1
0
mirror of https://github.com/zenorogue/hyperrogue.git synced 2024-11-24 21:37:18 +00:00
hyperrogue/floorshapes.cpp
2021-06-17 09:44:48 +02:00

1312 lines
41 KiB
C++

// Hyperbolic Rogue - Floor Shapes
// Copyright (C) 2011-2019 Zeno Rogue, see 'hyper.cpp' for details
/** \file floorshapes.cpp
* \brief Adjusting the floor shapes to various geometries.
*/
#include "hyper.h"
namespace hr {
#if CAP_SHAPES
#if HDR
struct qfloorinfo {
transmatrix spin;
const struct hpcshape *shape;
floorshape *fshape;
struct textureinfo *tinf;
int usershape;
};
extern qfloorinfo qfi;
#endif
EX vector<basic_textureinfo> floor_texture_vertices;
EX vector<glvertex> floor_texture_map;
EX struct renderbuffer *floor_textures;
void geometry_information::init_floorshapes() {
all_escher_floorshapes.clear();
all_plain_floorshapes = {
&shFloor, &shMFloor, &shMFloor2, &shMFloor3, &shMFloor4,
&shFullFloor, &shBigTriangle, &shTriheptaFloor, &shBigHepta
};
for(auto s: all_plain_floorshapes) s->is_plain = true;
auto init_escher = [this] (escher_floorshape& sh, int s0, int s1, int noft, int s2) {
sh.shapeid0 = s0;
sh.shapeid1 = s1;
sh.noftype = noft;
sh.shapeid2 = s2;
sh.scale = 1;
sh.is_plain = false;
all_escher_floorshapes.push_back(&sh);
};
init_escher(shStarFloor, 1, 2, 0, 0);
init_escher(shCloudFloor, 3, 4, 0, 0);
init_escher(shCrossFloor, 5, 6, 2, 54);
init_escher(shChargedFloor, 7, 385, 1, 10);
init_escher(shSStarFloor, 11, 12, 0, 0);
init_escher(shOverFloor, 13, 15, 1, 14);
init_escher(shTriFloor, 17, 18, 0, 385);
init_escher(shFeatherFloor, 19, 21, 1, 20);
init_escher(shBarrowFloor, 23, 24, 1, 25);
init_escher(shNewFloor, 26, 27, 2, 54);
init_escher(shTrollFloor, 28, 29, 0, 0);
init_escher(shButterflyFloor, 325, 326, 1, 178);
init_escher(shLavaFloor, 359, 360, 1, 178);
init_escher(shLavaSeabed, 386, 387, 1, 178);
init_escher(shSeabed, 334, 335, 0, 0);
init_escher(shCloudSeabed, 336, 337, 0, 0);
init_escher(shCaveSeabed, 338, 339, 2, 54);
init_escher(shPalaceFloor, 45, 46, 0, 385);
init_escher(shDemonFloor, 51, 50, 1, 178);
init_escher(shCaveFloor, 52, 53, 2, 54);
init_escher(shDesertFloor, 55, 56, 0, 4);
init_escher(shPowerFloor, 57, 58, 0, 12); /* dragon */
init_escher(shRoseFloor, 174, 175, 1, 173);
init_escher(shSwitchFloor, 377, 378, 1, 379);
init_escher(shTurtleFloor, 176, 177, 1, 178);
for(int i: {0,1,2})
init_escher(shRedRockFloor[i], 55, 56, 0, 0);
init_escher(shDragonFloor, 181, 182, 2, 183); /* dragon */
int ids = 0;
for(auto sh: all_plain_floorshapes) sh->id = ids++;
for(auto sh: all_escher_floorshapes) sh->id = ids++;
}
typedef pair<transmatrix, vector<transmatrix>> matrixitem;
struct mesher {
eGeometry g;
int sym;
ld bspi;
hyperpoint lcorner, rcorner, mfar[2], vfar[4];
};
mesher msh(eGeometry g, int sym, ld main, ld v0, ld v1, ld bspi, ld scale) {
main *= scale; v0 *= scale; v1 *= scale;
mesher m;
m.sym = sym;
m.bspi = bspi;
dynamicval<eGeometry> dg(geometry, g);
hyperpoint rot = xpush(v0) * xspinpush0(M_PI - M_PI/sym, main);
hyperpoint bnlfar = xpush(v0) * spin(M_PI) * rspintox(rot) * rspintox(rot) * rspintox(rot) * xpush0(hdist0(rot));
hyperpoint bnrfar = xpush(v0) * spin(M_PI) * spintox(rot) * spintox(rot) * spintox(rot) * xpush0(hdist0(rot));
m.lcorner = xspinpush0 (bspi-M_PI/sym, main);
m.rcorner = xspinpush0 (bspi+M_PI/sym, main);
m.mfar[0] = xspinpush0 (bspi, v0);
m.mfar[1] = xspinpush0 (bspi, v1);
m.vfar[0] = spin(bspi) * bnlfar;
m.vfar[2] = spin(bspi) * bnrfar;
m.vfar[1] = spin(-2*M_PI/sym) * m.vfar[2];
m.vfar[3] = spin(+2*M_PI/sym) * m.vfar[0];
return m;
}
struct matrixlist {
mesher o, n;
vector<matrixitem> v;
};
matrixitem genitem(const transmatrix& m1, const transmatrix& m2, int nsym) {
matrixitem mi;
mi.first = m1;
mi.second.resize(nsym);
for(int i=0; i<nsym; i++)
mi.second[i] = spin(2*M_PI*i/nsym) * m2;
return mi;
}
bool do_kleinize() { return S3 >= OINF; }
EX hyperpoint kleinize(hyperpoint h) {
if(GDIM == 2) return point3(h[0]/h[2], h[1]/h[2], 1);
else return point31(h[0]/h[3], h[1]/h[3], h[2]/h[3]);
}
EX hyperpoint may_kleinize(hyperpoint h) {
if(do_kleinize()) return kleinize(h);
else return h;
}
void addmatrix(matrixlist& matrices, hyperpoint o0, hyperpoint o1, hyperpoint o2, hyperpoint n0, hyperpoint n1, hyperpoint n2, int d, int osym, int nsym) {
if(do_kleinize()) o0 = kleinize(o0), o1 = kleinize(o1), o2 = kleinize(o2), n0 = kleinize(n0), n1 = kleinize(n1), n2 = kleinize(n2);
