| 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 | |
| 9 | namespace embree |
| 10 | { |
| 11 | /*! Quad Mesh */ |
| 12 | struct QuadMesh : public Geometry |
| 13 | { |
| 14 | /*! type of this geometry */ |
| 15 | static const Geometry::GTypeMask geom_type = Geometry::MTY_QUAD_MESH; |
| 16 | |
| 17 | /*! triangle indices */ |
| 18 | struct Quad |
| 19 | { |
| 20 | uint32_t v[4]; |
| 21 | |
| 22 | /*! outputs triangle indices */ |
| 23 | __forceinline friend embree_ostream operator<<(embree_ostream cout, const Quad& q) { |
| 24 | return cout << "Quad {" << q.v[0] << ", " << q.v[1] << ", " << q.v[2] << ", " << q.v[3] << " }" ; |
| 25 | } |
| 26 | }; |
| 27 | |
| 28 | public: |
| 29 | |
| 30 | /*! quad mesh construction */ |
| 31 | QuadMesh (Device* device); |
| 32 | |
| 33 | /* geometry interface */ |
| 34 | public: |
| 35 | void setMask(unsigned mask); |
| 36 | void setNumTimeSteps (unsigned int numTimeSteps); |
| 37 | void setVertexAttributeCount (unsigned int N); |
| 38 | void setBuffer(RTCBufferType type, unsigned int slot, RTCFormat format, const Ref<Buffer>& buffer, size_t offset, size_t stride, unsigned int num); |
| 39 | void* getBuffer(RTCBufferType type, unsigned int slot); |
| 40 | void updateBuffer(RTCBufferType type, unsigned int slot); |
| 41 | void commit(); |
| 42 | bool verify(); |
| 43 | void interpolate(const RTCInterpolateArguments* const args); |
| 44 | void addElementsToCount (GeometryCounts & counts) const; |
| 45 | |
| 46 | template<int N> |
| 47 | void interpolate_impl(const RTCInterpolateArguments* const args) |
| 48 | { |
| 49 | unsigned int primID = args->primID; |
| 50 | float u = args->u; |
| 51 | float v = args->v; |
| 52 | RTCBufferType bufferType = args->bufferType; |
| 53 | unsigned int bufferSlot = args->bufferSlot; |
| 54 | float* P = args->P; |
| 55 | float* dPdu = args->dPdu; |
| 56 | float* dPdv = args->dPdv; |
| 57 | float* ddPdudu = args->ddPdudu; |
| 58 | float* ddPdvdv = args->ddPdvdv; |
| 59 | float* ddPdudv = args->ddPdudv; |
| 60 | unsigned int valueCount = args->valueCount; |
| 61 | |
| 62 | /* calculate base pointer and stride */ |
| 63 | assert((bufferType == RTC_BUFFER_TYPE_VERTEX && bufferSlot < numTimeSteps) || |
| 64 | (bufferType == RTC_BUFFER_TYPE_VERTEX_ATTRIBUTE && bufferSlot <= vertexAttribs.size())); |
| 65 | const char* src = nullptr; |
| 66 | size_t stride = 0; |
| 67 | if (bufferType == RTC_BUFFER_TYPE_VERTEX_ATTRIBUTE) { |
| 68 | src = vertexAttribs[bufferSlot].getPtr(); |
| 69 | stride = vertexAttribs[bufferSlot].getStride(); |
| 70 | } else { |
| 71 | src = vertices[bufferSlot].getPtr(); |
| 72 | stride = vertices[bufferSlot].getStride(); |
| 73 | } |
| 74 | |
| 75 | for (unsigned int i=0; i<valueCount; i+=N) |
| 76 | { |
| 77 | const vbool<N> valid = vint<N>((int)i)+vint<N>(step) < vint<N>(int(valueCount)); |
| 78 | const size_t ofs = i*sizeof(float); |
| 79 | const Quad& tri = quad(i: primID); |
| 80 | const vfloat<N> p0 = mem<vfloat<N>>::loadu(valid,(float*)&src[tri.v[0]*stride+ofs]); |
| 81 | const vfloat<N> p1 = mem<vfloat<N>>::loadu(valid,(float*)&src[tri.v[1]*stride+ofs]); |
| 82 | const vfloat<N> p2 = mem<vfloat<N>>::loadu(valid,(float*)&src[tri.v[2]*stride+ofs]); |
| 83 | const vfloat<N> p3 = mem<vfloat<N>>::loadu(valid,(float*)&src[tri.v[3]*stride+ofs]); |
| 84 | const vbool<N> left = u+v <= 1.0f; |
| 85 | const vfloat<N> Q0 = select(left,p0,p2); |
| 86 | const vfloat<N> Q1 = select(left,p1,p3); |
