maxsize / loadmodel->lightmaps.width; int limith = sh_config.texture2d_maxsize / loadmodel->lightmaps.height; if (!q3bsp_mergeq3lightmaps.ival) { limitw = 1; limith = 1; } loadmodel->lightmaps.mergeh = loadmodel->lightmaps.mergew = 1; while (loadmodel->lightmaps.mergew*loadmodel->lightmaps.mergeh < maps) { //this could probably be smarter. if (loadmodel->lightmaps.mergew*2 <= limitw && loadmodel->lightmaps.mergew < loadmodel->lightmaps.mergeh) loadmodel->lightmaps.mergew *= 2; else if (loadmodel->lightmaps.mergeh*2 <= limith) loadmodel->lightmaps.mergeh *= 2; else if (loadmodel->lightmaps.mergew*2 <= limitw) loadmodel->lightmaps.mergew *= 2; else break; //can't expand in either direction. } } merge = loadmodel->lightmaps.mergew*loadmodel->lightmaps.mergeh; mergestride = loadmodel->lightmaps.mergew*mapstride; //q3bsp itself does not support deluxemapping. //the way it works is by interleaving the data in lightmap+deluxemap pairs. //the surface data makes no references to the deluxemap maps, they're implied by lightmap+1 //if no surface references an odd lightmap index then we know we have deluxemaps... assuming there are at least two lightmaps. //q3map2 likes generating null lightmaps, so beware false positives. //note that external lighting makes this even more fun. //if we have deluxemapping data then we split it here. beware externals. if (loadmodel->lightmaps.deluxemapping) { m = merge; while (m < maps) m += merge; loadmodel->lightdata = ZG_Malloc(&loadmodel->memgroup, mapsize*m*2); loadmodel->lightdatasize = mapsize*m*2; } else { m = merge; while (m < maps) m += merge; loadmodel->lightdata = ZG_Malloc(&loadmodel->memgroup, mapsize*m); loadmodel->lightdatasize = mapsize*m; } if (!loadmodel->lightdata) return; //be careful here, q3bsp deluxemapping is done using interleaving. we want to unoverbright ONLY lightmaps and not deluxemaps. for (m = 0; m < maps; m++) { out = loadmodel->lightdata; //figure out which merged lightmap we're putting it into out += (m/merge)*merge*mapsize * (loadmodel->lightmaps.deluxemapping?2:1); //and the submap s = m%merge; t = s/loadmodel->lightmaps.mergew; s = s%loadmodel->lightmaps.mergew; out += s*mapstride; out += t*mergestride*loadmodel->lightmaps.height; //q3bsp has 4-fold overbrights, so if we're not using overbrights then we basically need to scale the values up by 4 //this will require clamping, which can result in oversaturation of channels, meaning discolouration for (t = 0; t < loadmodel->lightmaps.height; t++) { for (s = 0; s < loadmodel->lightmaps.width; s++) { float i; vec3_t l; l[0] = *in++; l[1] = *in++; l[2] = *in++; VectorScale(l, scale, l); //it should be noted that this maths is wrong if you're trying to use srgb lightmaps. i = max(l[0], max(l[1], l[2])); if (i > 255) VectorScale(l, 255/i, l); //clamp the brightest channel, scaling the others down to retain chromiance. *out++ = l[0]; *out++ = l[1]; *out++ = l[2]; } out += mergestride-mapstride; } if (r_lightmap_saturation.value != 1.0f) SaturateR8G8B8(out, mapsize, r_lightmap_saturation.value); if (loadmodel->lightmaps.deluxemapping) { out -= mergestride*loadmodel->lightmaps.height; out += merge*mapsize; //no gamma for deluxemap for (t = 0; t < loadmodel->lightmaps.height; t++) { for (s = 0; s < loadmodel->lightmaps.width; s++) { *out++ = in[0]; *out++ = in[1]; *out++ = in[2]; in += 3; } out += mergestride-mapstride; } } } /*for (; m%merge; m++) { out = loadmodel->lightdata; //figure out which merged lightmap we're putting it into out += (m/merge)*merge*mapsize * (loadmodel->lightmaps.deluxemapping?2:1); //and the submap out += (m%merge)*mapsize; for(s = 0; s < mapsize; s+=3) { out[s+0] = 0; out[s+1] = 255; out[s+2] = 0; } }*/ } static qboolean CModQ3_LoadLightgrid (model_t *loadmodel, qbyte *mod_base, lump_t *l) { dq3gridlight_t *in; dq3gridlight_t *out; q3lightgridinfo_t *grid; int count; in = (void *)(mod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",loadmodel->name); return false; } count = l->filelen / sizeof(*in); grid = ZG_Malloc(&loadmodel->memgroup, sizeof(q3lightgridinfo_t) + count*sizeof(*out)); grid->lightgrid = (dq3gridlight_t*)(grid+1); out = grid->lightgrid; loadmodel->lightgrid = grid; grid->numlightgridelems = count; // lightgrid is all 8 bit memcpy ( out, in, count*sizeof(*out) ); return true; } #ifdef RFBSPS static qboolean CModRBSP_LoadLightgrid (model_t *loadmodel, qbyte *mod_base, lump_t *elements, lump_t *indexes) { unsigned short *iin; rbspgridlight_t *ein; unsigned short *iout; rbspgridlight_t *eout; q3lightgridinfo_t *grid; int ecount; int icount; int i; ein = (void *)(mod_base + elements->fileofs); iin = (void *)(mod_base + indexes->fileofs); if (indexes->filelen % sizeof(*iin) || elements->filelen % sizeof(*ein)) { Con_Printf (CON_ERROR "MOD_LoadBmodel: funny lump size in %s\n",loadmodel->name); return false; } icount = indexes->filelen / sizeof(*iin); ecount = elements->filelen / sizeof(*ein); grid = ZG_Malloc(&loadmodel->memgroup, sizeof(q3lightgridinfo_t) + ecount*sizeof(*eout) + icount*sizeof(*iout)); grid->rbspelements = (rbspgridlight_t*)((char *)grid + sizeof(q3lightgridinfo_t)); grid->rbspindexes = (unsigned short*)((char *)grid + sizeof(q3lightgridinfo_t) + ecount*sizeof(*eout)); eout = grid->rbspelements; iout = grid->rbspindexes; loadmodel->lightgrid = grid; grid->numlightgridelems = icount; // elements are all 8 bit memcpy ( eout, ein, ecount*sizeof(*eout) ); for (i = 0; i < icount; i++) iout[i] = LittleShort(iin[i]); return true; } #endif #endif #endif #if !defined(SERVERONLY) && defined(Q2BSPS) qbyte *ReadPCXPalette(qbyte *buf, int len, qbyte *out); static int CM_GetQ2Palette (void) { char *f; size_t sz; sz = FS_LoadFile("pics/colormap.pcx", (void**)&f); if (!f) { Con_Printf (CON_WARNING "Couldn't find pics/colormap.pcx\n"); return -1; } if (!ReadPCXPalette(f, sz, d_q28to24table)) { Con_Printf (CON_WARNING "Couldn't read pics/colormap.pcx\n"); FS_FreeFile(f); return -1; } FS_FreeFile(f); #if 1 { float inf; qbyte palette[768]; qbyte *pal; int i; pal = d_q28to24table; for (i=0 ; i<768 ; i++) { inf = ((pal[i]+1)/256.0)*255 + 0.5; if (inf < 0) inf = 0; if (inf > 255) inf = 255; palette[i] = inf; } memcpy (d_q28to24table, palette, sizeof(palette)); } #endif return 0; } #endif #if 0 static void CM_OpenAllPortals(model_t *mod, char *ents) //this is a compleate hack. About as compleate as possible. { //q2 levels contain a thingie called area portals. Basically, doors can seperate two areas and //the engine knows when this portal is open, and weather to send ents from both sides of the door //or not. It's not just ents, but also walls. We want to just open them by default and hope the //progs knows how to close them. char style[8]; char name[64]; if (!map_autoopenportals.value) return; while(*ents) { if (*ents == '{') //an entity { ents++; *style = '\0'; *name = '\0'; while (*ents) { ents = COM_Parse(ents); if (!strcmp(com_token, "classname")) { ents = COM_ParseOut(ents, name, sizeof(name)); } else if (!strcmp(com_token, "style")) { ents = COM_ParseOut(ents, style, sizeof(style)); } else if (*com_token == '}') break; else ents = COM_Parse(ents); //other field ents++; } if (!strcmp(name, "func_areaportal")) { CMQ2_SetAreaPortalState(mod, atoi(style), true); } } ents++; } } #endif #if !defined(CLIENTONLY) && defined(Q3BSPS) static void CalcClusterPHS(cminfo_t *prv, int cluster) { int j, k, l, index; int bitbyte; unsigned int *dest, *src; qbyte *scan; int numclusters = prv->q3pvs->numclusters; int rowbytes = prv->q3pvs->rowsize; int rowwords = rowbytes / sizeof(int); scan = (qbyte *)prv->q3pvs->data; dest = (unsigned int *)(prv->q3phs->data); dest += rowwords*cluster; scan += rowbytes*cluster; for (j=0 ; j= numclusters) { // if (!buggytools) // Con_Printf ("CM_CalcPHS: Bad bit(s) in PVS (%i >= %i)\n", index, numclusters); // pad bits should be 0 // buggytools = true; } else { src = (unsigned int *)(prv->q3pvs->data) + index*rowwords; for (l=0 ; lphscalced[cluster>>3] |= 1<<(cluster&7); } static void CMQ3_CalcPHS (model_t *mod) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int rowbytes, rowwords; int i, j, k, l, index; int bitbyte; unsigned int *dest, *src; qbyte *scan; int count, vcount; int numclusters; qboolean buggytools = false; extern cvar_t sv_calcphs; Con_DPrintf ("Building PHS...\n"); prv->q3phs = ZG_Malloc(&mod->memgroup, sizeof(*prv->q3phs) + prv->q3pvs->rowsize * prv->q3pvs->numclusters); rowwords = prv->q3pvs->rowsize / sizeof(int); rowbytes = prv->q3pvs->rowsize; memset ( prv->q3phs, 0, sizeof(*prv->q3phs) + prv->q3pvs->rowsize * prv->q3pvs->numclusters ); prv->q3phs->rowsize = prv->q3pvs->rowsize; prv->q3phs->numclusters = numclusters = prv->q3pvs->numclusters; if (!numclusters) return; vcount = 0; for (i=0 ; i>3] & (1<<(j&7)) ) { vcount++; } } } count = 0; scan = (qbyte *)prv->q3pvs->data; dest = (unsigned int *)(prv->q3phs->data); if (sv_calcphs.ival >= 2) { //delay-calculate it. prv->phscalced = ZG_Malloc(&mod->memgroup, (prv->q3pvs->numclusters+7)/8); memcpy(dest, scan, rowbytes*numclusters); Con_DPrintf ("Average clusters visible / total: %i / %i\n" , vcount/numclusters, numclusters); } else if (!sv_calcphs.ival) { //disable calcs (behaves like broadcast so just fill with 1s) memset(dest, 0xff, rowbytes*numclusters); } else { //the original slow logic like q3. for (i=0 ; i= numclusters) { if (!buggytools) Con_Printf ("CM_CalcPHS: Bad bit(s) in PVS (%i >= %i)\n", index, numclusters); // pad bits should be 0 buggytools = true; } else { src = (unsigned int *)(prv->q3pvs->data) + index*rowwords; for (l=0 ; l>3] & (1<<(j&7)) ) count++; } Con_DPrintf ("Average clusters visible / hearable / total: %i / %i / %i\n" , vcount/numclusters, count/numclusters, numclusters); } } #endif /* static qbyte *CM_LeafnumPVS (model_t *model, int leafnum, qbyte *buffer, unsigned int buffersize) { return CM_ClusterPVS(model, CM_LeafCluster(model, leafnum), buffer, buffersize); } */ #ifndef SERVERONLY #define GLQ2BSP_LightPointValues GLQ1BSP_LightPointValues extern int r_dlightframecount; static void Q2BSP_MarkLights (dlight_t *light, dlightbitmask_t bit, mnode_t *node) { mplane_t *splitplane; float dist; msurface_t *surf; int i; if (node->contents != -1) { mleaf_t *leaf = (mleaf_t *)node; msurface_t **mark; i = leaf->nummarksurfaces; mark = leaf->firstmarksurface; while(i--!