matrices.v.push_back(genitem(inverse(spin(2*M_PI*d/osym)*build_matrix(o0, o1, o2,C02)), spin(2*M_PI*d/nsym)*build_matrix(n0, n1, n2,C02), nsym));
}
matrixlist hex_matrices, hept_matrices;
void generate_matrices(matrixlist& matrices, const mesher& o, const mesher& n) {
matrices.v.clear();
matrices.o = o;
matrices.n = n;
for(int d=0; d<o.sym; d++) {
hyperpoint center = hpxy(0,0);
int d1 = d&1;
addmatrix(matrices, center, o.lcorner, o.rcorner, center, n.lcorner, n.rcorner, d, o.sym, n.sym);
addmatrix(matrices, o.mfar[d1], o.lcorner, o.rcorner, n.mfar[d1], n.lcorner, n.rcorner, d, o.sym, n.sym);
addmatrix(matrices, o.mfar[d1], o.lcorner, o.vfar[d1], n.mfar[d1], n.lcorner, n.vfar[d1], d, o.sym, n.sym);
addmatrix(matrices, o.mfar[d1], o.rcorner, o.vfar[d1+2], n.mfar[d1], n.rcorner, n.vfar[d1+2], d, o.sym, n.sym);
}
}
int nsym0;
void generate_matrices_scale(ld scale, int noft) {
mesher ohex = msh(gNormal, 6, 0.329036, 0.566256, 0.620672, 0, 1);
mesher ohept = msh(gNormal, 7, hexf7, hcrossf7, hcrossf7, M_PI/7, 1);
if(!BITRUNCATED) {
mesher nall = msh(geometry, S7, cgi.rhexf, cgi.tessf, cgi.tessf, -M_PI, scale);
bool use = geosupport_football() < 2;
if(use && noft == 1) {
mesher opure = msh(gNormal, 7, 0.620672, 1.090550, 1.090550, M_PI/7, 1);
generate_matrices(hept_matrices, opure, nall);
}
else if(use && noft == 2) {
mesher oeuc = msh(gNormal, 6, sqrt(3)/6, .5, .5, 0, 1);
generate_matrices(hept_matrices, oeuc, nall);
}
else if(use && noft == 3) {
generate_matrices(hept_matrices, ohex, nall);
}
else {
generate_matrices(hex_matrices, ohex, nall);
generate_matrices(hept_matrices, ohept, nall);
}
}
else {
generate_matrices(hex_matrices, ohex, msh(geometry, S6, cgi.hexvdist, cgi.hexhexdist, cgi.hcrossf, (S3-3)*M_PI/S3, scale));
generate_matrices(hept_matrices, ohept, msh(geometry, S7, cgi.rhexf, cgi.hcrossf, cgi.hcrossf, M_PI/S7, scale));
}
}
void geometry_information::bshape2(hpcshape& sh, PPR prio, int shapeid, matrixlist& m) {
auto& matrices = m.v;
int osym = m.o.sym;
int nsym = m.n.sym;
int whereis = 0;
while(polydata[whereis] != NEWSHAPE || polydata[whereis+1] != shapeid) whereis++;
int rots = polydata[whereis+2]; int sym = polydata[whereis+3];
whereis += 4;
int qty = 0;
while(polydata[whereis + 2*qty] != NEWSHAPE) qty++;
vector<hyperpoint> lst;
for(int i=0; i<qty; i++) {
dynamicval<eGeometry> dg(geometry, gNormal);
lst.push_back(hpxy(polydata[whereis+2*i], polydata[whereis+2*i+1]));
}
if(sym == 2)
for(int i=qty-1; i>=0; i--) {
dynamicval<eGeometry> dg(geometry, gNormal);
lst.push_back(hpxy(polydata[whereis+2*i], -polydata[whereis+2*i+1]));
}
hyperpoint lstmid = hpxyz(0,0,0);
for(auto pp: lst) lstmid += pp;
transmatrix T = spin(-m.o.bspi);
while((spin(2*M_PI / rots) * T* lstmid)[0] < (T*lstmid)[0])
T = spin(2*M_PI / rots) * T;
while((spin(-2*M_PI / rots) * T* lstmid)[0] < (T*lstmid)[0])
T = spin(-2*M_PI / rots) * T;
T = spin(m.o.bspi) * T;
for(auto &pp: lst) pp = T * pp;
if(osym % rots && rots % osym) printf("warning: rotation oddity (shapeid %d, osym=%d rots=%d)\n", shapeid, osym, rots);
if(rots > osym && rots % osym == 0) {
int rep = rots / osym;
int s = lst.size();
for(int i=0; i<s*(rep-1); i++)
lst.push_back(spin(2*M_PI/rots) * lst[i]);
rots /= rep;
}
bshape(sh, prio);
for(int r=0; r<nsym; r+=osym/rots) {
for(hyperpoint h: lst) {
hyperpoint nh = may_kleinize(h);
int mapped = 0;
for(auto& m: matrices) {
hyperpoint z = m.first * h;
if(z[0] > -1e-5 && z[1] > -1e-5 && z[2] > -1e-5) {
nh = m.second[r] * z, mapped++;
}
}
if(mapped == 0) printf("warning: not mapped (shapeid %d)\n", shapeid);
hpcpush(mid(nh, nh));
}
}
hpcpush(hpc[last->s]);
}
template<class T> void sizeto(T& t, int n) {
if(isize(t) <= n) t.resize(n+1);
}
void geometry_information::bshape_regular(floorshape &fsh, int id, int sides, ld shift, ld size, cell *c) {
sizeto(fsh.b, id);
sizeto(fsh.shadow, id);
#if CAP_BT
if(bt::in()) {
const int STEP = vid.texture_step;
for(int t=0; t<2; t++) {
if(t == 0)
bshape(fsh.b[id], fsh.prio);
if(t == 1)
bshape(fsh.shadow[id], fsh.prio);
for(int i=0; i<sides; i++) {
hyperpoint h0 = bt::get_corner_horo_coordinates(c, i) * size;
hyperpoint h1 = bt::get_corner_horo_coordinates(c, i+1) * size;
if(t) h0 *= SHADMUL, h1 *= SHADMUL;
hyperpoint hd = (h1 - h0) / STEP;
for(int j=0; j<STEP; j++) {
hpcpush(bt::get_horopoint(h0 + hd * j));
if(geometry == gBinary4 && among(i, 2, 4)) break;
if(geometry == gBinaryTiling && among(i, 0, 4)) break;
if(geometry == gTernary && among(i, 3, 5)) break;
}
}
hpcpush(hpc[last->s]);
