1 | // Copyright 2009-2021 Intel Corporation |
2 | // SPDX-License-Identifier: Apache-2.0 |
3 | |
4 | #pragma once |
5 | |
6 | #include "geometry.h" |
7 | #include "buffer.h" |
8 | #include "../subdiv/half_edge.h" |
9 | #include "../subdiv/tessellation_cache.h" |
10 | #include "../subdiv/catmullclark_coefficients.h" |
11 | #include "../subdiv/patch.h" |
12 | #include "../../common/algorithms/parallel_map.h" |
13 | #include "../../common/algorithms/parallel_set.h" |
14 | |
15 | namespace embree |
16 | { |
17 | class SubdivMesh : public Geometry |
18 | { |
19 | ALIGNED_CLASS_(16); |
20 | public: |
21 | |
22 | typedef HalfEdge::Edge Edge; |
23 | |
24 | /*! type of this geometry */ |
25 | static const Geometry::GTypeMask geom_type = Geometry::MTY_SUBDIV_MESH; |
26 | |
27 | /*! structure used to sort half edges using radix sort by their key */ |
28 | struct KeyHalfEdge |
29 | { |
30 | KeyHalfEdge() {} |
31 | |
32 | KeyHalfEdge (uint64_t key, HalfEdge* edge) |
33 | : key(key), edge(edge) {} |
34 | |
35 | __forceinline operator uint64_t() const { |
36 | return key; |
37 | } |
38 | |
39 | friend __forceinline bool operator<(const KeyHalfEdge& e0, const KeyHalfEdge& e1) { |
40 | return e0.key < e1.key; |
41 | } |
42 | |
43 | public: |
44 | uint64_t key; |
45 | HalfEdge* edge; |
46 | }; |
47 | |
48 | public: |
49 | |
50 | /*! subdiv mesh construction */ |
51 | SubdivMesh(Device* device); |
52 | |
53 | public: |
54 | void setMask (unsigned mask); |
55 | void setSubdivisionMode (unsigned int topologyID, RTCSubdivisionMode mode); |
56 | void setVertexAttributeTopology(unsigned int vertexAttribID, unsigned int topologyID); |
57 | void setNumTimeSteps (unsigned int numTimeSteps); |
58 | void setVertexAttributeCount (unsigned int N); |
59 | void setTopologyCount (unsigned int N); |
60 | void setBuffer(RTCBufferType type, unsigned int slot, RTCFormat format, const Ref<Buffer>& buffer, size_t offset, size_t stride, unsigned int num); |
61 | void* getBuffer(RTCBufferType type, unsigned int slot); |
62 | void updateBuffer(RTCBufferType type, unsigned int slot); |
63 | void setTessellationRate(float N); |
64 | bool verify(); |
65 | void commit(); |
66 | void addElementsToCount (GeometryCounts & counts) const; |
67 | void setDisplacementFunction (RTCDisplacementFunctionN func); |
68 | unsigned int getFirstHalfEdge(unsigned int faceID); |
69 | unsigned int getFace(unsigned int edgeID); |
70 | unsigned int getNextHalfEdge(unsigned int edgeID); |
71 | unsigned int getPreviousHalfEdge(unsigned int edgeID); |
72 | unsigned int getOppositeHalfEdge(unsigned int topologyID, unsigned int edgeID); |
73 | |
74 | public: |
75 | |
76 | /*! return the number of faces */ |
77 | size_t numFaces() const { |
78 | return faceVertices.size(); |
79 | } |
80 | |
81 | /*! return the number of edges */ |
82 | size_t numEdges() const { |
83 | return topology[0].vertexIndices.size(); |
84 | } |
85 | |
86 | /*! return the number of vertices */ |
87 | size_t numVertices() const { |
88 | return vertices[0].size(); |
89 | } |
90 | |
91 | /*! calculates the bounds of the i'th subdivision patch at the j'th timestep */ |
92 | __forceinline BBox3fa bounds(size_t i, size_t j = 0) const { |
93 | return topology[0].getHalfEdge(i)->bounds(vertices[j]); |
94 | } |
95 | |
96 | /*! check if the i'th primitive is valid */ |
97 | __forceinline bool valid(size_t i) const { |
98 | return topology[0].valid(i) && !invalidFace(i); |
99 | } |
100 | |
101 | /*! check if the i'th primitive is valid for the j'th time range */ |
102 | __forceinline bool valid(size_t i, size_t j) const { |
103 | return topology[0].valid(i) && !invalidFace(i,j); |
104 | } |
105 | |
106 | /*! prints some statistics */ |
107 | void printStatistics(); |
108 | |
109 | /*! initializes the half edge data structure */ |
110 | void initializeHalfEdgeStructures (); |
111 | |
112 | public: |
113 | |
114 | /*! returns the vertex buffer for some time step */ |
115 | __forceinline const BufferView<Vec3fa>& getVertexBuffer( const size_t t = 0 ) const { |