=0) { surf = *mark++; if (surf->dlightframe != r_dlightframecount) { surf->dlightbits = 0; surf->dlightframe = r_dlightframecount; } surf->dlightbits |= bit; } return; } splitplane = node->plane; dist = DotProduct (light->origin, splitplane->normal) - splitplane->dist; if (dist > light->radius) { Q2BSP_MarkLights (light, bit, node->children[0]); return; } if (dist < -light->radius) { Q2BSP_MarkLights (light, bit, node->children[1]); return; } // mark the polygons surf = cl.worldmodel->surfaces + node->firstsurface; for (i=0 ; inumsurfaces ; i++, surf++) { if (surf->dlightframe != r_dlightframecount) { surf->dlightbits = 0u; surf->dlightframe = r_dlightframecount; } surf->dlightbits |= bit; } Q2BSP_MarkLights (light, bit, node->children[0]); Q2BSP_MarkLights (light, bit, node->children[1]); } #ifndef SERVERONLY static void GLR_Q2BSP_StainNode (mnode_t *node, float *parms) { mplane_t *splitplane; float dist; msurface_t *surf; int i; if (node->contents != -1) return; splitplane = node->plane; dist = DotProduct ((parms+1), splitplane->normal) - splitplane->dist; if (dist > (*parms)) { GLR_Q2BSP_StainNode (node->children[0], parms); return; } if (dist < (-*parms)) { GLR_Q2BSP_StainNode (node->children[1], parms); return; } // mark the polygons surf = cl.worldmodel->surfaces + node->firstsurface; for (i=0 ; inumsurfaces ; i++, surf++) { if (surf->flags&~(SURF_DONTWARP|SURF_PLANEBACK)) continue; Surf_StainSurf(surf, parms); } GLR_Q2BSP_StainNode (node->children[0], parms); GLR_Q2BSP_StainNode (node->children[1], parms); } #endif #endif static void CM_BuildBIH(model_t *mod, int submodel) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; cmodel_t *sub = &prv->cmodels[submodel]; struct bihleaf_s *bihleaf, *l; size_t bihleafs, i; //undo any bih damage on the copy mod->cnodes = NULL; mod->funcs.NativeTrace = CM_NativeTrace; mod->funcs.NativeContents = CM_NativeContents; mod->funcs.PointContents = Q2BSP_PointContents; if (!q3bsp_bihtraces.ival) return; //skip this. fall back on other stuff. if (mod->fromgame != fg_quake3) return; bihleafs = sub->num_brushes; for (i = 0; i < sub->num_patches; i++) bihleafs += prv->patches[sub->firstpatch + i].numfacets; for (i = 0; i < sub->num_cmeshes; i++) bihleafs += prv->cmeshes[sub->firstcmesh + i].numincidies/3; bihleaf = l = BZ_Malloc(sizeof(*bihleaf)*bihleafs); //now we have enough storage, spit them out providing bounds info. for (i = 0; i < sub->num_brushes; i++) { q2cbrush_t *b = &prv->brushes[sub->firstbrush+i]; l->type = BIH_BRUSH; l->data.brush = b; l->data.contents = b->contents; VectorCopy(b->absmins, l->mins); VectorCopy(b->absmaxs, l->maxs); l++; } #ifdef Q3BSPS for (i = 0; i < sub->num_patches; i++) { q3cpatch_t *p = &prv->patches[sub->firstpatch+i]; size_t j; for (j = 0; j < p->numfacets; j++) { q2cbrush_t *b = &p->facets[j]; l->type = BIH_PATCHBRUSH; l->data.patchbrush = b; l->data.contents = b->contents; VectorCopy(b->absmins, l->mins); VectorCopy(b->absmaxs, l->maxs); l++; } } #endif #ifdef Q3BSPS for (i = 0; i < sub->num_cmeshes; i++) { q3cmesh_t *m = &prv->cmeshes[sub->firstcmesh+i]; size_t j; for (j = 0; j+2 < m->numincidies; j+=3) { index_t *v = m->indicies+j; vec_t *v1 = m->xyz_array[v[0]], *v2 = m->xyz_array[v[1]], *v3 = m->xyz_array[v[2]]; l->type = BIH_TRIANGLE; l->data.tri.xyz = m->xyz_array; l->data.tri.indexes = v; l->data.contents = m->surface->c.value; VectorCopy(v1, l->mins); VectorCopy(v1, l->maxs); AddPointToBounds(v2, l->mins, l->maxs); AddPointToBounds(v3, l->mins, l->maxs); l++; } } #endif BIH_Build(mod, bihleaf, l-bihleaf); BZ_Free(bihleaf); } /* ================== CM_LoadMap Loads in the map and all submodels ================== */ static cmodel_t *CM_LoadMap (model_t *mod, qbyte *filein, size_t filelen, qboolean clientload) { unsigned *buf; int i; q2dheader_t header; int length; qboolean noerrors = true; model_t *wmod = mod; char loadname[32]; qbyte *mod_base = (qbyte *)filein; bspx_header_t *bspx = NULL; unsigned int checksum; #ifdef Q3BSPS extern cvar_t gl_overbright; #endif #ifndef SERVERONLY void (*buildmeshes)(model_t *mod, msurface_t *surf, builddata_t *cookie) = NULL; qbyte *facedata = NULL; unsigned int facesize = 0; #endif cminfo_t *prv; COM_FileBase (mod->name, loadname, sizeof(loadname)); // free old stuff mod->meshinfo = prv = ZG_Malloc(&mod->memgroup, sizeof(*prv)); prv->numcmodels = 0; prv->numvisibility = 0; mod->type = mod_brush; if (!mod->name[0]) { prv->cmodels = ZG_Malloc(&mod->memgroup, 1 * sizeof(*prv->cmodels)); mod->leafs = ZG_Malloc(&mod->memgroup, 1 * sizeof(*mod->leafs)); prv->numcmodels = 1; prv->numareas = 1; mod->checksum = mod->checksum2 = 0; prv->cmodels[0].headnode = (mnode_t*)mod->leafs; //directly start with the empty leaf return &prv->cmodels[0]; // cinematic servers won't have anything at all } // // load the file // buf = (unsigned *)filein; length = filelen; if (!buf) { Con_Printf (CON_ERROR "Couldn't load %s\n", mod->name); return NULL; } checksum = LittleLong (Com_BlockChecksum (buf, length)); header = *(q2dheader_t *)(buf); header.ident = LittleLong(header.ident); header.version = LittleLong(header.version); ClearBounds(mod->mins, mod->maxs); switch(header.version) { default: Con_Printf (CON_ERROR "Quake 2 or Quake 3 based BSP with unknown header (%s: %i should be %i or %i)\n" , mod->name, header.version, BSPVERSION_Q2, BSPVERSION_Q3); return NULL; break; #ifdef Q3BSPS #ifdef RFBSPS case BSPVERSION_RBSP: //rbsp/fbsp #endif case BSPVERSION_RTCW: //rtcw case BSPVERSION_Q3: #ifdef RFBSPS if (header.ident == (('F'<<0)+('B'<<8)+('S'<<16)+('P'<<24))) { mod->lightmaps.width = 512; mod->lightmaps.height = 512; } else #endif { mod->lightmaps.width = 128; mod->lightmaps.height = 128; } prv->mapisq3 = true; mod->fromgame = fg_quake3; for (i=0 ; i filelen) { Con_Printf (CON_ERROR "WARNING: q3bsp %s truncated (lump %i, %i+%i > %u)\n", mod->name, i, header.lumps[i].fileofs, header.lumps[i].filelen, (unsigned int)filelen); header.lumps[i].filelen = filelen - header.lumps[i].fileofs; if (header.lumps[i].filelen < 0) header.lumps[i].filelen = 0; } } } /* #ifndef SERVERONLY GLMod_LoadVertexes (mod, cmod_base, &header.lumps[Q3LUMP_DRAWVERTS]); // GLMod_LoadEdges (mod, cmod_base, &header.lumps[Q3LUMP_EDGES]); // GLMod_LoadSurfedges (mod, cmod_base, &header.lumps[Q3LUMP_SURFEDGES]); GLMod_LoadLighting (mod, cmod_base, &header.lumps[Q3LUMP_LIGHTMAPS]); #endif CModQ3_LoadShaders (mod, cmod_base, &header.lumps[Q3LUMP_SHADERS]); CModQ3_LoadPlanes (mod, cmod_base, &header.lumps[Q3LUMP_PLANES]); CModQ3_LoadLeafBrushes (mod, cmod_base, &header.lumps[Q3LUMP_LEAFBRUSHES]); CModQ3_LoadBrushes (mod, cmod_base, &header.lumps[Q3LUMP_BRUSHES]); CModQ3_LoadBrushSides (mod, cmod_base, &header.lumps[Q3LUMP_BRUSHSIDES]); #ifndef SERVERONLY CMod_LoadTexInfo (mod, cmod_base, &header.lumps[Q3LUMP_SHADERS]); CMod_LoadFaces (mod, cmod_base, &header.lumps[Q3LUMP_SURFACES]); // GLMod_LoadMarksurfaces (mod, cmod_base, &header.lumps[Q3LUMP_LEAFFACES]); #endif CMod_LoadVisibility (mod, cmod_base, &header.lumps[Q3LUMP_VISIBILITY]); CModQ3_LoadSubmodels (mod, cmod_base, &header.lumps[Q3LUMP_MODELS]); CModQ3_LoadLeafs (mod, cmod_base, &header.lumps[Q3LUMP_LEAFS]); CModQ3_LoadNodes (mod, cmod_base, &header.lumps[Q3LUMP_NODES]); // CMod_LoadAreas (mod, cmod_base, &header.lumps[Q3LUMP_AREAS]); // CMod_LoadAreaPortals (mod, cmod_base, &header.lumps[Q3LUMP_AREAPORTALS]); CMod_LoadEntityString (mod, cmod_base, &header.lumps[Q3LUMP_ENTITIES]); */ prv->faces = NULL; bspx = BSPX_Setup(mod, mod_base, filelen, header.lumps, Q3LUMPS_TOTAL); //q3 maps have built in 4-fold overbright. //if we're not rendering with that, we need to brighten the lightmaps in order to keep the darker parts the same brightness. we loose the 2 upper bits. those bright areas become uniform and indistinct. //this is used for both the lightmap AND vertex lighting //FIXME: when not using overbrights, we suffer a loss of precision. gl_overbright.flags |= CVAR_RENDERERLATCH; BuildLightMapGammaTable(1, (1<<(2-gl_overbright.ival))); prv->mapisq3 = true; noerrors = noerrors && CModQ3_LoadShaders (mod, mod_base, &header.lumps[Q3LUMP_SHADERS]); noerrors = noerrors && CModQ3_LoadPlanes (mod, mod_base, &header.lumps[Q3LUMP_PLANES]); #ifdef RFBSPS if (header.version == BSPVERSION_RBSP) { noerrors = noerrors && CModRBSP_LoadBrushSides (mod, mod_base, &header.lumps[Q3LUMP_BRUSHSIDES]); noerrors = noerrors && CModRBSP_LoadVertexes (mod, mod_base, &header.lumps[Q3LUMP_DRAWVERTS]); } else #endif { noerrors = noerrors && CModQ3_LoadBrushSides (mod, mod_base, &header.lumps[Q3LUMP_BRUSHSIDES]); noerrors = noerrors && CModQ3_LoadVertexes (mod, mod_base, &header.lumps[Q3LUMP_DRAWVERTS]); } noerrors = noerrors && CModQ3_LoadBrushes (mod, mod_base, &header.lumps[Q3LUMP_BRUSHES]); noerrors = noerrors && CModQ3_LoadLeafBrushes (mod, mod_base, &header.lumps[Q3LUMP_LEAFBRUSHES]); #ifdef RFBSPS if (header.version == BSPVERSION_RBSP) noerrors = noerrors && CModRBSP_LoadFaces (mod, mod_base, &header.lumps[Q3LUMP_SURFACES]); else #endif noerrors = noerrors && CModQ3_LoadFaces (mod, mod_base, &header.lumps[Q3LUMP_SURFACES]); if (noerrors) Mod_LoadEntities (mod, mod_base, &header.lumps[Q3LUMP_ENTITIES]); #ifndef SERVERONLY if (qrenderer != QR_NONE) { #ifdef RFBSPS if (header.version == BSPVERSION_RBSP) noerrors = noerrors && CModRBSP_LoadLightgrid (mod, mod_base, &header.lumps[Q3LUMP_LIGHTGRID], &header.lumps[RBSPLUMP_LIGHTINDEXES]); else #endif noerrors = noerrors && CModQ3_LoadLightgrid (mod, mod_base, &header.lumps[Q3LUMP_LIGHTGRID]); noerrors = noerrors && CModQ3_LoadIndexes (mod, mod_base, &header.lumps[Q3LUMP_DRAWINDEXES]); if (header.version != BSPVERSION_RTCW) noerrors = noerrors && CModQ3_LoadFogs (mod, mod_base, &header.lumps[Q3LUMP_FOGS]); else mod->numfogs = 0; facedata = (void *)(mod_base + header.lumps[Q3LUMP_SURFACES].fileofs); #ifdef RFBSPS if (header.version == BSPVERSION_RBSP) { noerrors = noerrors && CModRBSP_LoadRFaces (mod, mod_base, &header.lumps[Q3LUMP_SURFACES]); buildmeshes = CModRBSP_BuildSurfMesh; facesize = sizeof(rbspface_t); mod->lightmaps.surfstyles = 4; } else #endif { noerrors = noerrors && CModQ3_LoadRFaces (mod, mod_base, &header.lumps[Q3LUMP_SURFACES]); buildmeshes = CModQ3_BuildSurfMesh; facesize = sizeof(q3dface_t); mod->lightmaps.surfstyles = 1; } if (noerrors) { i = header.lumps[Q3LUMP_LIGHTMAPS].filelen / (mod->lightmaps.width*mod->lightmaps.height*3); mod->lightmaps.deluxemapping = !