1 | // Copyright 2009-2021 Intel Corporation |
2 | // SPDX-License-Identifier: Apache-2.0 |
3 | |
4 | #pragma once |
5 | |
6 | #include "../common/geometry.h" |
7 | #include "../common/buffer.h" |
8 | #include "half_edge.h" |
9 | #include "catmullclark_coefficients.h" |
10 | |
11 | namespace embree |
12 | { |
13 | struct __aligned(64) FinalQuad { |
14 | Vec3fa vtx[4]; |
15 | }; |
16 | |
17 | template<typename Vertex, typename Vertex_t = Vertex> |
18 | struct __aligned(64) CatmullClark1RingT |
19 | { |
20 | ALIGNED_STRUCT_(64); |
21 | |
22 | int border_index; //!< edge index where border starts |
23 | unsigned int face_valence; //!< number of adjacent quad faces |
24 | unsigned int edge_valence; //!< number of adjacent edges (2*face_valence) |
25 | float vertex_crease_weight; //!< weight of vertex crease (0 if no vertex crease) |
26 | DynamicStackArray<float,16,MAX_RING_FACE_VALENCE> crease_weight; //!< edge crease weights for each adjacent edge |
27 | float vertex_level; //!< maximum level of all adjacent edges |
28 | float edge_level; //!< level of first edge |
29 | unsigned int eval_start_index; //!< topology dependent index to start evaluation |
30 | unsigned int eval_unique_identifier; //!< topology dependent unique identifier for this ring |
31 | Vertex vtx; //!< center vertex |
32 | DynamicStackArray<Vertex,32,MAX_RING_EDGE_VALENCE> ring; //!< ring of neighboring vertices |
33 | |
34 | public: |
35 | CatmullClark1RingT () |
36 | : eval_start_index(0), eval_unique_identifier(0) {} // FIXME: default constructor should be empty |
37 | |
38 | /*! calculates number of bytes required to serialize this structure */ |
39 | __forceinline size_t bytes() const |
40 | { |
41 | size_t ofs = 0; |
42 | ofs += sizeof(border_index); |
43 | ofs += sizeof(face_valence); |
44 | assert(2*face_valence == edge_valence); |
45 | ofs += sizeof(vertex_crease_weight); |
46 | ofs += face_valence*sizeof(float); |
47 | ofs += sizeof(vertex_level); |
48 | ofs += sizeof(edge_level); |
49 | ofs += sizeof(eval_start_index); |
50 | ofs += sizeof(eval_unique_identifier); |
51 | ofs += sizeof(vtx); |
52 | ofs += edge_valence*sizeof(Vertex); |
53 | return ofs; |
54 | } |
55 | |
56 | template<typename Ty> |
57 | static __forceinline void store(char* ptr, size_t& ofs, const Ty& v) { |
58 | *(Ty*)&ptr[ofs] = v; ofs += sizeof(Ty); |
59 | } |
60 | |
61 | template<typename Ty> |
62 | static __forceinline void load(char* ptr, size_t& ofs, Ty& v) { |
63 | v = *(Ty*)&ptr[ofs]; ofs += sizeof(Ty); |
64 | } |
65 | |
66 | /*! serializes the ring to some memory location */ |
67 | __forceinline void serialize(char* ptr, size_t& ofs) const |
68 | { |
69 | store(ptr,ofs,border_index); |
70 | store(ptr,ofs,face_valence); |
71 | store(ptr,ofs,vertex_crease_weight); |
72 | for (size_t i=0; i<face_valence; i++) |
73 | store(ptr,ofs,crease_weight[i]); |
74 | store(ptr,ofs,vertex_level); |
75 | store(ptr,ofs,edge_level); |
76 | store(ptr,ofs,eval_start_index); |
77 | store(ptr,ofs,eval_unique_identifier); |
78 | Vertex_t::storeu(&ptr[ofs],vtx); ofs += sizeof(Vertex); |
79 | for (size_t i=0; i<edge_valence; i++) { |
80 | Vertex_t::storeu(&ptr[ofs],ring[i]); ofs += sizeof(Vertex); |
81 | } |
82 | } |
83 | |
84 | /*! deserializes the ring from some memory location */ |
85 | __forceinline void deserialize(char* ptr, size_t& ofs) |
86 | { |
87 | load(ptr,ofs,border_index); |
88 | load(ptr,ofs,face_valence); |
89 | edge_valence = 2*face_valence; |
90 | load(ptr,ofs,vertex_crease_weight); |
91 | for (size_t i=0; i<face_valence; i++) |
92 | load(ptr,ofs,crease_weight[i]); |
93 | load(ptr,ofs,vertex_level); |
94 | load(ptr,ofs,edge_level); |
95 | load(ptr,ofs,eval_start_index); |
96 | load(ptr,ofs,eval_unique_identifier); |
97 | vtx = Vertex_t::loadu(&ptr[ofs]); ofs += sizeof(Vertex); |
98 | for (size_t i=0; i<edge_valence; i++) { |
99 | ring[i] = Vertex_t::loadu(&ptr[ofs]); ofs += sizeof(Vertex); |