116 | return vertices[t]; |
117 | } |
118 | |
119 | /* returns tessellation level of edge */ |
120 | __forceinline float getEdgeLevel(const size_t i) const |
121 | { |
122 | if (levels) return clamp(levels[i],1.0f,4096.0f); // FIXME: do we want to limit edge level? |
123 | else return clamp(tessellationRate,1.0f,4096.0f); // FIXME: do we want to limit edge level? |
124 | } |
125 | |
126 | public: |
127 | RTCDisplacementFunctionN displFunc; //!< displacement function |
128 | |
129 | /*! all buffers in this section are provided by the application */ |
130 | public: |
131 | |
132 | /*! the topology contains all data that may differ when |
133 | * interpolating different user data buffers */ |
134 | struct Topology |
135 | { |
136 | public: |
137 | |
138 | /*! Default topology construction */ |
139 | Topology () : halfEdges(nullptr,0) {} |
140 | |
141 | /*! Topology initialization */ |
142 | Topology (SubdivMesh* mesh); |
143 | |
144 | /*! make the class movable */ |
145 | public: |
146 | Topology (Topology&& other) // FIXME: this is only required to workaround compilation issues under Windows |
147 | : mesh(std::move(other.mesh)), |
148 | vertexIndices(std::move(other.vertexIndices)), |
149 | subdiv_mode(std::move(other.subdiv_mode)), |
150 | halfEdges(std::move(other.halfEdges)), |
151 | halfEdges0(std::move(other.halfEdges0)), |
152 | halfEdges1(std::move(other.halfEdges1)) {} |
153 | |
154 | Topology& operator= (Topology&& other) // FIXME: this is only required to workaround compilation issues under Windows |
155 | { |
156 | mesh = std::move(other.mesh); |
157 | vertexIndices = std::move(other.vertexIndices); |
158 | subdiv_mode = std::move(other.subdiv_mode); |
159 | halfEdges = std::move(other.halfEdges); |
160 | halfEdges0 = std::move(other.halfEdges0); |
161 | halfEdges1 = std::move(other.halfEdges1); |
162 | return *this; |
163 | } |
164 | |
165 | public: |
166 | /*! check if the i'th primitive is valid in this topology */ |
167 | __forceinline bool valid(size_t i) const |
168 | { |
169 | if (unlikely(subdiv_mode == RTC_SUBDIVISION_MODE_NO_BOUNDARY)) { |
170 | if (getHalfEdge(i)->faceHasBorder()) return false; |
171 | } |
172 | return true; |
173 | } |
174 | |
175 | /*! updates the interpolation mode for the topology */ |
176 | void setSubdivisionMode (RTCSubdivisionMode mode); |
177 | |
178 | /*! marks all buffers as modified */ |
179 | void update (); |
180 | |
181 | /*! verifies index array */ |
182 | bool verify (size_t numVertices); |
183 | |
184 | /*! initializes the half edge data structure */ |
185 | void initializeHalfEdgeStructures (); |
186 | |
187 | private: |
188 | |
189 | /*! recalculates the half edges */ |
190 | void calculateHalfEdges(); |
191 | |
192 | /*! updates half edges when recalculation is not necessary */ |
193 | void updateHalfEdges(); |
194 | |
195 | /*! user input data */ |
196 | public: |
197 | |
198 | SubdivMesh* mesh; |
199 | |
200 | /*! indices of the vertices composing each face */ |
201 | BufferView<unsigned int> vertexIndices; |
202 | |
203 | /*! subdiv interpolation mode */ |
204 | RTCSubdivisionMode subdiv_mode; |
205 | |
206 | /*! generated data */ |
207 | public: |
208 | |
209 | /*! returns the start half edge for face f */ |
210 | __forceinline const HalfEdge* getHalfEdge ( const size_t f ) const { |
211 | return &halfEdges[mesh->faceStartEdge[f]]; |
212 | } |
213 | |
214 | /*! Half edge structure, generated by initHalfEdgeStructures */ |
215 | mvector<HalfEdge> halfEdges; |
216 | |
217 | /*! the following data is only required during construction of the |
218 | * half edge structure and can be cleared for static scenes */ |
219 | private: |
220 | |
221 | /*! two arrays used to sort the half edges */ |
222 | std::vector<KeyHalfEdge> halfEdges0; |
223 | std::vector<KeyHalfEdge> halfEdges1; |
224 | }; |
225 | |
226 | /*! returns the start half edge for topology t and face f */ |