}
for(int k=0; k<SIDEPARS; k++) {
if(isize(fsh.gpside[k]) < c->type)
fsh.gpside[k].resize(c->type);
for(int i=0; i<c->type; i++) {
sizeto(fsh.gpside[k][i], id);
bshape(fsh.gpside[k][i][id], PPR::LAKEWALL);
hyperpoint h0 = bt::get_corner_horo_coordinates(c, i) * size;
hyperpoint h1 = bt::get_corner_horo_coordinates(c, i+1) * size;
hyperpoint hd = (h1 - h0) / STEP;
for(int j=0; j<=STEP; j++)
hpcpush(iddspin(c, i) * bt::get_horopoint(h0 + hd * j));
chasmifyPoly(dlow_table[k], dhi_table[k], k);
}
}
return;
}
#endif
bshape(fsh.b[id], fsh.prio);
for(int t=0; t<=sides; t++)
hpcpush(xspinpush0(t*2 * M_PI / sides + shift * M_PI / S42, size));
bshape(fsh.shadow[id], fsh.prio);
for(int t=0; t<=sides; t++)
hpcpush(xspinpush0(t*2 * M_PI / sides + shift * M_PI / S42, size * SHADMUL));
for(int k=0; k<SIDEPARS; k++) {
fsh.side[k].resize(2);
bshape(fsh.side[k][id], PPR::LAKEWALL);
hpcpush(xspinpush0(+M_PI/sides, size));
hpcpush(xspinpush0(-M_PI/sides, size));
chasmifyPoly(dlow_table[k], dhi_table[k], k);
}
}
#if CAP_IRR
namespace irr { void generate_floorshapes(); }
#endif
// !siid equals pseudohept(c)
void geometry_information::generate_floorshapes_for(int id, cell *c, int siid, int sidir) {
DEBBI(DF_POLY, ("generate_floorshapes_for ", id));
for(auto pfsh: all_plain_floorshapes) {
auto& fsh = *pfsh;
if(STDVAR && (standard_tiling() || bt::in())) {
ld hexside = fsh.rad0, heptside = fsh.rad1;
for(int k=0; k<SIDEPARS; k++) sizeto(fsh.side[k], id);
ld td = (PURE && !(S7&1)) ? S42+S6 : 0;
if(&fsh == &shBigHepta) td += S6;
if(S3 >= OINF && !(S7 & 1)) td = S42 * 1. / S7;
int b = 0;
if(S3 == 4 && BITRUNCATED) b += S14;
if(id)
bshape_regular(fsh, id, S7, td, heptside, c);
else if(PURE) {
if(&fsh == &shTriheptaFloor)
bshape_regular(fsh, 0, S7/2, 0, hexside, c);
else if(&fsh == &shBigTriangle)
bshape_regular(fsh, 0, S7/2, S12, hexside, c);
else
bshape_regular(fsh, 0, S7, td, heptside, c);
}
else if(&fsh == &shBigTriangle)
bshape_regular(fsh, 0, S3, b+S14, hexside, c);
else if(&fsh == &shTriheptaFloor)
bshape_regular(fsh, 0, S3, b, hexside, c);
else
bshape_regular(fsh, 0, S6, S7, hexside, c);
continue;
}
// special
ld sca = 3 * shFullFloor.rad0 / fsh.rad0;
vector<hyperpoint> cornerlist;
int cor = c->type;
if(&fsh == &shTriheptaFloor) {
if(!siid) {
for(int i=0; i<cor; i++)
cornerlist.push_back(midcorner(c, i, .49));
}
else {
for(int i=0; i<cor; i++) {
int ri = i;
if((i&1) == ((sidir+siid)&1)) ri--;
ri = c->c.fix(ri);
cornerlist.push_back(mid(get_corner_position(c, ri, 3.1), get_corner_position(c, c->c.fix(ri+1), 3.1)));
}
}
}
else if(&fsh == &shBigTriangle) {
if(!siid) {
for(int i=0; i<cor; i++) cornerlist.push_back(hpxy(0,0));
}
else if(geosupport_chessboard()) {
for(int i=0; i<cor; i++) {
hyperpoint nc = nearcorner(c, i);
cornerlist.push_back(mid_at(hpxy(0,0), nc, .94));
}
}
else {
for(int i=0; i<cor; i++) {
int ri = i;
if((i&1) != ((sidir+siid)&1)) ri--;
ri = c->c.fix(ri);
hyperpoint nc = nearcorner(c, ri);
cornerlist.push_back(mid_at(hpxy(0,0), nc, .94));
}
}
}
else if(&fsh == &shBigHepta) {
if(!siid) {
for(int i=0; i<cor; i++) {
hyperpoint nc = nearcorner(c, i);
cornerlist.push_back(mid_at(hpxy(0,0), nc, .94));
}
}
else {
for(int i=0; i<cor; i++) cornerlist.push_back(hpxy(0,0));
}
}
else if(arb::in()) {
vector<hyperpoint> actual;
for(int j=0; j<cor; j++)
actual.push_back(get_corner_position(c, j));
ld min_dist = 1e3;
for(int j=0; j<cor; j++)
for(int k=0; k<j; k++) {
ld dist = hdist(actual[j], actual[k]);
if(dist > 1e-6 && dist < min_dist)
min_dist = dist;
}
ld dist = min_dist * (1 - 3 / sca);
ld area = 0;
for(int j=0; j<cor; j++) {
hyperpoint current = kleinize(actual[j]);
hyperpoint last = kleinize(actual[j?j-1:cor-1]);
area += current[0] * last[1] - last[0] * current[1];
}
if(area < 0) dist = -dist;
for(int j=0; j<cor; j++) {
hyperpoint last = actual[j?j-1:cor-1];
hyperpoint current = ypush(1e-6 * randd()) * xpush(1e-6) * actual[j];
hyperpoint next = actual[j<cor-1?j+1:0];
auto T = gpushxto0(current);
last = T * last;
next = T * next;
hyperpoint a = rspintox(last) * ypush0(dist);
hyperpoint b = rspintox(last) * xpush(hdist0(last)) * ypush0(dist);
hyperpoint c = rspintox(next) * ypush0(-dist);
hyperpoint d = rspintox(next) * xpush(hdist0(next)) * ypush0(-dist);
hyperpoint h = linecross(a, b, c, d);
cornerlist.push_back(rgpushxto0(current) * h);
}
}
else {
for(int j=0; j<cor; j++)
cornerlist.push_back(get_corner_position(c, j, sca));
}
sizeto(fsh.b, id);