100 | } |
101 | } |
102 | |
103 | __forceinline bool hasBorder() const { |
104 | return border_index != -1; |
105 | } |
106 | |
107 | __forceinline const Vertex& front(size_t i) const { |
108 | assert(edge_valence>i); |
109 | return ring[i]; |
110 | } |
111 | |
112 | __forceinline const Vertex& back(size_t i) const { |
113 | assert(edge_valence>=i); |
114 | return ring[edge_valence-i]; |
115 | } |
116 | |
117 | __forceinline bool has_last_face() const { |
118 | return (size_t)border_index != (size_t)edge_valence-2; |
119 | } |
120 | |
121 | __forceinline bool has_opposite_front(size_t i) const { |
122 | return (size_t)border_index != 2*i; |
123 | } |
124 | |
125 | __forceinline bool has_opposite_back(size_t i) const { |
126 | return (size_t)border_index != ((size_t)edge_valence-2-2*i); |
127 | } |
128 | |
129 | __forceinline BBox3fa bounds() const |
130 | { |
131 | BBox3fa bounds ( vtx ); |
132 | for (size_t i = 0; i<edge_valence ; i++) |
133 | bounds.extend( ring[i] ); |
134 | return bounds; |
135 | } |
136 | |
137 | /*! initializes the ring from the half edge structure */ |
138 | __forceinline void init(const HalfEdge* const h, const char* vertices, size_t stride) |
139 | { |
140 | border_index = -1; |
141 | vtx = Vertex_t::loadu(vertices+h->getStartVertexIndex()*stride); |
142 | vertex_crease_weight = h->vertex_crease_weight; |
143 | |
144 | HalfEdge* p = (HalfEdge*) h; |
145 | |
146 | unsigned i=0; |
147 | unsigned min_vertex_index = (unsigned)-1; |
148 | unsigned min_vertex_index_face = (unsigned)-1; |
149 | edge_level = p->edge_level; |
150 | vertex_level = 0.0f; |
151 | |
152 | do |
153 | { |
154 | vertex_level = max(a: vertex_level,b: p->edge_level); |
155 | crease_weight[i/2] = p->edge_crease_weight; |
156 | assert(p->hasOpposite() || p->edge_crease_weight == float(inf)); |
157 | |
158 | /* store first two vertices of face */ |
159 | p = p->next(); |
160 | const unsigned index0 = p->getStartVertexIndex(); |
161 | ring[i++] = Vertex_t::loadu(vertices+index0*stride); |
162 | if (index0 < min_vertex_index) { min_vertex_index = index0; min_vertex_index_face = i>>1; } |
163 | p = p->next(); |
164 | |
165 | const unsigned index1 = p->getStartVertexIndex(); |
166 | ring[i++] = Vertex_t::loadu(vertices+index1*stride); |
167 | p = p->next(); |
168 | |
169 | /* continue with next face */ |
170 | if (likely(p->hasOpposite())) |
171 | p = p->opposite(); |
172 | |
173 | /* if there is no opposite go the long way to the other side of the border */ |
174 | else |
175 | { |
176 | /* find minimum start vertex */ |
177 | const unsigned index0 = p->getStartVertexIndex(); |
178 | if (index0 < min_vertex_index) { min_vertex_index = index0; min_vertex_index_face = i>>1; } |
179 | |
180 | /*! mark first border edge and store dummy vertex for face between the two border edges */ |
181 | border_index = i; |
182 | crease_weight[i/2] = inf; |
183 | ring[i++] = Vertex_t::loadu(vertices+index0*stride); |
184 | ring[i++] = vtx; // dummy vertex |
185 | |
186 | /*! goto other side of border */ |
187 | p = (HalfEdge*) h; |
188 | while (p->hasOpposite()) |
189 | p = p->opposite()->next(); |
190 | } |
191 | |
192 | } while (p != h); |
193 | |
194 | edge_valence = i; |
195 | face_valence = i >> 1; |
196 | eval_unique_identifier = min_vertex_index; |
197 | eval_start_index = min_vertex_index_face; |
198 | |
199 | assert( hasValidPositions() ); |
200 | } |
201 | |
202 | __forceinline void subdivide(CatmullClark1RingT& dest) const |
203 | { |
204 | dest.edge_level = 0.5f*edge_level; |
205 | dest.vertex_level = 0.5f*vertex_level; |
206 | dest.face_valence = face_valence; |
207 | dest.edge_valence = edge_valence; |
208 | dest.border_index = border_index; |
209 | dest.vertex_crease_weight = max(a: 0.0f,b: vertex_crease_weight-1.0f); |
210 | dest.eval_start_index = eval_start_index; |
211 | dest.eval_unique_identifier = eval_unique_identifier; |
