1
0
mirror of https://github.com/zenorogue/hyperrogue.git synced 2024-11-16 02:04:48 +00:00
hyperrogue/euclid.cpp

669 lines
18 KiB
C++
Raw Normal View History

// Hyperbolic Rogue -- Euclidean geometry, including 2D, 3D, and quotient spaces
2019-02-24 21:12:32 +00:00
// Copyright (C) 2011-2018 Zeno Rogue, see 'hyper.cpp' for details
namespace hr {
// 2D Euclidean space
// --- euclidean geometry ---
// NOTE: patterns assume that pair_to_vec(0,1) % 3 == 2!
// Thus, pair_to_vec(0,1) must not be e.g. a power of four
int pair_to_vec(int x, int y) {
return x + (y << 15);
}
pair<int, int> vec_to_pair(int vec) {
int x = vec & ((1<<15)-1);
int y = (vec >> 15);
if(x >= (1<<14)) x -= (1<<15), y++;
return {x, y};
}
namespace torusconfig {
// the configuration of the torus topology.
// torus cells are indexed [0..qty),
// where the cell to the right from i is indexed i+dx,
// and the cell to the down-right is numbered i+dy
// Changed with command line option: -tpar <qty>,<dx>,<dy>
// Ideally, qty, dx, and dy should have the same "modulo 3"
// values as the default -- otherwise the three-color
// pattern breaks. Also, they should have no common
// prime divisor.
int def_qty = 127*3, dx = 1, def_dy = -11*2;
int qty = def_qty, dy = def_dy;
int sdx = 12, sdy = 12;
// new values to change
int newqty, newdy, newsdx, newsdy;
int torus_cx, torus_cy;
vector<torusmode_info> tmodes = {
{"single row (hex)", TF_SINGLE | TF_HEX},
{"single row (squares)", TF_SINGLE | TF_SQUARE},
{"parallelogram (hex)", TF_SIMPLE | TF_HEX},
{"rectangle (squares)", TF_SIMPLE | TF_SQUARE},
{"rectangle (hex)", TF_WEIRD | TF_HEX},
{"Klein bottle (squares)", TF_SIMPLE | TF_KLEIN | TF_SQUARE},
{"Klein bottle (hex)", TF_WEIRD | TF_KLEIN | TF_HEX},
{"cylinder (squares)", TF_SIMPLE | TF_CYL },
{"cylinder (hex)", TF_SIMPLE | TF_CYL | TF_HEX},
{"Möbius band (squares)", TF_SIMPLE | TF_CYL | TF_KLEIN},
{"Möbius band (hex)", TF_SIMPLE | TF_CYL | TF_HEX | TF_KLEIN},
};
eTorusMode torus_mode, newmode;
flagtype tmflags() { return tmodes[torus_mode].flags; }
int getqty() {
if(tmflags() & TF_SINGLE)
return qty;
else
return sdx * sdy;
}
int getvec(int x, int y) {
if(tmflags() & TF_SINGLE)
return x * dx + y * dy;
else if(tmflags() & TF_SIMPLE)
return pair_to_vec(x, y);
else
return pair_to_vec(-y - 2 * x, 3 * y);
}
int id_to_vec(int id, bool mirrored = false) {
if(tmflags() & TF_SINGLE)
return id;
else {
int dx = id % sdx;
int dy = id / sdx;
if(mirrored)
dy = -dy, dx += sdx;
if(tmflags() & TF_SIMPLE)
return pair_to_vec(dx, dy);
else
return pair_to_vec(- 2 * dx - (dy & 1), 3 * dy);
}
}
pair<int, bool> vec_to_id_mirror(int vec) {
if(tmflags() & TF_SINGLE) {
return {gmod(vec, qty), false};
}
else {
int x, y;