(i&1); mod->lightmaps.count = max(mod->lightmaps.count, i); mod->lightmaps.deluxemapping_modelspace = true; //we assume true for q3bsp. for (i = 0; i < mod->numsurfaces && mod->lightmaps.deluxemapping; i++) { if (mod->surfaces[i].lightmaptexturenums[0] >= 0 && (mod->surfaces[i].lightmaptexturenums[0] & 1)) mod->lightmaps.deluxemapping = false; } { char deluxeMaps[64], *key; key = (char*)Mod_ParseWorldspawnKey(mod, "deluxeMaps", deluxeMaps, sizeof(deluxeMaps)); if (*key) { switch(atoi(key)) { case 0: mod->lightmaps.deluxemapping = false; break; case 1: // mod->lightmaps.deluxemapping = true; mod->lightmaps.deluxemapping_modelspace = true; break; case 2: // mod->lightmaps.deluxemapping = true; mod->lightmaps.deluxemapping_modelspace = false; break; } } } } if (noerrors) CModQ3_LoadLighting (mod, mod_base, &header.lumps[Q3LUMP_LIGHTMAPS]); //fixme: duplicated loading. } #endif noerrors = noerrors && CModQ3_LoadMarksurfaces (mod, mod_base, &header.lumps[Q3LUMP_LEAFSURFACES]); noerrors = noerrors && CModQ3_LoadLeafs (mod, mod_base, &header.lumps[Q3LUMP_LEAFS]); noerrors = noerrors && CModQ3_LoadNodes (mod, mod_base, &header.lumps[Q3LUMP_NODES]); noerrors = noerrors && CModQ3_LoadSubmodels (mod, mod_base, &header.lumps[Q3LUMP_MODELS]); noerrors = noerrors && CModQ3_LoadVisibility (mod, mod_base, &header.lumps[Q3LUMP_VISIBILITY]); if (!noerrors) { if (prv->faces) BZ_Free(prv->faces); return NULL; } #ifndef CLIENTONLY mod->funcs.FatPVS = Q23BSP_FatPVS; mod->funcs.EdictInFatPVS = Q23BSP_EdictInFatPVS; mod->funcs.FindTouchedLeafs = Q23BSP_FindTouchedLeafs; #endif mod->funcs.ClusterPVS = CM_ClusterPVS; mod->funcs.ClusterForPoint = CM_PointCluster; #ifndef SERVERONLY mod->funcs.LightPointValues = GLQ3_LightGrid; mod->funcs.StainNode = GLR_Q2BSP_StainNode; mod->funcs.MarkLights = Q2BSP_MarkLights; #endif mod->funcs.PointContents = Q2BSP_PointContents; mod->funcs.NativeTrace = CM_NativeTrace; mod->funcs.NativeContents = CM_NativeContents; #ifndef SERVERONLY //light grid info if (mod->lightgrid) { char gridsize[256], *key; char val[64]; float maxs; q3lightgridinfo_t *lg = mod->lightgrid; key = (char*)Mod_ParseWorldspawnKey(mod, "gridsize", gridsize, sizeof(gridsize)); key = COM_ParseOut(key, val, sizeof(val)); lg->gridSize[0] = atof(val); key = COM_ParseOut(key, val, sizeof(val)); lg->gridSize[1] = atof(val); key = COM_ParseOut(key, val, sizeof(val)); lg->gridSize[2] = atof(val); if ( lg->gridSize[0] < 1 || lg->gridSize[1] < 1 || lg->gridSize[2] < 1 ) { lg->gridSize[0] = 64; lg->gridSize[1] = 64; lg->gridSize[2] = 128; } for ( i = 0; i < 3; i++ ) { lg->gridMins[i] = lg->gridSize[i] * ceil( (prv->cmodels->mins[i] + 1) / lg->gridSize[i] ); maxs = lg->gridSize[i] * floor( (prv->cmodels->maxs[i] - 1) / lg->gridSize[i] ); lg->gridBounds[i] = (maxs - lg->gridMins[i])/lg->gridSize[i] + 1; } lg->gridBounds[3] = lg->gridBounds[1] * lg->gridBounds[0]; } #endif if (!CM_CreatePatchesForLeafs (mod, prv)) //for clipping { BZ_Free(prv->faces); return NULL; } #ifndef CLIENTONLY CMQ3_CalcPHS(mod); #endif // BZ_Free(map_verts); BZ_Free(prv->faces); break; #endif #ifdef Q2BSPS case BSPVERSION_Q2: case BSPVERSION_Q2W: mod->lightmaps.width = LMBLOCK_SIZE_MAX; mod->lightmaps.height = LMBLOCK_SIZE_MAX; prv->mapisq3 = false; mod->engineflags |= MDLF_NEEDOVERBRIGHT; for (i=0 ; ifuncs.FatPVS = Q23BSP_FatPVS; mod->funcs.EdictInFatPVS = Q23BSP_EdictInFatPVS; mod->funcs.FindTouchedLeafs = Q23BSP_FindTouchedLeafs; #endif mod->funcs.LightPointValues = NULL; mod->funcs.StainNode = NULL; mod->funcs.MarkLights = NULL; mod->funcs.ClusterPVS = CM_ClusterPVS; mod->funcs.ClusterForPoint = CM_PointCluster; mod->funcs.PointContents = Q2BSP_PointContents; mod->funcs.NativeTrace = CM_NativeTrace; mod->funcs.NativeContents = CM_NativeContents; break; #ifndef SERVERONLY default: { // load into heap noerrors = noerrors && Mod_LoadVertexes (mod, mod_base, &header.lumps[Q2LUMP_VERTEXES]); if (header.version == BSPVERSION_Q2W) /*noerrors = noerrors &&*/ Mod_LoadVertexNormals(mod, bspx, mod_base, &header.lumps[19]); noerrors = noerrors && Mod_LoadEdges (mod, mod_base, &header.lumps[Q2LUMP_EDGES], false); noerrors = noerrors && Mod_LoadSurfedges (mod, mod_base, &header.lumps[Q2LUMP_SURFEDGES]); noerrors = noerrors && CModQ2_LoadSurfaces (mod, mod_base, &header.lumps[Q2LUMP_TEXINFO]); noerrors = noerrors && CModQ2_LoadPlanes (mod, mod_base, &header.lumps[Q2LUMP_PLANES]); noerrors = noerrors && CModQ2_LoadTexInfo (mod, mod_base, &header.lumps[Q2LUMP_TEXINFO], loadname); if (noerrors) Mod_LoadEntities (mod, mod_base, &header.lumps[Q2LUMP_ENTITIES]); noerrors = noerrors && CModQ2_LoadFaces (mod, mod_base, &header.lumps[Q2LUMP_FACES], &header.lumps[Q2LUMP_LIGHTING], header.version == BSPVERSION_Q2W, bspx); noerrors = noerrors && Mod_LoadMarksurfaces (mod, mod_base, &header.lumps[Q2LUMP_LEAFFACES], false); noerrors = noerrors && CModQ2_LoadVisibility (mod, mod_base, &header.lumps[Q2LUMP_VISIBILITY]); noerrors = noerrors && CModQ2_LoadBrushSides (mod, mod_base, &header.lumps[Q2LUMP_BRUSHSIDES]); noerrors = noerrors && CModQ2_LoadBrushes (mod, mod_base, &header.lumps[Q2LUMP_BRUSHES]); noerrors = noerrors && CModQ2_LoadLeafBrushes (mod, mod_base, &header.lumps[Q2LUMP_LEAFBRUSHES]); noerrors = noerrors && CModQ2_LoadLeafs (mod, mod_base, &header.lumps[Q2LUMP_LEAFS]); noerrors = noerrors && CModQ2_LoadNodes (mod, mod_base, &header.lumps[Q2LUMP_NODES]); noerrors = noerrors && CModQ2_LoadSubmodels (mod, mod_base, &header.lumps[Q2LUMP_MODELS]); noerrors = noerrors && CModQ2_LoadAreas (mod, mod_base, &header.lumps[Q2LUMP_AREAS]); noerrors = noerrors && CModQ2_LoadAreaPortals (mod, mod_base, &header.lumps[Q2LUMP_AREAPORTALS]); if (!noerrors) { return NULL; } #ifndef CLIENTONLY mod->funcs.FatPVS = Q23BSP_FatPVS; mod->funcs.EdictInFatPVS = Q23BSP_EdictInFatPVS; mod->funcs.FindTouchedLeafs = Q23BSP_FindTouchedLeafs; #endif mod->funcs.LightPointValues = GLQ2BSP_LightPointValues; mod->funcs.StainNode = GLR_Q2BSP_StainNode; mod->funcs.MarkLights = Q2BSP_MarkLights; mod->funcs.ClusterPVS = CM_ClusterPVS; mod->funcs.ClusterForPoint = CM_PointCluster; mod->funcs.PointContents = Q2BSP_PointContents; mod->funcs.NativeTrace = CM_NativeTrace; mod->funcs.NativeContents = CM_NativeContents; } break; #endif } #endif } BSPX_LoadEnvmaps(mod, bspx, mod_base); #ifdef Q3BSPS { int x, y; for (x = 0; x < prv->numareas; x++) for (y = 0; y < prv->numareas; y++) prv->q3areas[x].numareaportals[y] = map_autoopenportals.ival; } #endif #ifdef Q2BSPS if (map_autoopenportals.value) memset (prv->q2portalopen, 1, sizeof(prv->q2portalopen)); //open them all. Used for progs that havn't got a clue. else memset (prv->q2portalopen, 0, sizeof(prv->q2portalopen)); //make them start closed. #endif FloodAreaConnections (prv); mod->checksum = mod->checksum2 = checksum; mod->nummodelsurfaces = mod->numsurfaces; memset(&mod->batches, 0, sizeof(mod->batches)); mod->vbos = NULL; mod->numsubmodels = CM_NumInlineModels(mod); mod->hulls[0].firstclipnode = prv->cmodels[0].headnode-mod->nodes; mod->rootnode = prv->cmodels[0].headnode; mod->nummodelsurfaces = prv->cmodels[0].numsurfaces; #ifndef SERVERONLY if (qrenderer != QR_NONE) { builddata_t *bd = NULL; if (buildmeshes) { bd = BZ_Malloc(sizeof(*bd) + facesize*mod->nummodelsurfaces); bd->buildfunc = buildmeshes; memcpy(bd+1, facedata + mod->firstmodelsurface*facesize, facesize*mod->nummodelsurfaces); } COM_AddWork(WG_MAIN, ModBrush_LoadGLStuff, mod, bd, 0, 0); } #endif //FIXME: q2bsp apparently doesn't report which brushes are part of which submodels. //FIXME: ALL patches? not just worldmodel? CM_BuildBIH(mod, 0); for (i=1 ; i< mod->numsubmodels ; i++) { cmodel_t *bm; char name[MAX_QPATH]; Q_snprintfz (name, sizeof(name), "*%i:%s", i, wmod->publicname); mod = Mod_FindName (name); *mod = *wmod; mod->archive = NULL; mod->entities_raw = NULL; mod->submodelof = wmod; Q_strncpyz(mod->publicname, name, sizeof(mod->publicname)); Q_snprintfz (mod->name, sizeof(mod->name), "*%i:%s", i, wmod->name); memset(&mod->memgroup, 0, sizeof(mod->memgroup)); bm = CM_InlineModel (wmod, name); mod->hulls[0].firstclipnode = -1; //no nodes, if (bm->headleaf) { mod->leafs = bm->headleaf; mod->nodes = NULL; mod->hulls[0].firstclipnode = -1; //make it refer directly to the first leaf, for things that still use numbers. mod->rootnode = (mnode_t*)bm->headleaf; } else { mod->leafs = wmod->leafs; mod->nodes = wmod->nodes; mod->hulls[0].firstclipnode = bm->headnode - mod->nodes; //determine the correct node index mod->rootnode = bm->headnode; } mod->nummodelsurfaces = bm->numsurfaces; mod->firstmodelsurface = bm->firstsurface; CM_BuildBIH(mod, i); memset(&mod->batches, 0, sizeof(mod->batches)); mod->vbos = NULL; VectorCopy (bm->maxs, mod->maxs); VectorCopy (bm->mins, mod->mins); #ifndef SERVERONLY mod->radius = RadiusFromBounds (mod->mins, mod->maxs); if (qrenderer != QR_NONE) { builddata_t *bd = NULL; if (buildmeshes) { bd = BZ_Malloc(sizeof(*bd) + facesize*mod->nummodelsurfaces); bd->buildfunc = buildmeshes; memcpy(bd+1, facedata + mod->firstmodelsurface*facesize, facesize*mod->nummodelsurfaces); } COM_AddWork(WG_MAIN, ModBrush_LoadGLStuff, mod, bd, i, 0); } #endif COM_AddWork(WG_MAIN, Mod_ModelLoaded, mod, NULL, MLS_LOADED, 0); } #ifdef TERRAIN mod->terrain = Mod_LoadTerrainInfo(mod, loadname, false); #endif return &prv->cmodels[0]; } /* ================== CM_InlineModel ================== */ static cmodel_t *CM_InlineModel (model_t *model, char *name) { cminfo_t *prv = (cminfo_t*)model->meshinfo; int num; if (!name) Host_Error("Bad model\n"); else if (name[0] != '*') Host_Error("Bad model\n"); num = atoi (name+1); if (num < 1 || num >= prv->numcmodels) Host_Error ("CM_InlineModel: bad number"); return &prv->cmodels[num]; } int CM_ClusterBytes (model_t *model) { #ifdef Q3BSPS if (model->fromgame == fg_quake3) { cminfo_t *prv = (cminfo_t*)model->meshinfo; return prv->q3pvs->rowsize ? prv->q3pvs->rowsize : model->pvsbytes; } else #endif return model->pvsbytes; } static int CM_NumInlineModels (model_t *model) { cminfo_t *prv = (cminfo_t*)model->meshinfo; return prv->numcmodels; } int CM_LeafContents (model_t *model, int leafnum) { if (leafnum < 0 || leafnum >= model->numleafs) Host_Error ("CM_LeafContents: bad number"); return model->leafs[leafnum].contents; } int CM_LeafCluster (model_t *model, int leafnum) { if (leafnum < 0 || leafnum >= model->numleafs) Host_Error ("CM_LeafCluster: bad number"); return model->leafs[leafnum].cluster; } int CM_LeafArea (model_t *model, int leafnum) { if (leafnum < 0 || leafnum >= model->numleafs) Host_Error ("CM_LeafArea: bad number"); return model->leafs[leafnum].area; } //======================================================================= #define PlaneDiff(point,plane) (((plane)->type < 3 ? (point)[(plane)->type] : DotProduct((point), (plane)->normal)) - (plane)->dist) mplane_t box_planes[6]; model_t box_model; q2cbrush_t box_brush; q2cbrushside_t box_sides[6]; static qboolean BM_NativeTrace(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 contents, trace_t *trace); static unsigned int BM_NativeContents(struct model_s *model, int hulloverride, const framestate_t *framestate, const vec3_t axis[3], const vec3_t p, const vec3_t mins, const vec3_t maxs) { unsigned int j; q2cbrushside_t *brushside = box_sides; for ( j = 0; j < 6; j++, brushside++ ) { if ( PlaneDiff (p, brushside->plane) > 0 ) return 0; } return FTECONTENTS_BODY; } /* =================== CM_InitBoxHull Set up the planes and nodes so that the six floats of a bounding box can just be stored out and get a proper clipping hull structure. =================== */ void CM_InitBoxHull (void) { int i; mplane_t *p; q2cbrushside_t *s; /* #ifndef CLIENTONLY box_model.funcs.FatPVS = Q2BSP_FatPVS; box_model.funcs.EdictInFatPVS = Q2BSP_EdictInFatPVS; box_model.funcs.FindTouchedLeafs = Q23BSP_FindTouchedLeafs; #endif #ifndef SERVERONLY box_model.funcs.MarkLights = Q2BSP_MarkLights; #endif box_model.funcs.ClusterPVS = CM_ClusterPVS; box_model.funcs.ClusterForPoint = CM_PointCluster; */ box_model.funcs.NativeContents = BM_NativeContents; box_model.funcs.NativeTrace = BM_NativeTrace; box_model.loadstate = MLS_LOADED; box_brush.contents = FTECONTENTS_BODY; box_brush.numsides = 6; box_brush.brushside = box_sides; for (i=0 ; i<6 ; i++) { //the pointers s = &box_sides[i]; p = &box_planes[i]; // brush sides s->plane = p; s->surface = &nullsurface; // planes p->type = ((i>=3)?i-3:i); p->signbits = 0; VectorClear (p->normal); p->normal[p->type] = ((i>=3)?-1:1); } } /* =================== CM_HeadnodeForBox To keep everything totally uniform, bounding boxes are turned into small BSP trees instead of being compared directly. =================== */ static void CM_SetTempboxSize (const vec3_t mins, const vec3_t maxs) { box_planes[0].dist = maxs[0]; box_planes[1].dist = maxs[1]; box_planes[2].dist = maxs[2]; box_planes[3].dist = -mins[0]; box_planes[4].dist = -mins[1]; box_planes[5].dist = -mins[2]; } model_t *CM_TempBoxModel(const vec3_t mins, const vec3_t maxs) { CM_SetTempboxSize(mins, maxs); return &box_model; } /* ================== CM_PointLeafnum_r ================== */ static int CM_PointLeafnum_r (model_t *mod, const vec3_t p, int num) { float d; mnode_t *node; mplane_t *plane; while (num >= 0) { node = mod->nodes + num; plane = node->plane; if (plane->type < 3) d = p[plane->type] - plane->dist; else d = DotProduct (plane->normal, p) - plane->dist; if (d < 0) num = node->childnum[1]; else num = node->childnum[0]; } return -1 - num; } int CM_PointLeafnum (model_t *mod, const vec3_t p) { if (!mod || mod->loadstate != MLS_LOADED) return 0; // sound may call this without map loaded return CM_PointLeafnum_r (mod, p, 0); } static int CM_PointCluster (model_t *mod, const vec3_t p, int *area) { int leaf; if (!mod || mod->loadstate != MLS_LOADED) return 0; // sound may call this without map loaded leaf = CM_PointLeafnum_r (mod, p, 0); if (area) *area = CM_LeafArea(mod, leaf); return CM_LeafCluster(mod, leaf); } /* ============= CM_BoxLeafnums Fills in a list of all the leafs touched ============= */ int leaf_count, leaf_maxcount; int *leaf_list; const float *leaf_mins, *leaf_maxs; int leaf_topnode; static void CM_BoxLeafnums_r (model_t *mod, int nodenum) { mplane_t *plane; mnode_t *node; int s; while (1) { if (nodenum < 0) { if (leaf_count >= leaf_maxcount) { // Com_Printf ("CM_BoxLeafnums_r: overflow\n"); return; } leaf_list[leaf_count++] = -1 - nodenum; return; } node = &mod->nodes[nodenum]; plane = node->plane; // s = BoxOnPlaneSide (leaf_mins, leaf_maxs, plane); s = BOX_ON_PLANE_SIDE(leaf_mins, leaf_maxs, plane); if (s == 1) nodenum = node->childnum[0]; else if (s == 2) nodenum = node->childnum[1]; else { // go down both if (leaf_topnode == -1) leaf_topnode = nodenum; CM_BoxLeafnums_r (mod, node->childnum[0]); nodenum = node->childnum[1]; } } } static int CM_BoxLeafnums_headnode (model_t *mod, const vec3_t mins, const vec3_t maxs, int *list, int listsize, int headnode, int *topnode) { leaf_list = list; leaf_count = 0; leaf_maxcount = listsize; leaf_mins = mins; leaf_maxs = maxs; leaf_topnode = -1; CM_BoxLeafnums_r (mod, headnode); if (topnode) *topnode = leaf_topnode; return leaf_count; } int CM_BoxLeafnums (model_t *mod, const vec3_t mins, const vec3_t maxs, int *list, int listsize, int *topnode) { return CM_BoxLeafnums_headnode (mod, mins, maxs, list, listsize, mod->hulls[0].firstclipnode, topnode); } /* ================== CM_PointContents ================== */ int CM_PointContents (model_t *mod, const vec3_t p) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i, j, contents; mleaf_t *leaf; q2cbrush_t *brush; q2cbrushside_t *brushside; if (!mod) // map not loaded return 0; i = CM_PointLeafnum_r (mod, p, mod->hulls[0].firstclipnode); if (mod->fromgame == fg_quake2) contents = mod->leafs[i].contents; //q2 is simple. else { leaf = &mod->leafs[i]; // if ( leaf->contents & CONTENTS_NODROP ) { // contents = CONTENTS_NODROP; // } else { contents = 0; // } for (i = 0; i < leaf->numleafbrushes; i++) { brush = prv->leafbrushes[leaf->firstleafbrush + i]; // check if brush actually adds something to contents if ( (contents & brush->contents) == brush->contents ) { continue; } brushside = brush->brushside; for ( j = 0; j < brush->numsides; j++, brushside++ ) { if ( PlaneDiff (p, brushside->plane) > 0 ) break; } if (j == brush->numsides) contents |= brush->contents; } } #ifdef TERRAIN if (mod->terrain) contents |= Heightmap_PointContents(mod, NULL, p); #endif return contents; } unsigned int CM_NativeContents(struct model_s *model, int hulloverride, const framestate_t *framestate, const vec3_t axis[3], const vec3_t p, const vec3_t mins, const vec3_t maxs) { cminfo_t *prv = (cminfo_t*)model->meshinfo; int contents; if (!DotProduct(mins, mins) && !DotProduct(maxs, maxs)) return CM_PointContents(model, p); if (!model) // map not loaded return 0; { int i, j, k; mleaf_t *leaf; q2cbrush_t *brush; q2cbrushside_t *brushside; vec3_t absmin, absmax; int leaflist[64]; k = CM_BoxLeafnums (model, absmin, absmax, leaflist, 64, NULL); contents = 0; for (k--; k >= 0; k--) { leaf = &model->leafs[leaflist[k]]; if (model->fromgame != fg_quake2) { //q3 is more complex for (i = 0; i < leaf->numleafbrushes; i++) { brush = prv->leafbrushes[leaf->firstleafbrush + i]; // check if brush actually adds something to contents if ( (contents & brush->contents) == brush->contents ) { continue; } brushside = brush->brushside; for ( j = 0; j < brush->numsides; j++, brushside++ ) { if ( PlaneDiff (p, brushside->plane) > 0 ) break; } if (j == brush->numsides) contents |= brush->contents; } } else //q2 is simple contents |= leaf->contents; } } return contents; } /* ================== CM_TransformedPointContents Handles offseting and rotation of the end points for moving and rotating entities ================== */ int CM_TransformedPointContents (model_t *mod, const vec3_t p, int headnode, const vec3_t origin, const vec3_t angles) { vec3_t p_l; vec3_t temp; vec3_t forward, right, up; // subtract origin offset VectorSubtract (p, origin, p_l); // rotate start and end into the models frame of reference if (angles[0] || angles[1] || angles[2]) { AngleVectors (angles, forward, right, up); VectorCopy (p_l, temp); p_l[0] = DotProduct (temp, forward); p_l[1] = -DotProduct (temp, right); p_l[2] = DotProduct (temp, up); } return CM_PointContents(mod, p); } /* =============================================================================== BOX TRACING =============================================================================== */ // 1/32 epsilon to keep floating point happy #define DIST_EPSILON (0.03125) static vec3_t trace_start, trace_end; static vec3_t trace_mins, trace_maxs; static vec3_t trace_extents; static vec3_t