212 | |
213 | /* calculate face points */ |
214 | Vertex_t S = Vertex_t(0.0f); |
215 | for (size_t i=0; i<face_valence; i++) |
216 | { |
217 | size_t face_index = i + eval_start_index; if (face_index >= face_valence) face_index -= face_valence; assert(face_index < face_valence); |
218 | size_t index0 = 2*face_index+0; if (index0 >= edge_valence) index0 -= edge_valence; assert(index0 < edge_valence); |
219 | size_t index1 = 2*face_index+1; if (index1 >= edge_valence) index1 -= edge_valence; assert(index1 < edge_valence); |
220 | size_t index2 = 2*face_index+2; if (index2 >= edge_valence) index2 -= edge_valence; assert(index2 < edge_valence); |
221 | S += dest.ring[index1] = ((vtx + ring[index1]) + (ring[index0] + ring[index2])) * 0.25f; |
222 | } |
223 | |
224 | /* calculate new edge points */ |
225 | size_t num_creases = 0; |
226 | array_t<size_t,MAX_RING_FACE_VALENCE> crease_id; |
227 | |
228 | for (size_t i=0; i<face_valence; i++) |
229 | { |
230 | size_t face_index = i + eval_start_index; |
231 | if (face_index >= face_valence) face_index -= face_valence; |
232 | const float edge_crease = crease_weight[face_index]; |
233 | dest.crease_weight[face_index] = max(a: edge_crease-1.0f,b: 0.0f); |
234 | |
235 | size_t index = 2*face_index; |
236 | size_t prev_index = face_index == 0 ? edge_valence-1 : 2*face_index-1; |
237 | size_t next_index = 2*face_index+1; |
238 | |
239 | const Vertex_t v = vtx + ring[index]; |
240 | const Vertex_t f = dest.ring[prev_index] + dest.ring[next_index]; |
241 | S += ring[index]; |
242 | |
243 | /* fast path for regular edge points */ |
244 | if (likely(edge_crease <= 0.0f)) { |
245 | dest.ring[index] = (v+f) * 0.25f; |
246 | } |
247 | |
248 | /* slower path for hard edge rule */ |
249 | else { |
250 | crease_id[num_creases++] = face_index; |
251 | dest.ring[index] = v*0.5f; |
252 | |
253 | /* even slower path for blended edge rule */ |
254 | if (unlikely(edge_crease < 1.0f)) { |
255 | dest.ring[index] = lerp((v+f)*0.25f,v*0.5f,edge_crease); |
256 | } |
257 | } |
258 | } |
259 | |
260 | /* compute new vertex using smooth rule */ |
261 | const float inv_face_valence = 1.0f / (float)face_valence; |
262 | const Vertex_t v_smooth = (Vertex_t) madd(inv_face_valence,S,(float(face_valence)-2.0f)*vtx)*inv_face_valence; |
263 | dest.vtx = v_smooth; |
264 | |
265 | /* compute new vertex using vertex_crease_weight rule */ |
266 | if (unlikely(vertex_crease_weight > 0.0f)) |
267 | { |
268 | if (vertex_crease_weight >= 1.0f) { |
269 | dest.vtx = vtx; |
270 | } else { |
271 | dest.vtx = lerp(v_smooth,vtx,vertex_crease_weight); |
272 | } |
273 | return; |
274 | } |
275 | |
276 | /* no edge crease rule and dart rule */ |
277 | if (likely(num_creases <= 1)) |
278 | return; |
279 | |
280 | /* compute new vertex using crease rule */ |
281 | if (likely(num_creases == 2)) |
282 | { |
283 | /* update vertex using crease rule */ |
284 | const size_t crease0 = crease_id[0], crease1 = crease_id[1]; |
285 | const Vertex_t v_sharp = (Vertex_t)(ring[2*crease0] + 6.0f*vtx + ring[2*crease1]) * (1.0f / 8.0f); |
286 | dest.vtx = v_sharp; |
287 | |
288 | /* update crease_weights using chaikin rule */ |
289 | const float crease_weight0 = crease_weight[crease0], crease_weight1 = crease_weight[crease1]; |
290 | dest.crease_weight[crease0] = max(a: 0.25f*(3.0f*crease_weight0 + crease_weight1)-1.0f,b: 0.0f); |
291 | dest.crease_weight[crease1] = max(a: 0.25f*(3.0f*crease_weight1 + crease_weight0)-1.0f,b: 0.0f); |
292 | |
293 | /* interpolate between sharp and smooth rule */ |
294 | const float v_blend = 0.5f*(crease_weight0+crease_weight1); |
295 | if (unlikely(v_blend < 1.0f)) { |
296 | dest.vtx = lerp(v_smooth,v_sharp,v_blend); |
297 | } |
298 | } |
299 | |
300 | /* compute new vertex using corner rule */ |
301 | else { |
302 | dest.vtx = vtx; |
303 | } |
304 | } |
305 | |
306 | __forceinline bool isRegular1() const |