bshape(fsh.b[id], fsh.prio);
if(cor == 2) {
/* give digons some width */
for(int i=0; i<cor; i++) hpcpush(spin(-.1) * cornerlist[i]), hpcpush(spin(+.1) * cornerlist[i]);
hpcpush(spin(-.1) * cornerlist[0]);
}
else if(&fsh == &shTriheptaFloor && cor == 4 && siid)
/* trihepta floors generate digons too */
for(int i=0; i<=cor; i++) hpcpush(spin((i&1) ? .1 : -.1) * cornerlist[i%cor]);
else
for(int i=0; i<=cor; i++) hpcpush(cornerlist[i%cor]);
sizeto(fsh.shadow, id);
bshape(fsh.shadow[id], fsh.prio);
for(int i=0; i<=cor; i++)
hpcpush(mid_at(hpxy(0,0), cornerlist[i%cor], SHADMUL));
for(int k=0; k<SIDEPARS; k++) {
if(isize(fsh.gpside[k]) < cor)
fsh.gpside[k].resize(cor);
for(int cid=0; cid<cor; cid++) {
sizeto(fsh.gpside[k][cid], id);
bshape(fsh.gpside[k][cid][id], fsh.prio);
hpcpush(iddspin_side(c, cid) * cornerlist[cid]);
hpcpush(iddspin_side(c, cid) * cornerlist[(cid+1)%cor]);
chasmifyPoly(dlow_table[k], dhi_table[k], k);
}
}
}
for(auto pfsh: all_escher_floorshapes) {
auto& fsh = *pfsh;
sizeto(fsh.b, id);
sizeto(fsh.shadow, id);
if(STDVAR && standard_tiling()) {
generate_matrices_scale(fsh.scale, fsh.noftype);
if(PURE && geosupport_football() < 2) {
bshape2(fsh.b[id], fsh.prio, fsh.shapeid2 ? fsh.shapeid2 : fsh.shapeid1, hept_matrices);
}
else {
if(id == 0) bshape2(fsh.b[0], fsh.prio, fsh.shapeid0, hex_matrices);
if(id == 1) bshape2(fsh.b[1], fsh.prio, fsh.shapeid1, hept_matrices);
}
generate_matrices_scale(fsh.scale * SHADMUL, fsh.noftype);
if(PURE && geosupport_football() < 2) {
bshape2(fsh.shadow[id], fsh.prio, fsh.shapeid2 ? fsh.shapeid2 : fsh.shapeid1, hept_matrices);
}
else {
if(id == 0) bshape2(fsh.shadow[0], fsh.prio, fsh.shapeid0, hex_matrices);
if(id == 1) bshape2(fsh.shadow[1], fsh.prio, fsh.shapeid1, hept_matrices);
}
}
else {
generate_matrices_scale(fsh.scale, fsh.noftype);
auto& m = (siid && geosupport_football() == 2) ? hex_matrices : hept_matrices;
int cor = c->type;
m.n.sym = cor;
int v = sidir+siid;
for(auto& mvi: m.v) mvi.second.resize(cor);
for(int ii=0; ii<2; ii++) {
int i = 0;
for(int d=0; d<m.o.sym; d++) {
hyperpoint center = hpxy(0,0);
for(int cid=0; cid<cor; cid++) {
hyperpoint nlcorner = get_corner_position(c, (d+cid+v+1) % cor, 3 / fsh.scale * (ii ? 1/SHADMUL : 1));
hyperpoint nrcorner = get_corner_position(c, (d+cid+v+2) % cor, 3 / fsh.scale * (ii ? 1/SHADMUL : 1));
hyperpoint nfar = nearcorner(c, (d+cid+v+1) % cor);
hyperpoint nlfar = farcorner(c, (d+cid+v+1) % cor, 0);
hyperpoint nrfar = farcorner(c, (d+cid+v+1) % cor, 1);
m.v[i].second[cid] = build_matrix(center, nlcorner, nrcorner,C02);
m.v[i+1].second[cid] = build_matrix(nfar, nlcorner, nrcorner,C02);
m.v[i+2].second[cid] = build_matrix(nfar, nlcorner, nlfar,C02);
m.v[i+3].second[cid] = build_matrix(nfar, nrcorner, nrfar,C02);
}
i += 4;
}
if(i != isize(m.v)) printf("warning: i=%d sm=%d\n", i, isize(m.v));
bshape2((ii?fsh.shadow:fsh.b)[id], fsh.prio, (fsh.shapeid2 && geosupport_football() < 2) ? fsh.shapeid2 : siid?fsh.shapeid0:fsh.shapeid1, m);
}
}
}
#if MAXMDIM >= 4
if(WDIM == 2 && GDIM == 3) {
finishshape();
for(auto pfsh: all_plain_floorshapes) {
auto& fsh = *pfsh;
for(int i=fsh.shadow[id].s; i<fsh.shadow[id].e; i++)
hpc[i] = orthogonal_move(hpc[i], FLOOR - human_height / 100);
for(int k=0; k<SIDEPARS; k++) {
sizeto(fsh.levels[k], id);
bshape(fsh.levels[k][id], fsh.prio);
last->flags |= POLY_TRIANGLES;
last->tinf = &floor_texture_vertices[fsh.id];
last->texture_offset = 0;
#if CAP_BT
if(bt::in())
for(int t=0; t<c->type; t++)
texture_order([&] (ld x, ld y) {
hyperpoint left = bt::get_corner_horo_coordinates(c, t);
hyperpoint right = bt::get_corner_horo_coordinates(c, t+1);
hpcpush(orthogonal_move(bt::get_horopoint(left * x + right * y), dfloor_table[k]));
});
else
#endif
if(1) {
int s = fsh.b[id].s;
int e = fsh.b[id].e-1;
for(int t=0; t<e-s; t++) {
hyperpoint v1 = may_kleinize(hpc[s+t]) - C0;
hyperpoint v2 = may_kleinize(hpc[s+t+1]) - C0;
texture_order([&] (ld x, ld y) {
hpcpush(
orthogonal_move(
normalize(C0 + v1 * x + v2 * y)
, dfloor_table[k])
);
});
}
}
finishshape();
ensure_vertex_number(fsh.levels[k][id]);
}
for(int co=0; co<2; co++) {
sizeto(fsh.cone[co], id);
bshape(fsh.cone[co][id], fsh.prio);
last->flags |= POLY_TRIANGLES;
last->tinf = &floor_texture_vertices[fsh.id];
last->texture_offset = 0;
ld h = (FLOOR - WALL) / (co+1);
ld top = co ? (FLOOR + WALL) / 2 : WALL;
#if CAP_BT
if(bt::in())
for(int t=0; t<c->type; t++)
texture_order([&] (ld x, ld y) {