tie(x,y) = vec_to_pair(vec);
bool mirror = false;
if(tmflags() & TF_KLEIN) {
if(tmflags() & TF_WEIRD) {
x = gmod(x, 4 * sdx);
mirror = x > 0 && x <= 2 * sdx;
}
else {
x = gmod(x, 2 * sdx);
mirror = x >= sdx;
}
if(mirror) y = -y;
}
if(tmflags() & TF_WEIRD) {
y /= 3; x = (x + (y&1)) / -2;
}
x = gmod(x, sdx), y = gmod(y, sdy);
return {y * sdx + x, mirror};
}
}
int vec_to_id(int vec) {
return vec_to_id_mirror(vec).first;
}
void torus_test() {
printf("Testing torus vec_to_pair/pair_to_vec...\n");
for(int x=-10; x<=10; x++)
for(int y=-10; y<=10; y++) {
auto p = vec_to_pair(pair_to_vec(x, y));
if(p.first != x || p.second != y)
printf("Failed for (%d,%d) -> [%d] -> (%d,%d)\n", x, y, pair_to_vec(x,y), p.first, p.second);
}
printf("Testing id_to_vec / vec_to_id...\n");
for(int i=0; i < getqty(); i++)
for(int m=0; m< (torus_mode == tmKlein ? 2 : 1); m++)
if(vec_to_id_mirror(id_to_vec(i, m)) != pair<int,bool> (i,m))
printf("Failed for id %d.%d [%d] (%d.%d)\n", i, m, id_to_vec(i,m), vec_to_id(id_to_vec(i,m)), vec_to_id_mirror(id_to_vec(i,m)).second);
}
int tester = addHook(hooks_tests, 0, torus_test);
void activate() {
auto& gi(ginf[gTorus]);
if(tmflags() & TF_HEX)
gi.vertex = 3, gi.sides = 6, gi.tiling_name = "{6,3}";
else
gi.vertex = 4, gi.sides = 4, gi.tiling_name = "{4,4}";
flagtype& flags = gi.flags;
set_flag(flags, qNONORIENTABLE, tmflags() & TF_KLEIN);
set_flag(flags, qBOUNDED, !(tmflags() & TF_CYL));
int i = 0;
if(tmflags() & TF_KLEIN) i++;
if(tmflags() & TF_CYL) i+=2;
const char *quonames[4] = {"torus", "Klein bottle", "cylinder", "Möbius band"};
gi.quotient_name = quonames[i];
}
int dscalar(gp::loc e1, gp::loc e2) {
return 2 * (e1.first * e2.first + e1.second*e2.second) + (S3 == 3 ? e1.first*e2.second + e2.first * e1.second : 0);
}
int dcross(gp::loc e1, gp::loc e2) {
return e1.first * e2.second - e1.second*e2.first;
}
gp::loc sdxy() { return gp::loc(sdx, sdy); }
int mobius_dir_basic() {
int dscalars[6];
for(int a=0; a<SG6; a++)
dscalars[a] = dscalar(gp::eudir(a), sdxy());
for(int a=0; a<SG6; a++)
for(int b=0; b<SG6; b++)
if(a != b && dscalars[a] == dscalars[b]) {
return (a + b) % SG6;
}
return -1;
}
bool mobius_symmetric(bool square, int dx, int dy) {
dynamicval<eGeometry> g(geometry, square ? gEuclidSquare : gEuclid);
dynamicval<int> gx(sdx, dx);
dynamicval<int> gy(sdy, dy);
return mobius_dir_basic() != -1;
}
void mobius_flip(int&x, int& y) {
int d = mobius_dir_basic();
int a, b;
if(d == 0) a = 1, b = SG6-1;
else a = 0, b = d;
auto p1 = gp::eudir(a);
auto p2 = gp::eudir(b);
// x = sdx * s + px * t
// y = sdy * s + py * t
// py * x = py * sdx * s + px * py * t
// px * y = px * sdy * s + px + py * t
// py * x - px * y = py * sdx * s - px * sdy * s