307 | { |
308 | if (border_index == -1) { |
309 | if (face_valence == 4) return true; |
310 | } else { |
311 | if (face_valence < 4) return true; |
312 | } |
313 | return false; |
314 | } |
315 | |
316 | __forceinline size_t numEdgeCreases() const |
317 | { |
318 | ssize_t numCreases = 0; |
319 | for (size_t i=0; i<face_valence; i++) { |
320 | numCreases += crease_weight[i] > 0.0f; |
321 | } |
322 | return numCreases; |
323 | } |
324 | |
325 | enum Type { |
326 | TYPE_NONE = 0, //!< invalid type |
327 | TYPE_REGULAR = 1, //!< regular patch when ignoring creases |
328 | TYPE_REGULAR_CREASES = 2, //!< regular patch when considering creases |
329 | TYPE_GREGORY = 4, //!< gregory patch when ignoring creases |
330 | TYPE_GREGORY_CREASES = 8, //!< gregory patch when considering creases |
331 | TYPE_CREASES = 16 //!< patch has crease features |
332 | }; |
333 | |
334 | __forceinline Type type() const |
335 | { |
336 | /* check if there is an edge crease anywhere */ |
337 | const size_t numCreases = numEdgeCreases(); |
338 | const bool noInnerCreases = hasBorder() ? numCreases == 2 : numCreases == 0; |
339 | |
340 | Type crease_mask = (Type) (TYPE_REGULAR | TYPE_GREGORY); |
341 | if (noInnerCreases ) crease_mask = (Type) (crease_mask | TYPE_REGULAR_CREASES | TYPE_GREGORY_CREASES); |
342 | if (numCreases != 0) crease_mask = (Type) (crease_mask | TYPE_CREASES); |
343 | |
344 | /* calculate if this vertex is regular */ |
345 | bool hasBorder = border_index != -1; |
346 | if (face_valence == 2 && hasBorder) { |
347 | if (vertex_crease_weight == 0.0f ) return (Type) (crease_mask & (TYPE_REGULAR | TYPE_REGULAR_CREASES | TYPE_GREGORY | TYPE_GREGORY_CREASES | TYPE_CREASES)); |
348 | else if (vertex_crease_weight == float(inf)) return (Type) (crease_mask & (TYPE_REGULAR | TYPE_REGULAR_CREASES | TYPE_GREGORY | TYPE_GREGORY_CREASES | TYPE_CREASES)); |
349 | else return TYPE_CREASES; |
350 | } |
351 | else if (vertex_crease_weight != 0.0f) return TYPE_CREASES; |
352 | else if (face_valence == 3 && hasBorder) return (Type) (crease_mask & (TYPE_REGULAR | TYPE_REGULAR_CREASES | TYPE_GREGORY | TYPE_GREGORY_CREASES | TYPE_CREASES)); |
353 | else if (face_valence == 4 && !hasBorder) return (Type) (crease_mask & (TYPE_REGULAR | TYPE_REGULAR_CREASES | TYPE_GREGORY | TYPE_GREGORY_CREASES | TYPE_CREASES)); |
354 | else return (Type) (crease_mask & (TYPE_GREGORY | TYPE_GREGORY_CREASES | TYPE_CREASES)); |
355 | } |
356 | |
357 | __forceinline bool isFinalResolution(float res) const { |
358 | return vertex_level <= res; |
359 | } |
360 | |
361 | /* computes the limit vertex */ |
362 | __forceinline Vertex getLimitVertex() const |
363 | { |
364 | /* return hard corner */ |
365 | if (unlikely(std::isinf(vertex_crease_weight))) |
366 | return vtx; |
367 | |
368 | /* border vertex rule */ |
369 | if (unlikely(border_index != -1)) |
370 | { |
371 | const unsigned int second_border_index = border_index+2 >= int(edge_valence) ? 0 : border_index+2; |
372 | return (4.0f * vtx + (ring[border_index] + ring[second_border_index])) * 1.0f/6.0f; |
373 | } |
374 | |
375 | Vertex_t F( 0.0f ); |
376 | Vertex_t E( 0.0f ); |
377 | |
378 | assert(eval_start_index < face_valence); |
379 | |
380 | for (size_t i=0; i<face_valence; i++) { |
381 | size_t index = i+eval_start_index; |
382 | if (index >= face_valence) index -= face_valence; |
383 | F += ring[2*index+1]; |
384 | E += ring[2*index]; |
385 | } |
386 | |
387 | const float n = (float)face_valence; |
388 | return (Vertex_t)(n*n*vtx+4.0f*E+F) / ((n+5.0f)*n); |
389 | } |
390 | |
391 | /* gets limit tangent in the direction of egde vtx -> ring[0] */ |
392 | __forceinline Vertex getLimitTangent() const |
393 | { |
394 | if (unlikely(std::isinf(vertex_crease_weight))) |
395 | return ring[0] - vtx; |
396 | |
397 | /* border vertex rule */ |
398 | if (unlikely(border_index != -1)) |
399 | { |