hyperpoint left = bt::get_corner_horo_coordinates(c, t);
hyperpoint right = bt::get_corner_horo_coordinates(c, t+1);
hpcpush(orthogonal_move(bt::get_horopoint(left * x + right * y), top + h * (x+y)));
});
else
#endif
if(1) {
int s = fsh.b[id].s;
int e = fsh.b[id].e-1;
for(int t=0; t<e-s; t++) {
hyperpoint v1 = may_kleinize(hpc[s+t]) - C0;
hyperpoint v2 = may_kleinize(hpc[s+t+1]) - C0;
texture_order([&] (ld x, ld y) { hpcpush(orthogonal_move(normalize(C0 + v1 * x + v2 * y), top + h * (x+y))); });
}
}
finishshape();
ensure_vertex_number(fsh.cone[co][id]);
}
for(int l=0; l<SIDEPARS; l++) {
for(auto& li: fsh.side[l])
bind_floor_texture(li, fsh.id);
if(isize(fsh.gpside[l]) < c->type)
fsh.gpside[l].resize(c->type);
for(auto& gs: fsh.gpside[l]) {
for(auto& li: gs)
bind_floor_texture(li, fsh.id);
}
}
}
for(auto pfsh: all_escher_floorshapes) {
auto& fsh = *pfsh;
for(int l=0; l<SIDEPARS; l++) {
fsh.levels[l] = shFullFloor.levels[l];
fsh.shadow = shFullFloor.shadow;
for(auto& li: fsh.levels[l]) bind_floor_texture(li, fsh.id);
fsh.side[l] = shFullFloor.side[l];
for(auto& li: fsh.side[l]) bind_floor_texture(li, fsh.id);
if(isize(fsh.gpside[l]) < c->type)
fsh.gpside[l].resize(c->type);
for(int e=0; e<c->type; e++) {
fsh.gpside[l][e] = shFullFloor.gpside[l][e];
for(auto& li: fsh.gpside[l][e])
bind_floor_texture(li, fsh.id);
}
fsh.cone[0] = shFullFloor.cone[0];
fsh.cone[1] = shFullFloor.cone[1];
for(int c=0; c<2; c++)
for(auto& li: fsh.cone[c])
bind_floor_texture(li, fsh.id);
}
}
finishshape();
}
#endif
}
void geometry_information::generate_floorshapes() {
DEBBI(DF_POLY, ("generate_floorshapes"));
heptagon modelh;
cell model;
model.master = &modelh;
modelh.c7 = &model;
model.type = modelh.type = S7;
auto mmerge1 = [&] (int i, int j) { model.c.setspin(i, j, false); modelh.c.setspin(i, j, false); };
auto mmerge = [&] (int i, int j) { mmerge1(i, j); mmerge1(j, i); };
for(int i=0; i<S7; i++) {
model.move(i) = &model;
modelh.move(i) = &modelh;
model.c.setspin(i, i, false);
modelh.c.setspin(i, i, false);
}
if(WDIM == 3) ;
#if CAP_IRR
else if(IRREGULAR) {
DEBBI(DF_POLY, ("generate_floorshapes: irregular"));
int cc = isize(irr::cells);
for(int id=0; id<cc; id++) {
irr::cellindex[&model] = id;
auto& vs = irr::cells[id];
model.type = isize(vs.vertices);
int siid = !vs.is_pseudohept;
int sidir = 0;
if(siid) sidir = irr::cells[vs.neid[0]].is_pseudohept;
generate_floorshapes_for(id, &model, !vs.is_pseudohept, sidir);
}
printf("done\n");
}
#endif
else if(GOLDBERG_INV) { /* will be generated on the fly */ }
else if(inforder::mixed()) { /* will be generated on the fly */ }
#if CAP_BT
else if(kite::in()) {
dynamicval<bool> ncor(approx_nearcorner, true);
for(int i=0; i<2; i++) {
modelh.s = hstate(i); /* kite/dart shape */
kite::no_adj = true;
generate_floorshapes_for(i, &model, 0, 0);
kite::no_adj = false;
}
}
#endif
#if CAP_ARCM
else if(arcm::in()) {
arcm::parent_index_of(&modelh) = 0;
auto &ac = arcm::current;
for(int i=0; i<2*ac.N + 2; i++) {
if(ac.regular && i>=2 && i < 2*ac.N) continue;
arcm::id_of(&modelh) = i;
model.type = isize(ac.triangles[i]);
if(DUAL) model.type /= 2, arcm::parent_index_of(&modelh) = !(i&1);
if(BITRUNCATED)
generate_floorshapes_for(i, &model, !arcm::pseudohept(&model), arcm::pseudohept(&model) ? 0 : 1^(i&1));
else if(geosupport_football() == 2)
generate_floorshapes_for(i, &model, !arcm::pseudohept(&model), i >= 4 ? 1 : 0);
else
generate_floorshapes_for(i, &model, 0, 0);
}
}
#endif
else if(arb::in()) {
auto& c = arb::current;
int n = isize(c.shapes);
vector<cell> models(n);
vector<heptagon> modelh(n);
for(int i=0; i<n; i++) {
auto &ms = models[i];
auto &mh = modelh[i];
for(auto& t: ms.c.move_table) t = nullptr;
for(auto& t: mh.c.move_table) t = nullptr;
}
for(int i=0; i<n; i++) {
auto &ms = models[i];
auto &mh = modelh[i];
ms.master = &mh;
mh.c7 = &ms;
mh.zebraval = i;
auto& sh = c.shapes[i];
ms.type = mh.type = sh.size();
}
for(int i=0; i<n; i++) {
auto &ms = models[i];
auto &mh = modelh[i];
auto& sh = c.shapes[i];
for(int j=0; j<sh.size(); j++) {
auto& co = sh.connections[j];
mh.c.connect(j, &modelh[get<0>(co)], get<1>(co), get<2>(co));
ms.c.connect(j, &models[get<0>(co)], get<1>(co), get<2>(co));
}
}
for(int i=0; i<n; i++) generate_floorshapes_for(i, &models[i], 0, 0);
}
else if(geometry == gBinary4) {
for(int i: {0,1}) {
modelh.zebraval = i;
mmerge(2, 4); mmerge(0, 3); mmerge(1, 3); mmerge(i, 3);
generate_floorshapes_for(i, &model, 1, 0);