trace_absmins, trace_absmaxs; static vec3_t trace_up; //capsule points upwards in this direction static vec3_t trace_capsulesize; //radius, up, down static float trace_truefraction; static float trace_nearfraction; static trace_t trace_trace; static int trace_contents; static enum { shape_isbox, shape_iscapsule, shape_ispoint } trace_shape; // optimized case static void CM_FinalizeBrush(q2cbrush_t *brush) { vecV_t verts[256]; vec4_t planes[256]; int i, j; ClearBounds(brush->absmins, brush->absmaxs); for (i = 0; i < brush->numsides; i++) { VectorCopy(brush->brushside[i].plane->normal, planes[i]); planes[i][3] = brush->brushside[i].plane->dist; } for (i = 0; i < brush->numsides; i++) { //most brushes are axial, which can save some a little loadtime if (planes[i][0] == 1) brush->absmaxs[0] = planes[i][3]; else if (planes[i][1] == 1) brush->absmaxs[1] = planes[i][3]; else if (planes[i][2] == 1) brush->absmaxs[2] = planes[i][3]; else if (planes[i][0] == -1) brush->absmins[0] = -planes[i][3]; else if (planes[i][1] == -1) brush->absmins[1] = -planes[i][3]; else if (planes[i][2] == -1) brush->absmins[2] = -planes[i][3]; else { j = Fragment_ClipPlaneToBrush(verts, countof(verts), planes, sizeof(planes[0]), brush->numsides, planes[i]); while (j-- > 0) AddPointToBounds(verts[j], brush->absmins, brush->absmaxs); } } } /* ================ CM_ClipBoxToBrush ================ */ static void CM_ClipBoxToBrush (vec3_t mins, vec3_t maxs, vec3_t p1, vec3_t p2, trace_t *trace, q2cbrush_t *brush) { int i, j; mplane_t *plane, *clipplane; float dist; float enterfrac, leavefrac; vec3_t ofs; float d1, d2; qboolean getout, startout; float f; q2cbrushside_t *side, *leadside; float nearfrac=0; enterfrac = -1; leavefrac = 2; clipplane = NULL; if (!brush->numsides) return; getout = false; startout = false; leadside = NULL; for (i=0 ; inumsides ; i++) { side = brush->brushside+i; plane = side->plane; switch(trace_shape) { default: case shape_isbox: // general box case // push the plane out apropriately for mins/maxs // FIXME: use signbits into 8 way lookup for each mins/maxs for (j=0 ; j<3 ; j++) { if (plane->normal[j] < 0) ofs[j] = maxs[j]; else ofs[j] = mins[j]; } dist = DotProduct (ofs, plane->normal); dist = plane->dist - dist; break; capsuledist(dist,plane,mins,maxs) case shape_ispoint: // special point case dist = plane->dist; break; } d1 = DotProduct (p1, plane->normal) - dist; d2 = DotProduct (p2, plane->normal) - dist; if (d2 > 0) getout = true; // endpoint is not in solid if (d1 > 0) startout = true; // if completely in front of face, no intersection if (d1 > 0 && d2 >= d1) return; if (d1 <= 0 && d2 <= 0) continue; // crosses face if (d1 > d2) { // enter f = (d1) / (d1-d2); if (f > enterfrac) { enterfrac = f; nearfrac = (d1-DIST_EPSILON) / (d1-d2); clipplane = plane; leadside = side; } } else { // leave f = (d1) / (d1-d2); if (f < leavefrac) leavefrac = f; } } if (!startout) { // original point was inside brush trace->startsolid = true; if (!getout) trace->allsolid = true; return; } if (enterfrac <= leavefrac) { if (enterfrac > -1 && enterfrac <= trace_truefraction) { if (enterfrac < 0) enterfrac = 0; trace_nearfraction = nearfrac; trace_truefraction = enterfrac; trace->plane.dist = clipplane->dist; VectorCopy(clipplane->normal, trace->plane.normal); trace->surface = &(leadside->surface->c); trace->contents = brush->contents; } } } #ifdef Q3BSPS static void CM_ClipBoxToPlanes (vec3_t trmins, vec3_t trmaxs, vec3_t p1, vec3_t p2, trace_t *trace, vec3_t plmins, vec3_t plmaxs, mplane_t *plane, int numplanes, q2csurface_t *surf) { int i, j; mplane_t *clipplane; float dist; float enterfrac, leavefrac; vec3_t ofs; float d1, d2; qboolean getout, startout; float f; // q2cbrushside_t *side, *leadside; static mplane_t bboxplanes[6] = //we change the dist, but nothing else { {{1, 0, 0}}, {{0, 1, 0}}, {{0, 0, 1}}, {{-1, 0, 0}}, {{0, -1, 0}}, {{0, 0, -1}}, }; float nearfrac=0; enterfrac = -1; leavefrac = 2; clipplane = NULL; getout = false; startout = false; // leadside = NULL; for (i=0 ; inormal[j] < 0) ofs[j] = trmaxs[j]; else ofs[j] = trmins[j]; } dist = DotProduct (ofs, plane->normal); dist = plane->dist - dist; break; capsuledist(dist,plane,trmins,trmaxs) case shape_ispoint: // special point case dist = plane->dist; break; } d1 = DotProduct (p1, plane->normal) - dist; d2 = DotProduct (p2, plane->normal) - dist; if (d2 > 0) getout = true; // endpoint is not in solid if (d1 > 0) startout = true; // if completely in front of face, no intersection if (d1 > 0 && d2 >= d1) return; if (d1 <= 0 && d2 <= 0) continue; // crosses face if (d1 > d2) { // enter f = (d1) / (d1-d2); if (f > enterfrac) { enterfrac = f; nearfrac = (d1-DIST_EPSILON) / (d1-d2); clipplane = plane; // leadside = side; } } else { // leave f = (d1) / (d1-d2); if (f < leavefrac) leavefrac = f; } } //bevel the brush axially (to match the player's bbox), in case that wasn't already done for (i=0, plane = bboxplanes; i<6 ; i++, plane++) { if (i < 3) { //positive normal dist = trmins[i]; plane->dist = plmaxs[i]; dist = plane->dist - dist; d1 = p1[i] - dist; d2 = p2[i] - dist; } else { //negative normal j = i-3; dist = -trmaxs[j]; plane->dist = -plmins[j]; dist = plane->dist - dist; d1 = -p1[j] - dist; d2 = -p2[j] - dist; } if (d2 > 0) getout = true; // endpoint is not in solid if (d1 > 0) startout = true; // if completely in front of face, no intersection if (d1 > 0 && d2 >= d1) return; if (d1 <= 0 && d2 <= 0) continue; // crosses face if (d1 > d2) { // enter f = (d1) / (d1-d2); if (f > enterfrac) { enterfrac = f; nearfrac = (d1-DIST_EPSILON) / (d1-d2); clipplane = plane; // leadside = side; } } else { // leave f = (d1) / (d1-d2); if (f < leavefrac) leavefrac = f; } } if (!startout) { // original point was inside brush trace->startsolid = true; if (!getout) trace->allsolid = true; return; } if (enterfrac <= leavefrac) { if (enterfrac > -1 && enterfrac <= trace_truefraction) { if (enterfrac < 0) enterfrac = 0; trace_nearfraction = nearfrac; trace_truefraction = enterfrac; trace->plane.dist = clipplane->dist; VectorCopy(clipplane->normal, trace->plane.normal); trace->surface = surf; trace->contents = surf->value; } } } static void Mod_Trace_Trisoup_(vecV_t *posedata, index_t *indexes, size_t numindexes, vec3_t start, vec3_t end, vec3_t mins, vec3_t maxs, trace_t *trace, q2csurface_t *surf) { size_t i; int j; float *p1, *p2, *p3; vec3_t edge1, edge2, edge3; mplane_t planes[5]; vec3_t tmins, tmaxs; for (i = 0; i < numindexes; i+=3) { p1 = posedata[indexes[i+0]]; p2 = posedata[indexes[i+1]]; p3 = posedata[indexes[i+2]]; //determine the triangle extents, and skip the triangle if we're completely out of bounds for (j = 0; j < 3; j++) { tmins[j] = p1[j]; if (tmins[j] > p2[j]) tmins[j] = p2[j]; if (tmins[j] > p3[j]) tmins[j] = p3[j]; if (trace_absmaxs[j]+(1/8.f) < tmins[j]) break; tmaxs[j] = p1[j]; if (tmaxs[j] < p2[j]) tmaxs[j] = p2[j]; if (tmaxs[j] < p3[j]) tmaxs[j] = p3[j]; if (trace_absmins[j]-(1/8.f) > tmaxs[j]) break; } //skip any triangles which are completely outside the trace bounds if (j < 3) continue; VectorSubtract(p1, p2, edge1); VectorSubtract(p3, p2, edge2); VectorSubtract(p1, p3, edge3); CrossProduct(edge1, edge2, planes[0].normal); VectorNormalize(planes[0].normal); planes[0].dist = DotProduct(p1, planes[0].normal); VectorNegate(planes[0].normal, planes[1].normal); planes[1].dist = -planes[0].dist + 4; //determine edges //FIXME: use adjacency info CrossProduct(edge1, planes[0].normal, planes[2].normal); VectorNormalize(planes[2].normal); planes[2].dist = DotProduct(p2, planes[2].normal); CrossProduct(planes[0].normal, edge2, planes[3].normal); VectorNormalize(planes[3].normal); planes[3].dist = DotProduct(p3, planes[3].normal); CrossProduct(planes[0].normal, edge3, planes[4].normal); VectorNormalize(planes[4].normal); planes[4].dist = DotProduct(p1, planes[4].normal); CM_ClipBoxToPlanes(mins, maxs, start, end, trace, tmins, tmaxs, planes, 5, surf); } } /* static void CM_ClipBoxToMesh (vec3_t mins, vec3_t maxs, vec3_t p1, vec3_t p2, trace_t *trace, mesh_t *mesh) { trace_truefraction = trace->truefraction; trace_nearfraction = trace->fraction; Mod_Trace_Trisoup_(mesh->xyz_array, mesh->indexes, mesh->numindexes, p1, p2, mins, maxs, trace, &nullsurface.c); trace->truefraction = trace_truefraction; trace->fraction = trace_nearfraction; } */ static void CM_ClipBoxToPatch (vec3_t mins, vec3_t maxs, vec3_t p1, vec3_t p2, trace_t *trace, q2cbrush_t *brush) { int i, j; mplane_t *plane, *clipplane; float enterfrac, leavefrac, nearfrac = 0; vec3_t ofs; float d1, d2; float dist; qboolean startout; float f; q2cbrushside_t *side, *leadside; if (!brush->numsides) return; enterfrac = -1; leavefrac = 2; clipplane = NULL; startout = false; leadside = NULL; for (i=0 ; inumsides ; i++) { side = brush->brushside+i; plane = side->plane; // push the plane out apropriately for mins/maxs switch(trace_shape) { default: case shape_isbox: // general box case // FIXME: use signbits into 8 way lookup for each mins/maxs for (j=0 ; j<3 ; j++) { if (plane->normal[j] < 0) ofs[j] = maxs[j]; else ofs[j] = mins[j]; } dist = DotProduct (ofs, plane->normal); dist = plane->dist - dist; break; capsuledist(dist,plane,mins,maxs) case shape_ispoint: // special point case dist = plane->dist; break; } d1 = DotProduct (p1, plane->normal) - dist; d2 = DotProduct (p2, plane->normal) - dist; // if completely in front of face, no intersection if (d1 > 0 && d2 >= d1) return; if (d1 > 0) startout = true; if (d1 <= 0 && d2 <= 0) continue; // crosses face if (d1 > d2) { // enter f = (d1) / (d1-d2); if (f > enterfrac) { enterfrac = f; nearfrac = (d1-DIST_EPSILON) / (d1-d2); clipplane = plane; leadside = side; } } else { // leave f = (d1) / (d1-d2); if (f < leavefrac) leavefrac = f; } } if (!startout) { trace->startsolid = true; return; // original point is inside the patch } if (nearfrac <= leavefrac) { if (leadside && leadside->surface && enterfrac <= trace_truefraction) { if (enterfrac < 0) enterfrac = 0; trace_truefraction = enterfrac; trace_nearfraction = nearfrac; trace->plane.dist = clipplane->dist; VectorCopy(clipplane->normal, trace->plane.normal); trace->surface = &leadside->surface->c; trace->contents = brush->contents; } else if (enterfrac < trace_truefraction) leavefrac=0; } } #endif /* ================ CM_TestBoxInBrush ================ */ static void CM_TestBoxInBrush (vec3_t mins, vec3_t maxs, vec3_t p1, trace_t *trace, q2cbrush_t *brush) { int i, j; mplane_t *plane; float dist; vec3_t ofs; float d1; q2cbrushside_t *side; if (!brush->numsides) return; for (i=0 ; inumsides ; i++) { side = brush->brushside+i; plane = side->plane; switch(trace_shape) { default: case shape_isbox: // general box case // push the plane out apropriately for mins/maxs // FIXME: use signbits into 8 way lookup for each mins/maxs for (j=0 ; j<3 ; j++) { if (plane->normal[j] < 0) ofs[j] = maxs[j]; else ofs[j] = mins[j]; } dist = DotProduct (ofs, plane->normal); dist = plane->dist - dist; break; capsuledist(dist,plane,mins,maxs) case shape_ispoint: dist = plane->dist; break; } d1 = DotProduct (p1, plane->normal) - dist; // if completely in front of face, no intersection if (d1 > 0) return; } // inside this brush trace->startsolid = trace->allsolid = true; trace->contents |= brush->contents; } #ifdef Q3BSPS static void CM_TestBoxInPatch (vec3_t mins, vec3_t maxs, vec3_t p1, trace_t *trace, q2cbrush_t *brush) { int i, j; mplane_t *plane; vec3_t ofs, ofs2; float dist, thickness; float d1; q2cbrushside_t *side; if (!brush->numsides) return; i = 0; //front plane { side = brush->brushside+i; plane = side->plane; switch(trace_shape) { default: case shape_isbox: for (j=0 ; j<3 ; j++) { if (plane->normal[j] < 0) ofs[j] = maxs[j], ofs2[j] = mins[j]; else ofs[j] = mins[j], ofs2[j] = maxs[j]; } dist = DotProduct (ofs, plane->normal); thickness = DotProduct (ofs2, plane->normal)-dist; dist = plane->dist - dist; break; case shape_iscapsule: dist = DotProduct(trace_up, plane->normal); thickness = dist*(trace_capsulesize[(dist<0)?2:1]) + trace_capsulesize[0]*2; dist = dist*(trace_capsulesize[(dist<0)?1:2]) - trace_capsulesize[0]; dist = plane->dist - dist; break; case shape_ispoint: dist = plane->dist; thickness = 0; break; } d1 = DotProduct (p1, plane->normal) - dist; // if completely in front of face, no intersection if (d1 > 0) return; //point is behind the front plane, so no real intersection. if (thickness < 0.25) thickness = 0.25; //FIXME: patches should probably be infinitely thin, but that makes stuff messy. if (d1 < -thickness) return; } for (i=1 ; inumsides ; i++) { side = brush->brushside+i; plane = side->plane; switch(trace_shape) { default: case shape_isbox: // general box case // push the plane out apropriately for mins/maxs // FIXME: use signbits into 8 way lookup for each mins/maxs for (j=0 ; j<3 ; j++) { if (plane->normal[j] < 0) ofs[j] = maxs[j]; else ofs[j] = mins[j]; } dist = DotProduct (ofs, plane->normal); dist = plane->dist - dist; break; capsuledist(dist,plane,mins,maxs) case shape_ispoint: dist = plane->dist; break; } d1 = DotProduct (p1, plane->normal) - dist; // if completely in front of face, no intersection if (d1 > 0) return; } // inside this patch trace->startsolid = trace->allsolid = true; trace->contents = brush->contents; } #endif /* ================ CM_TraceToLeaf ================ */ static void CM_TraceToLeaf (cminfo_t *prv, mleaf_t *leaf) { int k; q2cbrush_t *b; #ifdef Q3BSPS int patchnum, j; q3cpatch_t *patch; q3cmesh_t *cmesh; #endif if ( !(leaf->contents & trace_contents)) return; // trace line against all brushes in the leaf for (k=0 ; knumleafbrushes ; k++) { b = prv->leafbrushes[leaf->firstleafbrush+k]; if (b->checkcount == checkcount) continue; // already checked this brush in another leaf b->checkcount = checkcount; if ( !(b->contents & trace_contents)) continue; if (!BoundsIntersect(b->absmins, b->absmaxs, trace_absmins, trace_absmaxs)) continue; CM_ClipBoxToBrush (trace_mins, trace_maxs, trace_start, trace_end, &trace_trace, b); if (trace_nearfraction <= 0) return; } #ifdef Q3BSPS if (!prv->mapisq3 || map_noCurves.value) return; // trace line against all patches in the leaf for (k = 0; k < leaf->numleafpatches; k++) { patchnum = prv->leafpatches[leaf->firstleafpatch+k]; patch = &prv->patches[patchnum]; if (patch->checkcount == checkcount) continue; // already checked this patch in another leaf patch->checkcount = checkcount; if ( !(patch->surface->c.value & trace_contents) ) continue; if ( !BoundsIntersect(patch->absmins, patch->absmaxs, trace_absmins, trace_absmaxs) ) continue; for (j = 0; j < patch->numfacets; j++) { CM_ClipBoxToPatch (trace_mins, trace_maxs, trace_start, trace_end, &trace_trace, &patch->facets[j]); if (trace_nearfraction<=0) return; } } for (k = 0; k < leaf->numleafcmeshes; k++) { patchnum = prv->leafcmeshes[leaf->firstleafcmesh+k]; cmesh = &prv->cmeshes[patchnum]; if (cmesh->checkcount == checkcount) continue; // already checked this patch in another leaf cmesh->checkcount = checkcount; if ( !(cmesh->surface->c.value & trace_contents) ) continue; if ( !BoundsIntersect(cmesh->absmins, cmesh->absmaxs, trace_absmins, trace_absmaxs) ) continue; Mod_Trace_Trisoup_(cmesh->xyz_array, cmesh->indicies, cmesh->numincidies, trace_start, trace_end, trace_mins, trace_maxs, &trace_trace, &cmesh->surface->c); if (trace_nearfraction<=0) return; } #endif } /* ================ CM_TestInLeaf ================ */ static void CM_TestInLeaf (cminfo_t *prv, mleaf_t *leaf) { int k; q2cbrush_t *b; #ifdef Q3BSPS int patchnum, j; q3cmesh_t *cmesh; q3cpatch_t *patch; #endif if ( !(leaf->contents & trace_contents)) return; // trace line against all brushes in the leaf for (k=0 ; knumleafbrushes ; k++) { b = prv->leafbrushes[leaf->firstleafbrush+k]; if (b->checkcount == checkcount) continue; // already checked this brush in another leaf b->checkcount = checkcount; if (!(b->contents & trace_contents)) continue; if (!BoundsIntersect(b->absmins, b->absmaxs, trace_absmins, trace_absmaxs)) continue; CM_TestBoxInBrush (trace_mins, trace_maxs, trace_start, &trace_trace, b); if (!trace_trace.fraction) return; } #ifdef Q3BSPS if (!prv->mapisq3 || map_noCurves.value) return; // trace line against all patches in the leaf for (k = 0; k < leaf->numleafpatches; k++) { patchnum = prv->leafpatches[leaf->firstleafpatch+k]; patch = &prv->patches[patchnum]; if (patch->checkcount == checkcount) continue; // already checked this patch in another leaf patch->checkcount = checkcount; if ( !(patch->surface->c.value & trace_contents) ) continue; if ( !BoundsIntersect(patch->absmins, patch->absmaxs, trace_absmins, trace_absmaxs) ) continue; for (j = 0; j < patch->numfacets; j++) { CM_TestBoxInPatch (trace_mins, trace_maxs, trace_start, &trace_trace, &patch->facets[j]); if (!trace_trace.fraction) return; } } for (k = 0; k < leaf->numleafcmeshes; k++) { patchnum = prv->leafcmeshes[leaf->firstleafcmesh+k]; cmesh = &prv->cmeshes[patchnum]; if (cmesh->checkcount == checkcount) continue; // already checked this patch in another leaf cmesh->checkcount = checkcount; if ( !(cmesh->surface->c.value & trace_contents) ) continue; if ( !BoundsIntersect(cmesh->absmins, cmesh->absmaxs, trace_absmins, trace_absmaxs) ) continue; Mod_Trace_Trisoup_(cmesh->xyz_array, cmesh->indicies, cmesh->numincidies, trace_start, trace_end, trace_mins, trace_maxs, &trace_trace, &cmesh->surface->c); if (trace_nearfraction<=0) return; } #endif } /* ================== CM_RecursiveHullCheck ================== */ static void CM_RecursiveHullCheck (model_t *mod, int num, float p1f, float p2f, vec3_t p1, vec3_t p2) { mnode_t *node; mplane_t *plane; float t1, t2, offset; float frac, frac2; float idist; int i; vec3_t mid; int side; float midf; if (trace_truefraction <= p1f) return; // already hit something nearer // if < 0, we are in a leaf node if (num < 0) { CM_TraceToLeaf (mod->meshinfo, &mod->leafs[-1-num]); return; } // // find the point distances to the seperating plane // and the offset for the size of the box // node = mod->nodes + num; plane = node->plane; if (plane->type < 3) { t1 = p1[plane->type] - plane->dist; t2 = p2[plane->type] - plane->dist; offset = trace_extents[plane->type]; } else { t1 = DotProduct (plane->normal, p1) - plane->dist; t2 = DotProduct (plane->normal, p2) - plane->dist; if (trace_shape == shape_ispoint) offset = 0; else offset = fabs(trace_extents[0]*plane->normal[0]) + fabs(trace_extents[1]*plane->normal[1]) + fabs(trace_extents[2]*plane->normal[2]); } #if 0 CM_RecursiveHullCheck (node->childnum[0], p1f, p2f, p1, p2); CM_RecursiveHullCheck (node->childnum[1], p1f, p2f, p1, p2); return; #endif // see which sides we need to consider if (t1 >= offset && t2 >= offset) { CM_RecursiveHullCheck (mod, node->childnum[0], p1f, p2f, p1, p2); return; } if (t1 < -offset && t2 < -offset) { CM_RecursiveHullCheck (mod, node->childnum[1], p1f, p2f, p1, p2); return; } // put the crosspoint DIST_EPSILON pixels on the near side if (t1 < t2) { idist = 1.0/(t1-t2); side = 1; frac2 = (t1 + offset + DIST_EPSILON)*idist; frac = (t1 - offset + DIST_EPSILON)*idist; } else if (t1 > t2) { idist = 1.0/(t1-t2); side = 0; frac2 = (t1 - offset - DIST_EPSILON)*idist; frac = (t1 + offset + DIST_EPSILON)*idist; } else { side = 0; frac = 1; frac2 = 0; } // move up to the node 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]); CM_RecursiveHullCheck (mod, node->childnum[side], p1f, midf, p1, mid); // go past the node if (frac2 < 0) frac2 = 0; if (frac2 > 1) frac2 = 1; midf = p1f + (p2f - p1f)*frac2; for (i=0 ; i<3 ; i++) mid[i] = p1[i] + frac2*(p2[i] - p1[i]); CM_RecursiveHullCheck (mod, node->childnum[side^1], midf, p2f, mid, p2); } //====================================================================== /* ================== CM_BoxTrace ================== */ static trace_t CM_BoxTrace (model_t *mod, const vec3_t start, const vec3_t end, const vec3_t mins, const vec3_t maxs, qboolean capsule, int brushmask) { int i; vec3_t point; checkcount++; // for multi-check avoidance // fill in a default trace memset (&trace_trace, 0, sizeof(trace_trace)); trace_truefraction = 1; trace_nearfraction = 1; trace_trace.fraction = 1; trace_trace.truefraction = 1; trace_trace.surface = &(nullsurface.c); if (!mod) // map not loaded return trace_trace; trace_contents = brushmask; VectorCopy (start, trace_start); VectorCopy (end, trace_end); VectorCopy (mins, trace_mins); VectorCopy (maxs, trace_maxs); if (1) { VectorAdd(trace_maxs, trace_mins, point); VectorScale(point, 0.5, point); VectorAdd(trace_start, point, trace_start); VectorAdd(trace_end, point, trace_end); VectorSubtract(trace_mins, point, trace_mins); VectorSubtract(trace_maxs, point, trace_maxs); } // build a bounding box of the entire move (for patches) ClearBounds (trace_absmins, trace_absmaxs); //determine the type of trace that we're going to use, and the max extents if (trace_mins[0] == 0 && trace_mins[1] == 0 && trace_mins[2] == 0 && trace_maxs[0] == 0 && trace_maxs[1] == 0 && trace_maxs[2] == 0) { trace_shape = shape_ispoint; VectorSet (trace_extents, 1/32.0, 1/32.0, 1/32.0); //acedemic AddPointToBounds (trace_start, trace_absmins, trace_absmaxs); AddPointToBounds (trace_end, trace_absmins, trace_absmaxs); } else if (capsule) { float ext; trace_shape = shape_iscapsule; //determine the capsule sizes trace_capsulesize[0] = ((trace_maxs[0]-trace_mins[0]) + (trace_maxs[1]-trace_mins[1]))/4.0; trace_capsulesize[1] = trace_maxs[2]; trace_capsulesize[2] = trace_mins[2]; //make sure the mins_z/maxs_z isn't screwed. // if (trace_capsulesize[1]-trace_capsulesize[2] < trace_capsulesize[0]) // trace_capsulesize[1] = trace_capsulesize[0]+trace_capsulesize[2]; ext = (trace_capsulesize[1] > -trace_capsulesize[2])?trace_capsulesize[1]:-trace_capsulesize[2]; trace_capsulesize[1] -= trace_capsulesize[0]; trace_capsulesize[2] += trace_capsulesize[0]; trace_extents[0] = ext+1; trace_extents[1] = ext+1; trace_extents[2] = ext+1; //determine the total range VectorSubtract (trace_start, trace_extents, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorAdd (trace_start, trace_extents, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorSubtract (trace_end, trace_extents, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorAdd (trace_end, trace_extents, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); } else { VectorAdd (trace_start, trace_mins, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorAdd (trace_start, trace_maxs, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorAdd (trace_end, trace_mins, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); VectorAdd (trace_end, trace_maxs, point); AddPointToBounds (point, trace_absmins, trace_absmaxs); trace_shape = shape_isbox; trace_extents[0] = ((-trace_mins[0] > trace_maxs[0]) ? -trace_mins[0] : trace_maxs[0])+1; trace_extents[1] = ((-trace_mins[1] > trace_maxs[1]) ? -trace_mins[1] : trace_maxs[1])+1; trace_extents[2] = ((-trace_mins[2] > trace_maxs[2]) ? -trace_mins[2] : trace_maxs[2])+1; } trace_absmins[0] -= 1.0; trace_absmins[1] -= 1.0; trace_absmins[2] -= 1.0; trace_absmaxs[0] += 1.0; trace_absmaxs[1] += 1.0; trace_absmaxs[2] += 1.0; #if 0 if (0) { //treat *ALL* tests against the actual geometry instead of using any brushes. //also ignores the bsp etc. not fast. testing only. trace_ispoint = trace_mins[0] == 0 && trace_mins[1] == 0 && trace_mins[2] == 0 && trace_maxs[0] == 0 && trace_maxs[1] == 0 && trace_maxs[2] == 0; for (i = 0; i < mod->numsurfaces; i++) { CM_ClipBoxToMesh(trace_mins, trace_maxs, trace_start, trace_end, &trace_trace, mod->surfaces[i].mesh); } } else if (0) { trace_ispoint = trace_mins[0] == 0 && trace_mins[1] == 0 && trace_mins[2] == 0 && trace_maxs[0] == 0 && trace_maxs[1] == 0 && trace_maxs[2] == 0; for (i = 0; i < mod->numleafs; i++) CM_TraceToLeaf(&mod->leafs[i]); } else #endif // // check for position test special case // if (start[0] == end[0] && start[1] == end[1] && start[2] == end[2]) { int leafs[1024]; int i, numleafs; int topnode; numleafs = CM_BoxLeafnums_headnode (mod, trace_absmins, trace_absmaxs, leafs, sizeof(leafs)/sizeof(leafs[0]), mod->hulls[0].firstclipnode, &topnode); for (i=0 ; imeshinfo, &mod->leafs[leafs[i]]); if (trace_trace.allsolid) break; } VectorCopy (start, trace_trace.endpos); return trace_trace; } // // general aabb trace // else { CM_RecursiveHullCheck (mod, mod->hulls[0].firstclipnode, 0, 1, trace_start, trace_end); } if (trace_nearfraction == 1) { trace_trace.fraction = 1; VectorCopy (end, trace_trace.endpos); } else { if (trace_nearfraction<0) trace_nearfraction=0; trace_trace.fraction = trace_nearfraction; for (i=0 ; i<3 ; i++) trace_trace.endpos[i] = start[i] + trace_trace.fraction * (end[i] - start[i]); } return trace_trace; } static qboolean BM_NativeTrace(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 contents, trace_t *trace) { int i; memset (trace, 0, sizeof(*trace)); trace_truefraction = 1; trace_nearfraction = 1; trace->fraction = 1; trace->truefraction = 1; trace->surface = &(nullsurface.c); if (contents & FTECONTENTS_BODY) { trace_contents = contents; VectorCopy (start, trace_start); VectorCopy (end, trace_end); VectorCopy (mins, trace_mins); VectorCopy (maxs, trace_maxs); if (trace_mins[0] == 0 && trace_mins[1] == 0 && trace_mins[2] == 0 && trace_maxs[0] == 0 && trace_maxs[1] == 0 && trace_maxs[2] == 0) trace_shape = shape_ispoint; else if (capsule) trace_shape = shape_iscapsule; else trace_shape = shape_isbox; CM_ClipBoxToBrush (trace_mins, trace_maxs, trace_start, trace_end, trace, &box_brush); } if (trace_nearfraction == 1) { trace->fraction = 1; VectorCopy (trace_end, trace->endpos); } else { if (trace_nearfraction<0) trace_nearfraction=0; trace->fraction = trace_nearfraction; trace->truefraction = trace_truefraction; for (i=0 ; i<3 ; i++) trace->endpos[i] = trace_start[i] + trace->fraction * (trace_end[i] - trace_start[i]); } return trace->fraction != 1; } static qboolean CM_NativeTrace(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 contents, trace_t *trace) { if (axis) { vec3_t start_l; vec3_t end_l; start_l[0] = DotProduct(start, axis[0]); start_l[1] = DotProduct(start, axis[1]); start_l[2] = DotProduct(start, axis[2]); end_l[0] = DotProduct(end, axis[0]); end_l[1] = DotProduct(end, axis[1]); end_l[2] = DotProduct(end, axis[2]); VectorSet(trace_up, axis[0][2], -axis[1][2], axis[2][2]); *trace = CM_BoxTrace(model, start_l, end_l, mins, maxs, capsule, contents); #ifdef TERRAIN if (model->terrain) { trace_t hmt; Heightmap_Trace(model, forcehullnum, framestate, NULL, start, end, mins, maxs, capsule, contents, &hmt); if (hmt.fraction < trace->fraction) *trace = hmt; } #endif 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 { VectorSet(trace_up, 0, 0, 1); *trace = CM_BoxTrace(model, start, end, mins, maxs, capsule, contents); #ifdef TERRAIN if (model->terrain) { trace_t hmt; Heightmap_Trace(model, forcehullnum, framestate, NULL, start, end, mins, maxs, capsule, contents, &hmt); if (hmt.fraction < trace->fraction) *trace = hmt; } #endif } return trace->fraction != 1; } /* =============================================================================== PVS / PHS =============================================================================== */ /* =================== CM_DecompressVis =================== */ /* qbyte *Mod_Q2DecompressVis (qbyte *in, model_t *model) { static qbyte decompressed[MAX_MAP_LEAFS/8]; int c; qbyte *out; int row; row = (model->vis->numclusters+7)>>3; out = decompressed; if (!in) { // no vis info, so make all visible while (row) { *out++ = 0xff; row--; } return decompressed; } do { if (*in) { *out++ = *in++; continue; } c = in[1]; in += 2; while (c) { *out++ = 0; c--; } } while (out - decompressed < row); return decompressed; } #define DVIS_PVS 0 #define DVIS_PHS 1 qbyte *Mod_ClusterPVS (int cluster, model_t *model) { if (cluster == -1 || !model->vis) return mod_novis; return Mod_Q2DecompressVis ( (qbyte *)model->vis + model->vis->bitofs[cluster][DVIS_PVS], model); } */ static void CM_DecompressVis (model_t *mod, qbyte *in, qbyte *out, qboolean merge) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int c; qbyte *out_p; int row; row = (mod->numclusters+7)>>3; out_p = out; if (!in || !prv->numvisibility) { // no vis info, so make all visible while (row) { *out_p++ = 0xff; row--; } return; } if (merge) { do { if (*in) { *out_p++ |= *in++; continue; } out_p += in[1]; in += 2; } while (out_p - out < row); } else { do { if (*in) { *out_p++ = *in++; continue; } c = in[1]; in += 2; if ((out_p - out) + c > row) { c = row - (out_p - out); Con_DPrintf ("warning: Vis decompression overrun\n"); } while (c) { *out_p++ = 0; c--; } } while (out_p - out < row); } } static pvsbuffer_t pvsrow; static pvsbuffer_t phsrow; qbyte *CM_ClusterPVS (model_t *mod, int cluster, pvsbuffer_t *buffer, pvsmerge_t merge) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (!buffer) buffer = &pvsrow; if (buffer->buffersize < mod->pvsbytes) buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=mod->pvsbytes); if (mod->fromgame == fg_quake2) { if (cluster == -1) memset (buffer->buffer, 0, (mod->numclusters+7)>>3); else CM_DecompressVis (mod, ((qbyte*)prv->q2vis) + prv->q2vis->bitofs[cluster][DVIS_PVS], buffer->buffer, merge==PVM_MERGE); return buffer->buffer; } else { if (cluster != -1 && prv->q3pvs->numclusters) { if (merge == PVM_FAST) return (qbyte *)prv->q3pvs->data + cluster * prv->q3pvs->rowsize; else if (merge == PVM_REPLACE) memcpy(buffer->buffer, prv->q3pvs->data + cluster * prv->q3pvs->rowsize, mod->pvsbytes); else { int c; char *in = prv->q3pvs->data + cluster * prv->q3pvs->rowsize; for (c = 0; c < mod->pvsbytes; c+=4) *(int*)&buffer->buffer[c] |= *(int*)&in[c]; } } else { if (merge != PVM_MERGE) memset (buffer->buffer, 0, (mod->numclusters+7)>>3); } return buffer->buffer; } } qbyte *CM_ClusterPHS (model_t *mod, int cluster, pvsbuffer_t *buffer) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (!buffer) buffer = &phsrow; if (buffer->buffersize < mod->pvsbytes) buffer->buffer = BZ_Realloc(buffer->buffer, buffer->buffersize=mod->pvsbytes); if (mod->fromgame != fg_quake2) { if (cluster != -1 && prv->q3phs->numclusters) { if (prv->phscalced && !