400 | if (border_index != (int)edge_valence-2 ) { |
401 | return ring[0] - vtx; |
402 | } |
403 | else |
404 | { |
405 | const unsigned int second_border_index = border_index+2 >= int(edge_valence) ? 0 : border_index+2; |
406 | return (ring[second_border_index] - ring[border_index]) * 0.5f; |
407 | } |
408 | } |
409 | |
410 | Vertex_t alpha( 0.0f ); |
411 | Vertex_t beta ( 0.0f ); |
412 | |
413 | const size_t n = face_valence; |
414 | |
415 | assert(eval_start_index < face_valence); |
416 | |
417 | Vertex_t q( 0.0f ); |
418 | for (size_t i=0; i<face_valence; i++) |
419 | { |
420 | size_t index = i+eval_start_index; |
421 | if (index >= face_valence) index -= face_valence; |
422 | const float a = CatmullClarkPrecomputedCoefficients::table.limittangent_a(i: index,n); |
423 | const float b = CatmullClarkPrecomputedCoefficients::table.limittangent_b(i: index,n); |
424 | alpha += a * ring[2*index]; |
425 | beta += b * ring[2*index+1]; |
426 | } |
427 | |
428 | const float sigma = CatmullClarkPrecomputedCoefficients::table.limittangent_c(n); |
429 | return sigma * (alpha + beta); |
430 | } |
431 | |
432 | /* gets limit tangent in the direction of egde vtx -> ring[edge_valence-2] */ |
433 | __forceinline Vertex getSecondLimitTangent() const |
434 | { |
435 | if (unlikely(std::isinf(vertex_crease_weight))) |
436 | return ring[2] - vtx; |
437 | |
438 | /* border vertex rule */ |
439 | if (unlikely(border_index != -1)) |
440 | { |
441 | if (border_index != 2) { |
442 | return ring[2] - vtx; |
443 | } |
444 | else { |
445 | const unsigned int second_border_index = border_index+2 >= int(edge_valence) ? 0 : border_index+2; |
446 | return (ring[border_index] - ring[second_border_index]) * 0.5f; |
447 | } |
448 | } |
449 | |
450 | Vertex_t alpha( 0.0f ); |
451 | Vertex_t beta ( 0.0f ); |
452 | |
453 | const size_t n = face_valence; |
454 | |
455 | assert(eval_start_index < face_valence); |
456 | |
457 | for (size_t i=0; i<face_valence; i++) |
458 | { |
459 | size_t index = i+eval_start_index; |
460 | if (index >= face_valence) index -= face_valence; |
461 | |
462 | size_t prev_index = index == 0 ? face_valence-1 : index-1; // need to be bit-wise exact in cosf eval |
463 | const float a = CatmullClarkPrecomputedCoefficients::table.limittangent_a(i: prev_index,n); |
464 | const float b = CatmullClarkPrecomputedCoefficients::table.limittangent_b(i: prev_index,n); |
465 | alpha += a * ring[2*index]; |
466 | beta += b * ring[2*index+1]; |
467 | } |
468 | |
469 | const float sigma = CatmullClarkPrecomputedCoefficients::table.limittangent_c(n); |
470 | return sigma* (alpha + beta); |
471 | } |
472 | |
473 | /* gets surface normal */ |
474 | const Vertex getNormal() const { |
475 | return cross(getLimitTangent(),getSecondLimitTangent()); |
476 | } |
477 | |
478 | /* returns center of the n-th quad in the 1-ring */ |
479 | __forceinline Vertex getQuadCenter(const size_t index) const |
480 | { |
481 | const Vertex_t &p0 = vtx; |
482 | const Vertex_t &p1 = ring[2*index+0]; |
483 | const Vertex_t &p2 = ring[2*index+1]; |
484 | const Vertex_t &p3 = index == face_valence-1 ? ring[0] : ring[2*index+2]; |
485 | const Vertex p = (p0+p1+p2+p3) * 0.25f; |
486 | return p; |
487 | } |
488 | |
489 | /* returns center of the n-th edge in the 1-ring */ |
490 | __forceinline Vertex getEdgeCenter(const size_t index) const { |
491 | return (vtx + ring[index*2]) * 0.5f; |
492 | } |
493 | |
494 | bool hasValidPositions() const |
495 | { |
496 | for (size_t i=0; i<edge_valence; i++) { |
497 | if (!isvalid(ring[i])) |
498 | return false; |
499 | } |
500 | return true; |
501 | } |
502 | |
503 | friend __forceinline embree_ostream operator<<(embree_ostream o, const CatmullClark1RingT &c) |
504 | { |
505 | o << "vtx " << c.vtx << " size = " << c.edge_valence << ", " << |
506 | "hard_edge = " << c.border_index << ", face_valence " << c.face_valence << |