}
}
else if(geometry == gTernary) {
for(int i: {0,1,2}) {
modelh.zebraval = i;
mmerge(3, 5); for(int a=0; a<3; a++) mmerge1(a, 4); mmerge(4, i);
generate_floorshapes_for(i, &model, 1, 0);
}
}
else if(PURE && geometry != gBinaryTiling && geosupport_football() < 2) {
generate_floorshapes_for(0, &model, 1, 0);
}
else {
static hrmap_standard stdmap;
dynamicval<hrmap*> c(currentmap, &stdmap);
// cell model;
model.type = S6; generate_floorshapes_for(0, &model, 0, 0);
model.type = S7; generate_floorshapes_for(1, &model, bt::in() ? 0 : 1, 0);
}
}
#if CAP_GP
EX namespace gp {
EX void clear_plainshapes() {
for(int m=0; m<3; m++)
for(int sd=0; sd<8; sd++)
for(int i=0; i<32; i++)
for(int j=0; j<32; j++)
for(int k=0; k<8; k++)
cgi.gpdata->pshid[m][sd][i][j][k] = -1;
cgi.gpdata->nextid = 0;
}
void build_plainshape(int& id, gp::local_info& li, cell *c0, int siid, int sidir) {
id = cgi.gpdata->nextid++;
bool master = !(li.relative.first||li.relative.second);
int cor = master ? S7 : SG6;
if(master) li.last_dir = -1;
DEBB(DF_GP, (format("last=%d at=%d,%d tot=%d siid=%d sidir=%d cor=%d id=%d\n", li.last_dir, li.relative.first, li.relative.second, li.total_dir, siid, sidir, cor, id)));
cgi.generate_floorshapes_for(id, c0, siid, sidir);
cgi.finishshape();
cgi.extra_vertices();
}
EX int get_plainshape_id(cell *c) {
int siid, sidir;
cell *c1 = c;
auto f = [&] {
if(geosupport_threecolor() == 2) {
auto si = patterns::getpatterninfo(c1, patterns::PAT_COLORING, patterns::SPF_NO_SUBCODES);
siid = si.id>>2;
// if(siid == 2) si.dir++;
// if(siid != pattern_threecolor(c)) printf("threecolor mismatch\n");
// if(pattern_threecolor(createMov(c, c->fixd(si.dir))) != (siid+1)%3) printf("threecolor mismatch direction\n");
sidir = c1->c.fix(si.dir);
}
else if(geosupport_football() == 2) {
siid = !pseudohept(c1);
sidir = !ishex1(c1);
}
else if(geosupport_chessboard()) {
siid = !chessvalue(c1);
sidir = 0;
}
else {
siid = 0;
sidir = 0;
}
};
if(INVERSE && gp::variation_for(gp::param) == eVariation::goldberg) {
c1 = gp::get_mapped(c);
UIU(f());
}
else if(INVERSE) {
siid = 0;
sidir = 0;
}
else f();
auto& id = cgi.gpdata->pshid[siid][sidir][draw_li.relative.first&31][draw_li.relative.second&31][gmod(draw_li.total_dir, S6)];
if(id == -1 && sphere && isize(cgi.shFloor.b) > 0) {
forCellEx(c1, c) if(!gmatrix0.count(c1)) return 0;
}
if(id == -1) build_plainshape(id, draw_li, c, siid, sidir);
return id;
}
EX }
#endif
qfloorinfo qfi;
EX void set_no_floor() {
qfi.fshape = NULL;
qfi.shape = NULL;
qfi.tinf = NULL;
qfi.usershape = -1;
}
EX void set_floor(floorshape& sh) {
qfi.fshape = &sh;
qfi.shape = NULL;
qfi.tinf = NULL;
qfi.usershape = -1;
}
EX void set_floor(hpcshape& sh) {
qfi.shape = &sh;
qfi.fshape = NULL;
qfi.spin = Id;
qfi.tinf = NULL;
qfi.usershape = -1;
}
EX void set_floor(const transmatrix& spin, hpcshape& sh) {
qfi.shape = &sh;
qfi.fshape = NULL;
qfi.spin = spin;
qfi.usershape = -1;
}
EX int shvid(cell *c) {
return currentmap->shvid(c);
}
int hrmap_standard::shvid(cell *c) {
if(GOLDBERG)
return gp::get_plainshape_id(c);
#if CAP_IRR
else if(IRREGULAR)
return irr::cellindex[c];
#endif
else if(geosupport_football() == 2)
return pseudohept(c);
else if(inforder::mixed()) {
int t = c->type;
static vector<bool> computed;
if(isize(computed) <= t) computed.resize(t+1);
if(!computed[t]) {
computed[t] = true;
cell model;
heptagon modelh;
model.type = t;
modelh.type = t;
S7 = t;
for(int i=0; i<S7; i++) {
model.move(i) = &model;
modelh.move(i) = &modelh;
model.c.setspin(i, i, false);
modelh.c.setspin(i, i, false);
}
cgi.tessf = edge_of_triangle_with_angles(0, M_PI/t, M_PI/t);
cgi.crossf = cgi.tessf;
println(hlog, "generating floorshapes for ", t);
cgi.generate_floorshapes_for(t, &model, 0, 0);
cgi.finishshape();
cgi.extra_vertices();
}
return t;
}
else if(PURE)
return 0;
else
return ctof(c);
}
EX struct dqi_poly *draw_shapevec(cell *c, const shiftmatrix& V, const vector<hpcshape> &shv, color_t col, PPR prio IS(PPR::DEFAULT)) {
if(no_wall_rendering) return NULL;
if(!c) return &queuepolyat(V, shv[0], col, prio);
else if(WDIM == 3) return NULL;
else if(currentmap->strict_tree_rules()) return &queuepolyat(V, shv[shvid(c)], col, prio);
#if CAP_GP
else if(GOLDBERG) {
int id = gp::get_plainshape_id(c);
if(isize(shv) > id) return &queuepolyat(V, shv[id], col, prio);
return NULL;
}
#endif
#if CAP_IRR
else if(IRREGULAR) {
int id = irr::cellindex[c];
if(id < 0 || id >= isize(shv)) {