| 87 | const vfloat<N> Q2 = select(left,p3,p1); |
| 88 | const vfloat<N> U = select(left,u,vfloat<N>(1.0f)-u); |
| 89 | const vfloat<N> V = select(left,v,vfloat<N>(1.0f)-v); |
| 90 | const vfloat<N> W = 1.0f-U-V; |
| 91 | if (P) { |
| 92 | mem<vfloat<N>>::storeu(valid,P+i,madd(W,Q0,madd(U,Q1,V*Q2))); |
| 93 | } |
| 94 | if (dPdu) { |
| 95 | assert(dPdu); mem<vfloat<N>>::storeu(valid,dPdu+i,select(left,Q1-Q0,Q0-Q1)); |
| 96 | assert(dPdv); mem<vfloat<N>>::storeu(valid,dPdv+i,select(left,Q2-Q0,Q0-Q2)); |
| 97 | } |
| 98 | if (ddPdudu) { |
| 99 | assert(ddPdudu); mem<vfloat<N>>::storeu(valid,ddPdudu+i,vfloat<N>(zero)); |
| 100 | assert(ddPdvdv); mem<vfloat<N>>::storeu(valid,ddPdvdv+i,vfloat<N>(zero)); |
| 101 | assert(ddPdudv); mem<vfloat<N>>::storeu(valid,ddPdudv+i,vfloat<N>(zero)); |
| 102 | } |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | public: |
| 107 | |
| 108 | /*! returns number of vertices */ |
| 109 | __forceinline size_t numVertices() const { |
| 110 | return vertices[0].size(); |
| 111 | } |
| 112 | |
| 113 | /*! returns i'th quad */ |
| 114 | __forceinline const Quad& quad(size_t i) const { |
| 115 | return quads[i]; |
| 116 | } |
| 117 | |
| 118 | /*! returns i'th vertex of itime'th timestep */ |
| 119 | __forceinline const Vec3fa vertex(size_t i) const { |
| 120 | return vertices0[i]; |
| 121 | } |
| 122 | |
| 123 | /*! returns i'th vertex of itime'th timestep */ |
| 124 | __forceinline const char* vertexPtr(size_t i) const { |
| 125 | return vertices0.getPtr(i); |
| 126 | } |
| 127 | |
| 128 | /*! returns i'th vertex of itime'th timestep */ |
| 129 | __forceinline const Vec3fa vertex(size_t i, size_t itime) const { |
| 130 | return vertices[itime][i]; |
| 131 | } |
| 132 | |
| 133 | /*! returns i'th vertex of itime'th timestep */ |
| 134 | __forceinline const char* vertexPtr(size_t i, size_t itime) const { |
| 135 | return vertices[itime].getPtr(i); |
| 136 | } |
| 137 | |
| 138 | /*! calculates the bounds of the i'th quad */ |
| 139 | __forceinline BBox3fa bounds(size_t i) const |
| 140 | { |
| 141 | const Quad& q = quad(i); |
| 142 | const Vec3fa v0 = vertex(i: q.v[0]); |
| 143 | const Vec3fa v1 = vertex(i: q.v[1]); |
| 144 | const Vec3fa v2 = vertex(i: q.v[2]); |
| 145 | const Vec3fa v3 = vertex(i: q.v[3]); |
| 146 | return BBox3fa(min(a: v0,b: v1,c: v2,d: v3),max(a: v0,b: v1,c: v2,d: v3)); |
| 147 | } |
| 148 | |
| 149 | /*! calculates the bounds of the i'th quad at the itime'th timestep */ |
| 150 | __forceinline BBox3fa bounds(size_t i, size_t itime) const |
| 151 | { |
| 152 | const Quad& q = quad(i); |
| 153 | const Vec3fa v0 = vertex(i: q.v[0],itime); |
| 154 | const Vec3fa v1 = vertex(i: q.v[1],itime); |
| 155 | const Vec3fa v2 = vertex(i: q.v[2],itime); |
| 156 | const Vec3fa v3 = vertex(i: q.v[3],itime); |
| 157 | return BBox3fa(min(a: v0,b: v1,c: v2,d: v3),max(a: v0,b: v1,c: v2,d: v3)); |
| 158 | } |
| 159 | |
| 160 | /*! check if the i'th primitive is valid at the itime'th timestep */ |
| 161 | __forceinline bool valid(size_t i, size_t itime) const { |
| 162 | return valid(i, itime_range: make_range(begin: itime, end: itime)); |
| 163 | } |
| 164 | |
| 165 | /*! check if the i'th primitive is valid between the specified time range */ |
| 166 | __forceinline bool valid(size_t i, const range<size_t>& itime_range) const |
| 167 | { |
| 168 | const Quad& q = quad(i); |
| 169 | if (unlikely(q.v[0] >= numVertices())) return false; |
| 170 | if (unlikely(q.v[1] >= numVertices())) return false; |