507 | ", edge_level = " << c.edge_level << ", vertex_level = " << c.vertex_level << ", eval_start_index: " << c.eval_start_index << ", ring: " << embree_endl; |
508 | |
509 | for (unsigned int i=0; i<min(c.edge_valence,(unsigned int)MAX_RING_FACE_VALENCE); i++) { |
510 | o << i << " -> " << c.ring[i]; |
511 | if (i % 2 == 0) o << " crease = " << c.crease_weight[i/2]; |
512 | o << embree_endl; |
513 | } |
514 | return o; |
515 | } |
516 | }; |
517 | |
518 | typedef CatmullClark1RingT<Vec3fa,Vec3fa_t> CatmullClark1Ring3fa; |
519 | |
520 | template<typename Vertex, typename Vertex_t = Vertex> |
521 | struct __aligned(64) GeneralCatmullClark1RingT |
522 | { |
523 | ALIGNED_STRUCT_(64); |
524 | |
525 | typedef CatmullClark1RingT<Vertex,Vertex_t> CatmullClark1Ring; |
526 | |
527 | struct Face |
528 | { |
529 | __forceinline Face() {} |
530 | __forceinline Face (int size, float crease_weight) |
531 | : size(size), crease_weight(crease_weight) {} |
532 | |
533 | // FIXME: add member that returns total number of vertices |
534 | |
535 | int size; // number of vertices-2 of nth face in ring |
536 | float crease_weight; |
537 | }; |
538 | |
539 | Vertex vtx; |
540 | DynamicStackArray<Vertex,32,MAX_RING_EDGE_VALENCE> ring; |
541 | DynamicStackArray<Face,16,MAX_RING_FACE_VALENCE> faces; |
542 | unsigned int face_valence; |
543 | unsigned int edge_valence; |
544 | int border_face; |
545 | float vertex_crease_weight; |
546 | float vertex_level; //!< maximum level of adjacent edges |
547 | float edge_level; // level of first edge |
548 | bool only_quads; // true if all faces are quads |
549 | unsigned int eval_start_face_index; |
550 | unsigned int eval_start_vertex_index; |
551 | unsigned int eval_unique_identifier; |
552 | |
553 | public: |
554 | GeneralCatmullClark1RingT() |
555 | : eval_start_face_index(0), eval_start_vertex_index(0), eval_unique_identifier(0) {} |
556 | |
557 | __forceinline bool isRegular() const |
558 | { |
559 | if (border_face == -1 && face_valence == 4) return true; |
560 | return false; |
561 | } |
562 | |
563 | __forceinline bool has_last_face() const { |
564 | return border_face != (int)face_valence-1; |
565 | } |
566 | |
567 | __forceinline bool has_second_face() const { |
568 | return (border_face == -1) || (border_face >= 2); |
569 | } |
570 | |
571 | bool hasValidPositions() const |
572 | { |
573 | for (size_t i=0; i<edge_valence; i++) { |
574 | if (!isvalid(ring[i])) |
575 | return false; |
576 | } |
577 | return true; |
578 | } |
579 | |
580 | __forceinline void init(const HalfEdge* const h, const char* vertices, size_t stride) |
581 | { |
582 | only_quads = true; |
583 | border_face = -1; |
584 | vtx = Vertex_t::loadu(vertices+h->getStartVertexIndex()*stride); |
585 | vertex_crease_weight = h->vertex_crease_weight; |
586 | HalfEdge* p = (HalfEdge*) h; |
587 | |
588 | unsigned int e=0, f=0; |
589 | unsigned min_vertex_index = (unsigned)-1; |
590 | unsigned min_vertex_index_face = (unsigned)-1; |
591 | unsigned min_vertex_index_vertex = (unsigned)-1; |
592 | edge_level = p->edge_level; |
593 | vertex_level = 0.0f; |
594 | do |
595 | { |
596 | HalfEdge* p_prev = p->prev(); |
597 | HalfEdge* p_next = p->next(); |
598 | const float crease_weight = p->edge_crease_weight; |
599 | assert(p->hasOpposite() || p->edge_crease_weight == float(inf)); |
600 | vertex_level = max(a: vertex_level,b: p->edge_level); |
601 | |
602 | /* find minimum start vertex */ |
603 | unsigned vertex_index = p_next->getStartVertexIndex(); |
604 | if (vertex_index < min_vertex_index) { min_vertex_index = vertex_index; min_vertex_index_face = f; min_vertex_index_vertex = e; } |
605 | |
606 | /* store first N-2 vertices of face */ |
607 | unsigned int vn = 0; |
608 | for (p = p_next; p!=p_prev; p=p->next()) { |
609 | ring[e++] = Vertex_t::loadu(vertices+p->getStartVertexIndex()*stride); |
610 | vn++; |
611 | } |
612 | faces[f++] = Face(vn,crease_weight); |