return NULL;
}
return &queuepolyat(V, shv[id], col, prio);
}
#endif
#if CAP_ARCM
else if(arcm::in()) {
return &queuepolyat(V, shv[shvid(c)], col, prio);
}
#endif
else if(GOLDBERG && ishex1(c))
return &queuepolyat(V * pispin, shv[0], col, prio);
else if(!(S7&1) && PURE && !kite::in() && !a4) {
auto si = patterns::getpatterninfo(c, patterns::PAT_COLORING, 0);
if(si.id == 8) si.dir++;
transmatrix D = applyPatterndir(c, si);
return &queuepolyat(V*D, shv[shvid(c)], col, prio);
}
else
return &queuepolyat(V, shv[shvid(c)], col, prio);
}
EX void draw_floorshape(cell *c, const shiftmatrix& V, const floorshape &fsh, color_t col, PPR prio IS(PPR::DEFAULT)) {
if(no_wall_rendering) return;
draw_shapevec(c, V, fsh.b, col, prio);
}
EX void draw_qfi(cell *c, const shiftmatrix& V, color_t col, PPR prio IS(PPR::DEFAULT), vector<hpcshape> floorshape::* tab IS(&floorshape::b)) {
if(no_wall_rendering) return;
if(qfi.shape)
queuepolyat(V * qfi.spin, *qfi.shape, col, prio);
else if(qfi.usershape >= 0) {
mapeditor::drawUserShape(V * qfi.spin, mapeditor::sgFloor, qfi.usershape, col, c);
}
else if(!qfi.fshape) ;
#if CAP_TEXTURE
else if(qfi.tinf) {
auto& poly = queuetable(V * qfi.spin, qfi.tinf->vertices, isize(qfi.tinf->vertices), texture::config.mesh_color, texture::config.recolor(col), prio == PPR::DEFAULT ? PPR::FLOOR : prio);
poly.tinf = qfi.tinf;
poly.offset_texture = 0;
poly.flags = POLY_INVERSE;
}
#endif
else draw_shapevec(c, V, (qfi.fshape->*tab), col, prio);
}
EX bool floorshape_debug;
EX void viewmat() {
if(floorshape_debug) {
shiftmatrix V = ggmatrix(cwt.at);
for(int i=0; i<cwt.at->type; i++) {
shiftpoint ci = V * get_corner_position(cwt.at, i);
shiftpoint ci1 = V * get_corner_position(cwt.at, (i+1) % cwt.at->type);
shiftpoint cn = V * nearcorner(cwt.at, i);
shiftpoint cf0 = V * farcorner(cwt.at, i, 0);
shiftpoint cf1 = V * farcorner(cwt.at, i, 1);
queuestr(ci, 20, its(i), 0x0000FF, 1);
if(vid.grid)
queuestr(cn, 20, its(i), 0x00FF00, 1);
else
queuestr(V * currentmap->adj(cwt.at, i) * C0, 20, its(i), 0x00FFFF, 1);
queueline(V * C0, ci, 0xFFFFFFFF, 3);
queueline(ci, ci1, 0xFFFF00FF, 3);
queueline(ci, cn, 0xFF00FFFF, 3);
queueline(ci1, cn, 0xFF0000FF, 3);
queueline(ci, cf0, 0x00FFFFFF, 3);
queueline(cn, cf0, 0x00FF00FF, 3);
queueline(cn, cf1, 0x0000FFFF, 3);
}
}
}
auto floor_hook = arg::add1("-floordebug", [] { floorshape_debug = true; });
#endif
#if MAXMDIM < 4 || !CAP_GL
EX void ensure_vertex_number(basic_textureinfo& bti, int qty) {}
EX void ensure_vertex_number(hpcshape& sh) {}
EX void bind_floor_texture(hpcshape& li, int id) {}
#endif
#if MAXMDIM >= 4 && CAP_GL
EX ld floor_texture_square_size;
void draw_shape_for_texture(floorshape* sh) {
int id = sh->id;
const ld s1 = 1;
const ld s3 = 3 * s1;
const ld sd = s1/2;
ld gx = (id % 8) * s3 - 3.5 * s3;
ld gy = (id / 8) * s3 - 3.5 * s3;
if(1) {
dynamicval<ld> v(vid.linewidth, 8);
curvepoint(eupush(gx+s1, gy-s1) * C0);
curvepoint(eupush(gx+s1, gy+s1) * C0);
curvepoint(eupush(gx-s1, gy+s1) * C0);
curvepoint(eupush(gx-s1, gy-s1) * C0);
curvepoint(eupush(gx+s1, gy-s1) * C0);
queuecurve(shiftless(Id), 0x000000FF, 0xFFFFFFFF - 0x1010100 * (sh->pstrength * 24/10), PPR::LAKELEV);
}
poly_outline = 0xFFFFFFFF - 0x1010100 * (sh->pstrength * 3/2);
for(int a=-1; a<=1; a++)
for(int b=-1; b<=1; b++)
queuepoly(shiftless(eupush(gx+a, gy+b)), sh->b[0], 0xFFFFFFFF);
if(sh == &cgi.shCrossFloor) {
queuepoly(shiftless(eupush(gx, gy) * spin(M_PI/4)), cgi.shCross, 0x808080FF);
}
if(1) {
dynamicval<ld> v(vid.linewidth, 8);
curvepoint(eupush(gx+sd, gy-sd) * C0);
curvepoint(eupush(gx+sd, gy+sd) * C0);
curvepoint(eupush(gx-sd, gy+sd) * C0);
curvepoint(eupush(gx-sd, gy-sd) * C0);
curvepoint(eupush(gx+sd, gy-sd) * C0);
queuecurve(shiftless(Id), 0x40404000 + sh->fstrength * 192/10, 0, PPR::LINE);
}
for(int i=0; i<(ISMOBILE ? 10 : 1000); i++) {
hyperpoint h1 = hpxy(sd * (6*randd()-3), sd * (6*randd()-3));
hyperpoint h2 = hpxy(sd * (6*randd()-3), sd * (6*randd()-3));
ld d = hdist(h1, h2);
hyperpoint h3 = h1 + (h2-h1) /d * min(d, .1);
for(int a=0; a<4; a++) {
curvepoint(eupush(gx,gy) * eupush(spin(90*degree*a) * h1) * C0);
curvepoint(eupush(gx,gy) * eupush(spin(90*degree*a) * h3) * C0);
queuecurve(shiftless(Id), 0x10101010, 0, PPR::LINE);
}
}
auto& ftv = floor_texture_vertices[sh->id];
ftv.tvertices.clear();
ftv.texture_id = floor_textures->renderedTexture;
hyperpoint center = eupush(gx, gy) * C0;
hyperpoint v1 = hpxyz3(sd, sd, 0, 0);