(prv->phscalced[cluster>>3] & (1<<(cluster&7)))) CalcClusterPHS(prv, cluster); return (qbyte *)prv->q3phs->data + cluster * prv->q3phs->rowsize; } else { memset (buffer->buffer, 0, (mod->numclusters+7)>>3); return buffer->buffer; } } if (cluster == -1) memset (buffer->buffer, 0, (mod->numclusters+7)>>3); else CM_DecompressVis (mod, ((qbyte*)prv->q2vis) + prv->q2vis->bitofs[cluster][DVIS_PHS], buffer->buffer, false); return buffer->buffer; } static unsigned int SV_Q2BSP_FatPVS (model_t *mod, const vec3_t org, pvsbuffer_t *result, qboolean merge) { int leafs[64]; int i, j, count; vec3_t mins, maxs; for (i=0 ; i<3 ; i++) { mins[i] = org[i] - 8; maxs[i] = org[i] + 8; } count = CM_BoxLeafnums (mod, mins, maxs, leafs, countof(leafs), NULL); if (count < 1) Sys_Error ("SV_Q2FatPVS: count < 1"); // convert leafs to clusters for (i=0 ; ibuffersize < mod->pvsbytes) result->buffer = BZ_Realloc(result->buffer, result->buffersize=mod->pvsbytes); if (count == 1 && leafs[0] == -1) { //if the only leaf is the outside then broadcast it. memset(result->buffer, 0xff, mod->pvsbytes); i = count; } else { i = 0; if (!merge) mod->funcs.ClusterPVS(mod, leafs[i++], result, PVM_REPLACE); // or in all the other leaf bits for ( ; ifuncs.ClusterPVS(mod, leafs[i], result, PVM_MERGE); } } return mod->pvsbytes; } static int clientarea; unsigned int Q23BSP_FatPVS(model_t *mod, const vec3_t org, pvsbuffer_t *fte_restrict buffer, qboolean merge) {//fixme: this doesn't add areas int leafnum; leafnum = CM_PointLeafnum (mod, org); clientarea = CM_LeafArea (mod, leafnum); return SV_Q2BSP_FatPVS (mod, org, buffer, merge); } qboolean Q23BSP_EdictInFatPVS(model_t *mod, const pvscache_t *ent, const qbyte *pvs, const int *areas) { int i,l; int nullarea = (mod->fromgame == fg_quake2)?0:-1; if (areas) { for (i = 1; ; i++) { if (i > areas[0]) return false; //none of the camera's areas could see the entity if (areas[i] == ent->areanum) { if (areas[i] != nullarea) break; //else entity is fully outside the world, invisible to all... } else if (CM_AreasConnected (mod, areas[i], ent->areanum)) break; // doors can legally straddle two areas, so // we may need to check another one else if (ent->areanum2 != nullarea && CM_AreasConnected (mod, areas[i], ent->areanum2)) break; } } if (ent->num_leafs == -1) { // too many leafs for individual check, go by headnode if (!CM_HeadnodeVisible (mod, ent->headnode, pvs)) return false; } else { // check individual leafs for (i=0 ; i < ent->num_leafs ; i++) { l = ent->leafnums[i]; if (pvs[l >> 3] & (1 << (l&7) )) break; } if (i == ent->num_leafs) return false; // not visible } return true; } /* =============================================================================== AREAPORTALS =============================================================================== */ static void FloodArea_r (cminfo_t *prv, size_t areaidx, int floodnum) { size_t i; careaflood_t *flood = &prv->areaflood[areaidx]; if (flood->floodvalid == prv->floodvalid) { if (flood->floodnum == floodnum) return; Con_Printf ("FloodArea_r: reflooded\n"); return; } flood->floodnum = floodnum; flood->floodvalid = prv->floodvalid; switch(prv->mapisq3) { case true: #ifdef Q3BSPS for (i=0 ; inumareas ; i++) { if (prv->q3areas[areaidx].numareaportals[i]>0) FloodArea_r (prv, i, floodnum); } #endif break; case false: #ifdef Q2BSPS { q2carea_t *area = &prv->q2areas[areaidx]; q2dareaportal_t *p = &prv->q2areaportals[area->firstareaportal]; for (i=0 ; inumareaportals ; i++, p++) { if (prv->q2portalopen[p->portalnum]) FloodArea_r (prv, p->otherarea, floodnum); } } #endif break; } } /* ==================== FloodAreaConnections ==================== */ static void FloodAreaConnections (cminfo_t *prv) { size_t i; int floodnum; // all current floods are now invalid prv->floodvalid++; floodnum = 0; // area 0 is not used for (i=0 ; inumareas ; i++) { if (prv->areaflood[i].floodvalid == prv->floodvalid) continue; // already flooded into floodnum++; FloodArea_r (prv, i, floodnum); } } #ifdef Q2BSPS void CMQ2_SetAreaPortalState (model_t *mod, unsigned int portalnum, qboolean open) { cminfo_t *prv; if (!mod) return; prv = (cminfo_t*)mod->meshinfo; if (prv->mapisq3) return; if (portalnum > prv->numq2areaportals) Host_Error ("areaportal > numareaportals"); if (prv->q2portalopen[portalnum] == open) return; prv->q2portalopen[portalnum] = open; FloodAreaConnections (prv); return; } #endif #ifdef Q3BSPS void CMQ3_SetAreaPortalState (model_t *mod, unsigned int area1, unsigned int area2, qboolean open) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (!prv->mapisq3) return; // Host_Error ("CMQ3_SetAreaPortalState on non-q3 map"); if (area1 >= prv->numareas || area2 >= prv->numareas) Host_Error ("CMQ3_SetAreaPortalState: area > numareas"); if (open) { prv->q3areas[area1].numareaportals[area2]++; prv->q3areas[area2].numareaportals[area1]++; } else { prv->q3areas[area1].numareaportals[area2]--; prv->q3areas[area2].numareaportals[area1]--; } FloodAreaConnections(prv); } #endif qboolean VARGS CM_AreasConnected (model_t *mod, unsigned int area1, unsigned int area2) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (map_noareas.value) return true; if (area1 == ~0 || area2 == ~0) return area1 == area2; if (area1 > prv->numareas || area2 > prv->numareas) Host_Error ("area > numareas"); if (prv->areaflood[area1].floodnum == prv->areaflood[area2].floodnum) return true; return false; } /* ================= CM_WriteAreaBits Writes a length qbyte followed by a bit vector of all the areas that area in the same flood as the area parameter This is used by the client refreshes to cull visibility ================= */ int CM_WriteAreaBits (model_t *mod, qbyte *buffer, int area, qboolean merge) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; int i; int floodnum; int bytes; bytes = (prv->numareas+7)>>3; if (map_noareas.value || (area < 0 && !merge)) { // for debugging, send everything if (!merge) memset (buffer, 255, bytes); } else { if (!merge) memset (buffer, 0, bytes); floodnum = prv->areaflood[area].floodnum; for (i=0 ; inumareas ; i++) { if (prv->areaflood[i].floodnum == floodnum || !area) buffer[i>>3] |= 1<<(i&7); } } return bytes; } /* =================== CM_WritePortalState Returns a size+pointer to the data that needs to be written into a saved game. =================== */ size_t CM_WritePortalState (model_t *mod, void **data) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (mod->type == mod_brush && (mod->fromgame == fg_quake2 || mod->fromgame == fg_quake3)) { switch(prv->mapisq3) { #ifdef Q3BSPS case true: //endian issues. oh well. *data = prv->q3areas; return sizeof(prv->q3areas); #endif #ifdef Q2BSPS case false: *data = prv->q2portalopen; return sizeof(prv->q2portalopen); #endif default: break; } } *data = NULL; return 0; } /* =================== CM_ReadPortalState Reads the portal state from a savegame file and recalculates the area connections =================== */ qofs_t CM_ReadPortalState (model_t *mod, qbyte *ptr, qofs_t ptrsize) { cminfo_t *prv = (cminfo_t*)mod->meshinfo; if (mod->type == mod_brush && (mod->fromgame == fg_quake2 || mod->fromgame == fg_quake3)) { switch(prv->mapisq3) { #ifdef Q3BSPS case 1: if (ptrsize < sizeof(prv->q3areas)) Con_Printf("CM_ReadPortalState() expected %u, but only %u available\n",(unsigned int)sizeof(prv->q3areas),(unsigned int)ptrsize); else { memcpy(prv->q3areas, ptr, sizeof(prv->q3areas)); FloodAreaConnections (prv); return sizeof(prv->q3areas); } break; #endif #ifdef Q2BSPS case 0: if (ptrsize < sizeof(prv->q2portalopen)) Con_Printf("CM_ReadPortalState() expected %u, but only %u available\n",(unsigned int)sizeof(prv->q2portalopen),(unsigned int)ptrsize); else { memcpy(prv->q2portalopen, ptr, sizeof(prv->q2portalopen)); FloodAreaConnections (prv); return sizeof(prv->q2portalopen); } break; #endif default: break; } } return 0; } /* ============= CM_HeadnodeVisible Returns true if any leaf under headnode has a cluster that is potentially visible ============= */ qboolean CM_HeadnodeVisible (model_t *mod, int nodenum, const qbyte *visbits) { int leafnum; int cluster; mnode_t *node; if (nodenum < 0) { leafnum = -1-nodenum; cluster = mod->leafs[leafnum].cluster; if (cluster == -1) return false; if (visbits[cluster>>3] & (1<<(cluster&7))) return true; return false; } node = &mod->nodes[nodenum]; if (CM_HeadnodeVisible(mod, node->childnum[0], visbits)) return true; return CM_HeadnodeVisible(mod, node->childnum[1], visbits); } unsigned int Q2BSP_PointContents(model_t *mod, const vec3_t axis[3], const vec3_t p) { int pc; pc = CM_PointContents (mod, p); return pc; } qboolean QDECL Mod_LoadQ2BrushModel (model_t *mod, void *buffer, size_t fsize) { mod->fromgame = fg_quake2; return CM_LoadMap(mod, buffer, fsize, true) != NULL; } void CM_Init(void) //register cvars. { #define MAPOPTIONS "Map Cvar Options" Cvar_Register(&map_noareas, MAPOPTIONS); Cvar_Register(&map_noCurves, MAPOPTIONS); Cvar_Register(&map_autoopenportals, MAPOPTIONS); Cvar_Register(&q3bsp_surf_meshcollision_flag, MAPOPTIONS); Cvar_Register(&q3bsp_surf_meshcollision_force, MAPOPTIONS); Cvar_Register(&q3bsp_mergeq3lightmaps, MAPOPTIONS); Cvar_Register(&q3bsp_ignorestyles, MAPOPTIONS); Cvar_Register(&q3bsp_bihtraces, MAPOPTIONS); Cvar_Register(&r_subdivisions, MAPOPTIONS); CM_InitBoxHull (); } #endif