// s = (py * x - px * y) / (py * sdx - px * sdy)
int det = p1.second * sdx - p1.first * sdy;
int smul = p1.second * x - p1.first * y;
int tmul = sdx * y - sdy * x;
x = (tmul * p2.first + smul * sdx) / det;
y = (tmul * p2.second + smul * sdy) / det;
// println(hlog, make_pair(ox,oy), " [", d, "] ", make_pair(x,y), " p1 = ", p1, " p2 = ", p2, " det = ", det, " smul = ", smul, " tmul = ", tmul);
}
int mobius_dir(cell *c) {
if(c->type == 8) return mobius_dir_basic() * 2;
else return mobius_dir_basic();
}
bool be_canonical(int& x, int& y) {
using namespace torusconfig;
int periods = gdiv(dscalar(gp::loc(x,y), sdxy()), dscalar(sdxy(), sdxy()));
y -= sdy * periods;
x -= sdx * periods;
bool b = false;
if(nonorientable && (periods & 1)) {
mobius_flip(x, y);
b = true;
}
return b;
}
int cyldist(int id1, int id2) {
int x1, y1, x2, y2;
tie(x1, y1) = vec_to_pair(id1);
tie(x2, y2) = vec_to_pair(id2);
be_canonical(x1, y1);
be_canonical(x2, y2);
int dist = 1000000000;
for(int a1=-1; a1<=1; a1++)
for(int a2=-1; a2<=1; a2++) {
int ax1 = x1 + sdx * a1;
int ay1 = y1 + sdy * a1;
if(nonorientable && a1) mobius_flip(ax1, ay1);
int ax2 = x2 + sdx * a2;
int ay2 = y2 + sdy * a2;
if(nonorientable && a2) mobius_flip(ax2, ay2);
dist = min(dist, eudist(ax1 - ax2, ay1 - ay2));
}
return dist;
}
}
int euclid_getvec(int dx, int dy) {
if(euwrap) return torusconfig::getvec(dx, dy);
else return pair_to_vec(dx, dy);
}
template<class T> void build_euclidean_moves(cell *c, int vec, const T& builder) {
int x, y;
tie(x,y) = vec_to_pair(vec);
c->type = a4 ? (PURE || ((x^y^1) & 1) ? 4 : 8) : 6;
if(c->type == 4) {
int m = PURE ? 1 : 2;
builder(euclid_getvec(+1,+0), 0, 2 * m);
builder(euclid_getvec(+0,+1), 1, 3 * m);
builder(euclid_getvec(-1,+0), 2, 0 * m);
builder(euclid_getvec(+0,-1), 3, 1 * m);
}
else if(c->type == 8) {
builder(euclid_getvec(+1,+0), 0, 2);
builder(euclid_getvec(+1,+1), 1, 5);
builder(euclid_getvec(+0,+1), 2, 3);
builder(euclid_getvec(-1,+1), 3, 7);
builder(euclid_getvec(-1,+0), 4, 0);
builder(euclid_getvec(-1,-1), 5, 1);
builder(euclid_getvec(+0,-1), 6, 1);
builder(euclid_getvec(+1,-1), 7, 3);
}
else /* 6 */ {
builder(euclid_getvec(+1,+0), 0, 3);
builder(euclid_getvec(+0,+1), 1, 4);
builder(euclid_getvec(-1,+1), 2, 5);
builder(euclid_getvec(-1,+0), 3, 0);
builder(euclid_getvec(+0,-1), 4, 1);
builder(euclid_getvec(+1,-1), 5, 2);
}
}
struct hrmap_torus : hrmap {
vector<cell*> all;
vector<int> dists;
virtual vector<cell*>& allcells() { return all; }
cell *gamestart() {
return all[0];
}
hrmap_torus() {
using namespace torusconfig;
int q = getqty();
all.resize(q);
for(int i=0; i<q; i++) {
all[i] = newCell(8, encodeId(i));
}
for(int i=0; i<q; i++) {
int iv = id_to_vec(i);