613 | only_quads &= (vn == 2); |
614 | |
615 | /* continue with next face */ |
616 | if (likely(p->hasOpposite())) |
617 | p = p->opposite(); |
618 | |
619 | /* if there is no opposite go the long way to the other side of the border */ |
620 | else |
621 | { |
622 | /* find minimum start vertex */ |
623 | unsigned vertex_index = p->getStartVertexIndex(); |
624 | if (vertex_index < min_vertex_index) { min_vertex_index = vertex_index; min_vertex_index_face = f; min_vertex_index_vertex = e; } |
625 | |
626 | /*! mark first border edge and store dummy vertex for face between the two border edges */ |
627 | border_face = f; |
628 | faces[f++] = Face(2,inf); |
629 | ring[e++] = Vertex_t::loadu(vertices+p->getStartVertexIndex()*stride); |
630 | ring[e++] = vtx; // dummy vertex |
631 | |
632 | /*! goto other side of border */ |
633 | p = (HalfEdge*) h; |
634 | while (p->hasOpposite()) |
635 | p = p->opposite()->next(); |
636 | } |
637 | |
638 | } while (p != h); |
639 | |
640 | edge_valence = e; |
641 | face_valence = f; |
642 | eval_unique_identifier = min_vertex_index; |
643 | eval_start_face_index = min_vertex_index_face; |
644 | eval_start_vertex_index = min_vertex_index_vertex; |
645 | |
646 | assert( hasValidPositions() ); |
647 | } |
648 | |
649 | __forceinline void subdivide(CatmullClark1Ring& dest) const |
650 | { |
651 | dest.edge_level = 0.5f*edge_level; |
652 | dest.vertex_level = 0.5f*vertex_level; |
653 | dest.face_valence = face_valence; |
654 | dest.edge_valence = 2*face_valence; |
655 | dest.border_index = border_face == -1 ? -1 : 2*border_face; // FIXME: |
656 | dest.vertex_crease_weight = max(a: 0.0f,b: vertex_crease_weight-1.0f); |
657 | dest.eval_start_index = eval_start_face_index; |
658 | dest.eval_unique_identifier = eval_unique_identifier; |
659 | assert(dest.face_valence <= MAX_RING_FACE_VALENCE); |
660 | |
661 | /* calculate face points */ |
662 | Vertex_t S = Vertex_t(0.0f); |
663 | for (size_t face=0, v=eval_start_vertex_index; face<face_valence; face++) { |
664 | size_t f = (face + eval_start_face_index)%face_valence; |
665 | |
666 | Vertex_t F = vtx; |
667 | for (size_t k=v; k<=v+faces[f].size; k++) F += ring[k%edge_valence]; // FIXME: optimize |
668 | S += dest.ring[2*f+1] = F/float(faces[f].size+2); |
669 | v+=faces[f].size; |
670 | v%=edge_valence; |
671 | } |
672 | |
673 | /* calculate new edge points */ |
674 | size_t num_creases = 0; |
675 | array_t<size_t,MAX_RING_FACE_VALENCE> crease_id; |
676 | Vertex_t C = Vertex_t(0.0f); |
677 | for (size_t face=0, j=eval_start_vertex_index; face<face_valence; face++) |
678 | { |
679 | size_t i = (face + eval_start_face_index)%face_valence; |
680 | |
681 | const Vertex_t v = vtx + ring[j]; |
682 | Vertex_t f = dest.ring[2*i+1]; |
683 | if (i == 0) f += dest.ring[dest.edge_valence-1]; |
684 | else f += dest.ring[2*i-1]; |
685 | S += ring[j]; |
686 | dest.crease_weight[i] = max(faces[i].crease_weight-1.0f,0.0f); |
687 | |
688 | /* fast path for regular edge points */ |
689 | if (likely(faces[i].crease_weight <= 0.0f)) { |
690 | dest.ring[2*i] = (v+f) * 0.25f; |
691 | } |
692 | |
693 | /* slower path for hard edge rule */ |
694 | else { |
695 | C += ring[j]; crease_id[num_creases++] = i; |
696 | dest.ring[2*i] = v*0.5f; |
697 | |
698 | /* even slower path for blended edge rule */ |
699 | if (unlikely(faces[i].crease_weight < 1.0f)) { |
700 | dest.ring[2*i] = lerp((v+f)*0.25f,v*0.5f,faces[i].crease_weight); |
701 | } |
702 | } |
703 | j+=faces[i].size; |
704 | j%=edge_valence; |
705 | } |
706 | |
707 | /* compute new vertex using smooth rule */ |
708 | const float inv_face_valence = 1.0f / (float)face_valence; |
709 | const Vertex_t v_smooth = (Vertex_t) madd(inv_face_valence,S,(float(face_valence)-2.0f)*vtx)*inv_face_valence; |
710 | dest.vtx = v_smooth; |
711 | |
712 | /* compute new vertex using vertex_crease_weight rule */ |