| 171 | if (unlikely(q.v[2] >= numVertices())) return false; |
| 172 | if (unlikely(q.v[3] >= numVertices())) return false; |
| 173 | |
| 174 | for (size_t itime = itime_range.begin(); itime <= itime_range.end(); itime++) |
| 175 | { |
| 176 | if (!isvalid(v: vertex(i: q.v[0],itime))) return false; |
| 177 | if (!isvalid(v: vertex(i: q.v[1],itime))) return false; |
| 178 | if (!isvalid(v: vertex(i: q.v[2],itime))) return false; |
| 179 | if (!isvalid(v: vertex(i: q.v[3],itime))) return false; |
| 180 | } |
| 181 | |
| 182 | return true; |
| 183 | } |
| 184 | |
| 185 | /*! calculates the linear bounds of the i'th quad at the itimeGlobal'th time segment */ |
| 186 | __forceinline LBBox3fa linearBounds(size_t i, size_t itime) const { |
| 187 | return LBBox3fa(bounds(i,itime: itime+0),bounds(i,itime: itime+1)); |
| 188 | } |
| 189 | |
| 190 | /*! calculates the build bounds of the i'th primitive, if it's valid */ |
| 191 | __forceinline bool buildBounds(size_t i, BBox3fa* bbox = nullptr) const |
| 192 | { |
| 193 | const Quad& q = quad(i); |
| 194 | if (q.v[0] >= numVertices()) return false; |
| 195 | if (q.v[1] >= numVertices()) return false; |
| 196 | if (q.v[2] >= numVertices()) return false; |
| 197 | if (q.v[3] >= numVertices()) return false; |
| 198 | |
| 199 | for (unsigned int t=0; t<numTimeSteps; t++) |
| 200 | { |
| 201 | const Vec3fa v0 = vertex(i: q.v[0],itime: t); |
| 202 | const Vec3fa v1 = vertex(i: q.v[1],itime: t); |
| 203 | const Vec3fa v2 = vertex(i: q.v[2],itime: t); |
| 204 | const Vec3fa v3 = vertex(i: q.v[3],itime: t); |
| 205 | |
| 206 | if (unlikely(!isvalid(v0) || !isvalid(v1) || !isvalid(v2) || !isvalid(v3))) |
| 207 | return false; |
| 208 | } |
| 209 | |
| 210 | if (bbox) |
| 211 | *bbox = bounds(i); |
| 212 | |
| 213 | return true; |
| 214 | } |
| 215 | |
| 216 | /*! calculates the build bounds of the i'th primitive at the itime'th time segment, if it's valid */ |
| 217 | __forceinline bool buildBounds(size_t i, size_t itime, BBox3fa& bbox) const |
| 218 | { |
| 219 | const Quad& q = quad(i); |
| 220 | if (unlikely(q.v[0] >= numVertices())) return false; |
| 221 | if (unlikely(q.v[1] >= numVertices())) return false; |
| 222 | if (unlikely(q.v[2] >= numVertices())) return false; |
| 223 | if (unlikely(q.v[3] >= numVertices())) return false; |
| 224 | |
| 225 | assert(itime+1 < numTimeSteps); |
| 226 | const Vec3fa a0 = vertex(i: q.v[0],itime: itime+0); if (unlikely(!isvalid(a0))) return false; |
| 227 | const Vec3fa a1 = vertex(i: q.v[1],itime: itime+0); if (unlikely(!isvalid(a1))) return false; |
| 228 | const Vec3fa a2 = vertex(i: q.v[2],itime: itime+0); if (unlikely(!isvalid(a2))) return false; |
| 229 | const Vec3fa a3 = vertex(i: q.v[3],itime: itime+0); if (unlikely(!isvalid(a3))) return false; |
| 230 | const Vec3fa b0 = vertex(i: q.v[0],itime: itime+1); if (unlikely(!isvalid(b0))) return false; |
| 231 | const Vec3fa b1 = vertex(i: q.v[1],itime: itime+1); if (unlikely(!isvalid(b1))) return false; |
| 232 | const Vec3fa b2 = vertex(i: q.v[2],itime: itime+1); if (unlikely(!isvalid(b2))) return false; |
| 233 | const Vec3fa b3 = vertex(i: q.v[3],itime: itime+1); if (unlikely(!isvalid(b3))) return false; |
| 234 | |
| 235 | /* use bounds of first time step in builder */ |
| 236 | bbox = BBox3fa(min(a: a0,b: a1,c: a2,d: a3),max(a: a0,b: a1,c: a2,d: a3)); |
| 237 | return true; |
| 238 | } |
| 239 | |
| 240 | /*! calculates the linear bounds of the i'th primitive for the specified time range */ |