build_euclidean_moves(all[i], iv, [&] (int delta, int d, int d2) {
auto im = vec_to_id_mirror(iv + delta);
all[i]->move(d) = all[im.first];
all[i]->c.setspin(d, im.second, false);
});
}
for(cell *c: all) for(int d=0; d<c->type; d++) {
cell *c2 = c->move(d);
for(int d2=0; d2<c2->type; d2++)
if(c2->move(d2) == c)
c->c.setspin(d, d2, c->c.spin(d));
}
celllister cl(gamestart(), 100, 100000000, NULL);
dists.resize(q);
for(int i=0; i<isize(cl.lst); i++)
dists[decodeId(cl.lst[i]->master)] = cl.dists[i];
}
~hrmap_torus() {
for(cell *c: all) tailored_delete(c);
}
};
hrmap_torus *torusmap() {
return dynamic_cast<hrmap_torus*> (currentmap);
}
/* cell *getTorusId(int id) {
hrmap_torus *cur = torusmap();
if(!cur) return NULL;
return cur->all[id];
} */
struct hrmap_euclidean : hrmap {
cell *gamestart() {
return *(euclideanAtCreate(0).first);
}
struct euclideanSlab {
cell* a[256][256];
euclideanSlab() {
for(int y=0; y<256; y++) for(int x=0; x<256; x++)
a[y][x] = NULL;
}
~euclideanSlab() {
for(int y=0; y<256; y++) for(int x=0; x<256; x++)
if(a[y][x]) tailored_delete(a[y][x]);
}
};
static const int slabs = max_vec / 256;
euclideanSlab* euclidean[slabs][slabs];
hrmap_euclidean() {
for(int y=0; y<slabs; y++) for(int x=0; x<slabs; x++)
euclidean[y][x] = NULL;
}
euc_pointer at(int vec) {
auto p = vec_to_pair(vec);
int x = p.first, y = p.second;
bool mobius = false;
if(euwrap)
mobius = torusconfig::be_canonical(x, y);
euclideanSlab*& slab = euclidean[(y>>8)&(slabs-1)][(x>>8)&(slabs-1)];
if(!slab) slab = new hrmap_euclidean::euclideanSlab;
return make_pair(&(slab->a[y&255][x&255]), mobius);
}
map<int, struct cdata> eucdata;
~hrmap_euclidean() {
for(int y=0; y<slabs; y++) for(int x=0; x<slabs; x++)
if(euclidean[y][x]) {
tailored_delete(euclidean[y][x]);
euclidean[y][x] = NULL;
}
eucdata.clear();
}
};
cellwalker vec_to_cellwalker(int vec) {
if(!fulltorus) {
auto p = euclideanAtCreate(vec);
if(p.second)
return cellwalker(*p.first, torusconfig::mobius_dir(*p.first), true);
else
return cellwalker(*p.first, 0, false);
}
else {
hrmap_torus *cur = torusmap();
if(!cur) return cellwalker(NULL, 0);
auto p = torusconfig::vec_to_id_mirror(vec);
return cellwalker(cur->all[p.first], 0, p.second);
}
}
int cellwalker_to_vec(cellwalker cw) {
int id = decodeId(cw.at->master);
if(!fulltorus) {
if(nonorientable) {
auto ep = euclideanAt(id);
if(ep.second != cw.mirrored) {
int x, y;
tie(x, y) = vec_to_pair(id);
x += torusconfig::sdx;
y += torusconfig::sdy;
torusconfig::mobius_flip(x, y);
return pair_to_vec(x, y);
}
}
return id;
}
return torusconfig::id_to_vec(id, cw.mirrored);
}
int cell_to_vec(cell *c) {
int id = decodeId(c->master);
if(!fulltorus) return id;
return torusconfig::id_to_vec(id, false);
}