713 | if (unlikely(vertex_crease_weight > 0.0f)) |
714 | { |
715 | if (vertex_crease_weight >= 1.0f) { |
716 | dest.vtx = vtx; |
717 | } else { |
718 | dest.vtx = lerp(vtx,v_smooth,vertex_crease_weight); |
719 | } |
720 | return; |
721 | } |
722 | |
723 | if (likely(num_creases <= 1)) |
724 | return; |
725 | |
726 | /* compute new vertex using crease rule */ |
727 | if (likely(num_creases == 2)) { |
728 | const Vertex_t v_sharp = (Vertex_t)(C + 6.0f * vtx) * (1.0f / 8.0f); |
729 | const float crease_weight0 = faces[crease_id[0]].crease_weight; |
730 | const float crease_weight1 = faces[crease_id[1]].crease_weight; |
731 | dest.vtx = v_sharp; |
732 | dest.crease_weight[crease_id[0]] = max(a: 0.25f*(3.0f*crease_weight0 + crease_weight1)-1.0f,b: 0.0f); |
733 | dest.crease_weight[crease_id[1]] = max(a: 0.25f*(3.0f*crease_weight1 + crease_weight0)-1.0f,b: 0.0f); |
734 | const float v_blend = 0.5f*(crease_weight0+crease_weight1); |
735 | if (unlikely(v_blend < 1.0f)) { |
736 | dest.vtx = lerp(v_sharp,v_smooth,v_blend); |
737 | } |
738 | } |
739 | |
740 | /* compute new vertex using corner rule */ |
741 | else { |
742 | dest.vtx = vtx; |
743 | } |
744 | } |
745 | |
746 | void convert(CatmullClark1Ring& dst) const |
747 | { |
748 | dst.edge_level = edge_level; |
749 | dst.vertex_level = vertex_level; |
750 | dst.vtx = vtx; |
751 | dst.face_valence = face_valence; |
752 | dst.edge_valence = 2*face_valence; |
753 | dst.border_index = border_face == -1 ? -1 : 2*border_face; |
754 | for (size_t i=0; i<face_valence; i++) |
755 | dst.crease_weight[i] = faces[i].crease_weight; |
756 | dst.vertex_crease_weight = vertex_crease_weight; |
757 | for (size_t i=0; i<edge_valence; i++) dst.ring[i] = ring[i]; |
758 | |
759 | dst.eval_start_index = eval_start_face_index; |
760 | dst.eval_unique_identifier = eval_unique_identifier; |
761 | |
762 | assert( dst.hasValidPositions() ); |
763 | } |
764 | |
765 | |
766 | /* gets limit tangent in the direction of egde vtx -> ring[0] */ |
767 | __forceinline Vertex getLimitTangent() const |
768 | { |
769 | CatmullClark1Ring cc_vtx; |
770 | |
771 | /* fast path for quad only rings */ |
772 | if (only_quads) |
773 | { |
774 | convert(dst&: cc_vtx); |
775 | return cc_vtx.getLimitTangent(); |
776 | } |
777 | |
778 | subdivide(dest&: cc_vtx); |
779 | return 2.0f * cc_vtx.getLimitTangent(); |
780 | } |
781 | |
782 | /* gets limit tangent in the direction of egde vtx -> ring[edge_valence-2] */ |
783 | __forceinline Vertex getSecondLimitTangent() const |
784 | { |
785 | CatmullClark1Ring cc_vtx; |
786 | |
787 | /* fast path for quad only rings */ |
788 | if (only_quads) |
789 | { |
790 | convert(dst&: cc_vtx); |
791 | return cc_vtx.getSecondLimitTangent(); |
792 | } |
793 | |
794 | subdivide(dest&: cc_vtx); |
795 | return 2.0f * cc_vtx.getSecondLimitTangent(); |
796 | } |
797 | |
798 | |
799 | /* gets limit vertex */ |
800 | __forceinline Vertex getLimitVertex() const |
801 | { |
802 | CatmullClark1Ring cc_vtx; |
803 | |
804 | /* fast path for quad only rings */ |
805 | if (only_quads) |
806 | convert(dst&: cc_vtx); |
807 | else |
808 | subdivide(dest&: cc_vtx); |
809 | return cc_vtx.getLimitVertex(); |
810 | } |
811 | |
812 | friend __forceinline embree_ostream operator<<(embree_ostream o, const GeneralCatmullClark1RingT &c) |
813 | { |
814 | o << "vtx " << c.vtx << " size = " << c.edge_valence << ", border_face = " << c.border_face << ", " << " face_valence = " << c.face_valence << |
815 | ", edge_level = " << c.edge_level << ", vertex_level = " << c.vertex_level << ", ring: " << embree_endl; |
816 | for (size_t v=0, f=0; f<c.face_valence; v+=c.faces[f++].size) { |
817 | for (size_t i=v; i<v+c.faces[f].size; i++) { |
818 | o << i << " -> " << c.ring[i]; |
819 | if (i == v) o << " crease = " << c.faces[f].crease_weight; |
820 | o << embree_endl; |
821 | } |
822 | } |
823 | return o; |
824 | } |
825 | }; |
826 | } |
827 | |