hyperpoint v2 = hpxyz3(sd, -sd, 0, 0);
if(1) {
hyperpoint inmodel;
applymodel(shiftless(center), inmodel);
glvertex tmap;
tmap[0] = (1 + inmodel[0] * pconf.scale) / 2;
tmap[1] = (1 - inmodel[1] * pconf.scale) / 2;
applymodel(shiftless(center + v1), inmodel);
tmap[2] = (1 + inmodel[0] * pconf.scale) / 2 - tmap[0];
floor_texture_map[sh->id] = tmap;
}
auto tvec_at = [&] (ld x, ld y) {
hyperpoint h = center + v1 * x + v2 * y;
hyperpoint inmodel;
applymodel(shiftless(h), inmodel);
glvec2 v;
v[0] = (1 + inmodel[0] * pconf.scale) / 2;
v[1] = (1 - inmodel[1] * pconf.scale) / 2;
return v;
};
// SL2 needs 6 times more
texture_order([&] (ld x, ld y) {
auto v = tvec_at(x, y);
ftv.tvertices.push_back(glhr::makevertex(v[0], v[1], 0));
});
floor_texture_square_size = 2 * (tvec_at(1, 0)[0] - tvec_at(0, 0)[0]);
}
/** copy the texture vertices so that there are at least qty of them */
EX void ensure_vertex_number(basic_textureinfo& bti, int qty) {
int s = isize(bti.tvertices);
if(!s) return;
while(isize(bti.tvertices) <= qty) {
for(int i=0; i<s; i++) bti.tvertices.push_back(bti.tvertices[i]);
}
}
/** ensure_vertex_number for a hpcshape */
EX void ensure_vertex_number(hpcshape& sh) {
ensure_vertex_number(*sh.tinf, sh.e - sh.s);
}
EX void bind_floor_texture(hpcshape& li, int id) {
li.tinf = &floor_texture_vertices[id];
ensure_vertex_number(li);
}
#if HDR
const int FLOORTEXTURESIZE = 4096;
#endif
void geometry_information::make_floor_textures_here() {
require_shapes();
dynamicval<videopar> vi(vid, vid);
vid.xres = FLOORTEXTURESIZE;
vid.yres = FLOORTEXTURESIZE;
pconf.scale = 0.125;
pconf.camera_angle = 0;
pconf.alpha = 1;
dynamicval<ld> lw(vid.linewidth, 2);
floor_textures = new renderbuffer(vid.xres, vid.yres, vid.usingGL);
resetbuffer rb;
int q = isize(all_escher_floorshapes) + isize(all_plain_floorshapes);
floor_texture_vertices.resize(q);
floor_texture_map.resize(q);
auto cd = current_display;
cd->xtop = cd->ytop = 0;
cd->xsize = cd->ysize = FLOORTEXTURESIZE;
cd->xcenter = cd->ycenter = cd->scrsize = FLOORTEXTURESIZE/2;
cd->radius = cd->scrsize * pconf.scale;
floor_textures->enable();
#if CAP_VR
dynamicval<int> i(vrhr::state, 0);
#endif
floor_textures->clear(0); // 0xE8E8E8 = 1
// gradient vertices
vector<glhr::colored_vertex> gv;
gv.emplace_back(-1, -1, 0, 0, 0);
gv.emplace_back(+1, -1, 0, 0, 0);
gv.emplace_back(+1, +1, 1, 1, 1);
gv.emplace_back(-1, -1, 0, 0, 0);
gv.emplace_back(+1, +1, 1, 1, 1);
gv.emplace_back(-1, +1, 1, 1, 1);
#if CAP_RAY
dynamicval<bool> riu(ray::in_use, false);
#endif
if(1) {
current_display->next_shader_flags = GF_VARCOLOR;
dynamicval<eModel> m(pmodel, mdPixel);
current_display->set_all(0,0);
glhr::new_projection();
glhr::id_modelview();
glhr::prepare(gv);
glhr::set_depthtest(false);
glDrawArrays(GL_TRIANGLES, 0, isize(gv));
}
shOverFloor.pstrength = 20;
shFeatherFloor.pstrength = 40;
shFeatherFloor.fstrength = 5;
shTrollFloor.pstrength = 25;
shCaveFloor.pstrength = 40;
shCaveFloor.fstrength = 0;
shDesertFloor.pstrength = 30;
shDesertFloor.fstrength =10;
shRoseFloor.pstrength = 30;
shDragonFloor.pstrength = 30;
shBarrowFloor.pstrength = 40;
// all using Tortoise
for(auto v: all_escher_floorshapes) if(v->shapeid2 == 178) v->pstrength = 20;
ptds.clear();
for(auto v: all_plain_floorshapes) draw_shape_for_texture(v);
for(auto v: all_escher_floorshapes) draw_shape_for_texture(v);
drawqueue();
/*
SDL_Surface *sdark = floor_textures->render();
IMAGESAVE(sdark, "texture-test.png");
*/
rb.reset();
last_texture_step = vid.texture_step;
}
EX void make_floor_textures() {
if(noGUI || !vid.usingGL) return;
DEBBI(DF_POLY, ("make_floor_textures"));
dynamicval<eGeometry> g(geometry, gEuclidSquare);
dynamicval<eModel> gm(pmodel, mdDisk);
dynamicval<eVariation> va(variation, eVariation::pure);
dynamicval<geometryinfo1> gie(ginf[geometry].g, giEuclid2);
dynamicval<geometryinfo1> gih(ginf[gNormal].g, giHyperb2);
dynamicval<bool> a3(vid.always3, false);
dynamicval<bool> hq(inHighQual, true);
dynamicval<int> hd(darken, 0);
dynamicval<ld> hll(levellines, 0);
dynamicval<ld> gd(vid.depth, 1);
dynamicval<ld> gc(vid.camera, 1);
dynamicval<geometry_information*> dcgip(cgip, cgip);
dynamicval<eStereo> gvs(vid.stereo_mode, sOFF);
dynamicval<int> vgp(global_projection, 0);
check_cgi();
cgi.make_floor_textures_here();
/* update texture ID in existing cgi's */
for(auto& c: cgis) c.second.models_texture.texture_id = floor_textures->renderedTexture;
}
#endif
}