pair<int, int> cell_to_pair(cell *c) {
return vec_to_pair(cell_to_vec(c));
}
union heptacoder {
heptagon *h;
int id;
};
int decodeId(heptagon* h) {
heptacoder u;
u.h = h; return u.id;
}
heptagon* encodeId(int id) {
heptacoder u;
u.id = id;
return u.h;
}
// 3D Euclidean space
#if MAXMDIM == 4
namespace euclid3 {
2019-02-24 21:12:32 +00:00
typedef long long coord;
static const long long COORDMAX = (1<<16);
2019-03-02 00:19:31 +00:00
array<int, 3> getcoord(coord x) {
array<int, 3> res;
for(int k=0; k<3; k++) {
int next = x % COORDMAX;
if(next>COORDMAX/2) next -= COORDMAX;
if(next<-COORDMAX/2) next += COORDMAX;
res[k] = next;
x -= next;
x /= COORDMAX;
}
return res;
}
2019-03-01 03:18:19 +00:00
vector<coord> get_shifttable() {
static const coord D0 = 1;
static const coord D1 = COORDMAX;
static const coord D2 = COORDMAX * COORDMAX;
vector<coord> shifttable;
vector<transmatrix> tmatrix;
switch(geometry) {
case gCubeTiling:
shifttable = { +D0, +D1, +D2 };
break;
case gRhombic3:
shifttable = { D0+D1, D0+D2, D1+D2, D1-D2, D0-D2, D0-D1 };
break;
case gBitrunc3:
shifttable = { 2*D0, 2*D1, 2*D2, D0+D1+D2, D0+D1-D2, D0-D1-D2, D0-D1+D2 };
break;
default:
printf("euclid3::get_shifttable() called in geometry that is not euclid3");
exit(1);
}
// reverse everything
int s = isize(shifttable);
for(int i=0; i<s; i++) shifttable.push_back(-shifttable[i]);
return shifttable;
}
struct hrmap_euclid3 : hrmap {
vector<coord> shifttable;
vector<transmatrix> tmatrix;
2019-02-24 21:12:32 +00:00
map<coord, heptagon*> spacemap;
map<heptagon*, coord> ispacemap;
hrmap_euclid3() {
shifttable = get_shifttable();
tmatrix.resize(S7);
for(int i=0; i<S7; i++) tmatrix[i] = Id;
for(int i=0; i<S7; i++) for(int j=0; j<3; j++)
2019-03-02 00:19:31 +00:00
tmatrix[i][j][DIM] = getcoord(shifttable[i])[j];
2019-02-24 21:12:32 +00:00
getOrigin();
}
heptagon *getOrigin() {
return get_at(0);
}
heptagon *get_at(coord at) {
if(spacemap.count(at))
return spacemap[at];
else {
auto h = tailored_alloc<heptagon> (S7);
h->c7 = newCell(S7, h);
2019-02-24 21:12:32 +00:00
h->distance = 0;
h->cdata = NULL;
2019-03-02 00:19:31 +00:00
auto co = getcoord(at);
if(S7 != 14)
2019-03-02 00:19:31 +00:00
h->zebraval = gmod(co[0] + co[1] * 2 + co[2] * 4, 5);
else
2019-03-02 00:19:31 +00:00
h->zebraval = co[0] & 1;
2019-02-24 21:12:32 +00:00
spacemap[at] = h;
ispacemap[h] = at;
return h;
}
}
heptagon *build(heptagon *parent, int d, coord at) {
auto h = get_at(at);
h->c.connect((d+S7/2)%S7, parent, d, false);
2019-02-24 21:12:32 +00:00
return h;
}
heptagon *createStep(heptagon *parent, int d) {
return build(parent, d, ispacemap[parent] + shifttable[d]);
2019-02-24 21:12:32 +00:00
}
};
hrmap_euclid3* cubemap() {