227 | __forceinline const HalfEdge* getHalfEdge ( const size_t t , const size_t f ) const { |
228 | return topology[t].getHalfEdge(f); |
229 | } |
230 | |
231 | /*! buffer containing the number of vertices for each face */ |
232 | BufferView<unsigned int> faceVertices; |
233 | |
234 | /*! array of topologies */ |
235 | vector<Topology> topology; |
236 | |
237 | /*! vertex buffer (one buffer for each time step) */ |
238 | vector<BufferView<Vec3fa>> vertices; |
239 | |
240 | /*! user data buffers */ |
241 | vector<RawBufferView> vertexAttribs; |
242 | |
243 | /*! edge crease buffer containing edges (pairs of vertices) that carry edge crease weights */ |
244 | BufferView<Edge> edge_creases; |
245 | |
246 | /*! edge crease weights for each edge of the edge_creases buffer */ |
247 | BufferView<float> edge_crease_weights; |
248 | |
249 | /*! vertex crease buffer containing all vertices that carry vertex crease weights */ |
250 | BufferView<unsigned int> vertex_creases; |
251 | |
252 | /*! vertex crease weights for each vertex of the vertex_creases buffer */ |
253 | BufferView<float> vertex_crease_weights; |
254 | |
255 | /*! subdivision level for each half edge of the vertexIndices buffer */ |
256 | BufferView<float> levels; |
257 | float tessellationRate; // constant rate that is used when levels is not set |
258 | |
259 | /*! buffer that marks specific faces as holes */ |
260 | BufferView<unsigned> holes; |
261 | |
262 | /*! all data in this section is generated by initializeHalfEdgeStructures function */ |
263 | private: |
264 | |
265 | /*! number of half edges used by faces */ |
266 | size_t numHalfEdges; |
267 | |
268 | /*! fast lookup table to find the first half edge for some face */ |
269 | mvector<uint32_t> faceStartEdge; |
270 | |
271 | /*! fast lookup table to find the face for some half edge */ |
272 | mvector<uint32_t> halfEdgeFace; |
273 | |
274 | /*! set with all holes */ |
275 | parallel_set<uint32_t> holeSet; |
276 | |
277 | /*! fast lookup table to detect invalid faces */ |
278 | mvector<char> invalid_face; |
279 | |
280 | /*! test if face i is invalid in timestep j */ |
281 | __forceinline char& invalidFace(size_t i, size_t j = 0) { return invalid_face[i*numTimeSteps+j]; } |
282 | __forceinline const char& invalidFace(size_t i, size_t j = 0) const { return invalid_face[i*numTimeSteps+j]; } |
283 | |
284 | /*! interpolation cache */ |
285 | public: |
286 | static __forceinline size_t numInterpolationSlots4(size_t stride) { return (stride+15)/16; } |
287 | static __forceinline size_t numInterpolationSlots8(size_t stride) { return (stride+31)/32; } |
288 | static __forceinline size_t interpolationSlot(size_t prim, size_t slot, size_t stride) { |
289 | const size_t slots = numInterpolationSlots4(stride); |
290 | assert(slot < slots); |
291 | return slots*prim+slot; |
292 | } |
293 | std::vector<std::vector<SharedLazyTessellationCache::CacheEntry>> vertex_buffer_tags; |
294 | std::vector<std::vector<SharedLazyTessellationCache::CacheEntry>> vertex_attrib_buffer_tags; |
295 | std::vector<Patch3fa::Ref> patch_eval_trees; |
296 | |
297 | /*! the following data is only required during construction of the |
298 | * half edge structure and can be cleared for static scenes */ |
299 | private: |
300 | |
301 | /*! map with all vertex creases */ |
302 | parallel_map<uint32_t,float> vertexCreaseMap; |
303 | |
304 | /*! map with all edge creases */ |
305 | parallel_map<uint64_t,float> edgeCreaseMap; |
306 | |
307 | protected: |
308 | |
309 | /*! counts number of geometry commits */ |
310 | size_t commitCounter; |
311 | }; |
312 | |
313 | namespace isa |
314 | { |
315 | struct SubdivMeshISA : public SubdivMesh |
316 | { |
317 | SubdivMeshISA (Device* device) |
318 | : SubdivMesh(device) {} |
319 | |
320 | void interpolate(const RTCInterpolateArguments* const args); |
321 | void interpolateN(const RTCInterpolateNArguments* const args); |
322 | }; |
323 | } |
324 | |
325 | DECLARE_ISA_FUNCTION(SubdivMesh*, createSubdivMesh, Device*); |
326 | }; |
327 | |