return ((hrmap_euclid3*) currentmap);
2019-02-24 21:12:32 +00:00
}
hrmap* new_map() {
return new hrmap_euclid3;
2019-02-24 21:12:32 +00:00
}
heptagon *createStep(heptagon *parent, int d) {
return cubemap()->createStep(parent, d);
}
2019-02-27 13:40:56 +00:00
bool pseudohept(cell *c) {
coord co = cubemap()->ispacemap[c->master];
2019-03-02 00:19:31 +00:00
auto v = getcoord(co);
if(S7 == 12) {
for(int i=0; i<3; i++) if((v[i] & 1)) return false;
}
else {
for(int i=0; i<3; i++) if(!(v[i] & 1)) return false;
}
2019-02-27 13:40:56 +00:00
return true;
}
int dist_alt(cell *c) {
2019-02-27 14:50:26 +00:00
coord co = cubemap()->ispacemap[c->master];
2019-03-02 00:19:31 +00:00
auto v = getcoord(co);
if(S7 == 6) return v[2];
else if(S7 == 12) return (v[0] + v[1] + v[2]) / 2;
else return v[2]/2;
2019-02-27 14:50:26 +00:00
}
2019-02-24 21:12:32 +00:00
void draw() {
dq::visited.clear();
dq::enqueue(viewctr.at, cview());
auto cm = cubemap();
2019-02-24 21:12:32 +00:00
while(!dq::drawqueue.empty()) {
auto& p = dq::drawqueue.front();
heptagon *h = get<0>(p);
transmatrix V = get<1>(p);
2019-02-24 21:12:32 +00:00
dynamicval<ld> b(band_shift, get<2>(p));
bandfixer bf(V);
dq::drawqueue.pop();
2019-02-24 21:12:32 +00:00
cell *c = h->c7;
if(!do_draw(c, V)) continue;
drawcell(c, V, 0, false);
for(int i=0; i<S7; i++)
dq::enqueue(h->move(i), V * cm->tmatrix[i]);
2019-02-24 21:12:32 +00:00
}
}
transmatrix relative_matrix(heptagon *h2, heptagon *h1) {
auto cm = cubemap();
2019-03-02 00:19:31 +00:00
auto v = getcoord(cm->ispacemap[h2] - cm->ispacemap[h1]);
return eupush3(v[0], v[1], v[2]);
}
bool get_emerald(cell *c) {
auto v = getcoord(cubemap()->ispacemap[c->master]);
int s0 = 0, s1 = 0;
for(int i=0; i<3; i++) {
v[i] = gmod(v[i], 6);
int d = min(v[i], 6-v[i]);;
s0 += min(v[i], 6-v[i]);
s1 += 3-d;
}
if(s0 == s1) println(hlog, "equality");
return s0 > s1;
2019-02-24 21:12:32 +00:00
}
int celldistance(cell *c1, cell *c2) {
auto cm = cubemap();
2019-03-02 00:19:31 +00:00
auto v = getcoord(cm->ispacemap[c1->master] - cm->ispacemap[c2->master]);
if(S7 == 6)
2019-03-02 00:19:31 +00:00
return abs(v[0]) + abs(v[1]) + abs(v[2]);
else {
2019-03-02 00:19:31 +00:00
for(int i=0; i<3; i++) v[i] = abs(v[i]);
sort(v.begin(), v.end());
int dist = 0;
if(S7 == 12) {
2019-03-02 00:19:31 +00:00
int d = v[1] - v[0]; v[1] -= d; v[2] -= d;
dist += d;
2019-03-02 00:19:31 +00:00
int m = min((v[2] - v[0]) / 2, v[0]);
dist += 2 * d;
2019-03-02 00:19:31 +00:00
v[0] -= m; v[1] -= m; v[2] -= m;
if(v[0])
dist += (v[0] + v[1] + v[2]) / 2;
else
2019-03-02 00:19:31 +00:00
dist += v[2];
}
else {
2019-03-02 00:19:31 +00:00
dist = v[0] + (v[1] - v[0]) / 2 + (v[2] - v[0]) / 2;
}
return dist;
}
2019-02-24 21:12:32 +00:00
}
}
#endif
}