| 241 | __forceinline LBBox3fa linearBounds(size_t primID, const BBox1f& dt) const { |
| 242 | return LBBox3fa([&] (size_t itime) { return bounds(i: primID, itime); }, dt, time_range, fnumTimeSegments); |
| 243 | } |
| 244 | |
| 245 | /*! calculates the linear bounds of the i'th primitive for the specified time range */ |
| 246 | __forceinline bool linearBounds(size_t i, const BBox1f& dt, LBBox3fa& bbox) const |
| 247 | { |
| 248 | if (!valid(i, itime_range: timeSegmentRange(range: dt))) return false; |
| 249 | bbox = linearBounds(primID: i, dt); |
| 250 | return true; |
| 251 | } |
| 252 | |
| 253 | /*! get fast access to first vertex buffer */ |
| 254 | __forceinline float * getCompactVertexArray () const { |
| 255 | return (float*) vertices0.getPtr(); |
| 256 | } |
| 257 | |
| 258 | /* gets version info of topology */ |
| 259 | unsigned int getTopologyVersion() const { |
| 260 | return quads.modCounter; |
| 261 | } |
| 262 | |
| 263 | /* returns true if topology changed */ |
| 264 | bool topologyChanged(unsigned int otherVersion) const { |
| 265 | return quads.isModified(otherModCounter: otherVersion); // || numPrimitivesChanged; |
| 266 | } |
| 267 | |
| 268 | /* returns the projected area */ |
| 269 | __forceinline float projectedPrimitiveArea(const size_t i) const { |
| 270 | const Quad& q = quad(i); |
| 271 | const Vec3fa v0 = vertex(i: q.v[0]); |
| 272 | const Vec3fa v1 = vertex(i: q.v[1]); |
| 273 | const Vec3fa v2 = vertex(i: q.v[2]); |
| 274 | const Vec3fa v3 = vertex(i: q.v[3]); |
| 275 | return areaProjectedTriangle(v0,v1,v2: v3) + |
| 276 | areaProjectedTriangle(v0: v1,v1: v2,v2: v3); |
| 277 | } |
| 278 | |
| 279 | public: |
| 280 | BufferView<Quad> quads; //!< array of quads |
| 281 | BufferView<Vec3fa> vertices0; //!< fast access to first vertex buffer |
| 282 | vector<BufferView<Vec3fa>> vertices; //!< vertex array for each timestep |
| 283 | vector<BufferView<char>> vertexAttribs; //!< vertex attribute buffers |
| 284 | }; |
| 285 | |
| 286 | namespace isa |
| 287 | { |
| 288 | struct QuadMeshISA : public QuadMesh |
| 289 | { |
| 290 | QuadMeshISA (Device* device) |
| 291 | : QuadMesh(device) {} |
| 292 | |
| 293 | PrimInfo createPrimRefArray(mvector<PrimRef>& prims, const range<size_t>& r, size_t k, unsigned int geomID) const |
| 294 | { |
| 295 | PrimInfo pinfo(empty); |
| 296 | for (size_t j=r.begin(); j<r.end(); j++) |
| 297 | { |
| 298 | BBox3fa bounds = empty; |
| 299 | if (!buildBounds(i: j,bbox: &bounds)) continue; |
| 300 | const PrimRef prim(bounds,geomID,unsigned(j)); |
| 301 | pinfo.add_center2(prim); |
| 302 | prims[k++] = prim; |
| 303 | } |
| 304 | return pinfo; |
| 305 | } |
| 306 | |
| 307 | PrimInfo createPrimRefArrayMB(mvector<PrimRef>& prims, size_t itime, const range<size_t>& r, size_t k, unsigned int geomID) const |
| 308 | { |
| 309 | PrimInfo pinfo(empty); |
| 310 | for (size_t j=r.begin(); j<r.end(); j++) |
| 311 | { |
| 312 | BBox3fa bounds = empty; |
| 313 | if (!buildBounds(i: j,itime,bbox&: bounds)) continue; |
| 314 | const PrimRef prim(bounds,geomID,unsigned(j)); |
| 315 | pinfo.add_center2(prim); |
| 316 | prims[k++] = prim; |
| 317 | } |
| 318 | return pinfo; |
| 319 | } |
| 320 | |
| 321 | PrimInfoMB createPrimRefMBArray(mvector<PrimRefMB>& prims, const BBox1f& t0t1, const range<size_t>& r, size_t k, unsigned int geomID) const |
| 322 | { |
| 323 | PrimInfoMB pinfo(empty); |
| 324 | for (size_t j=r.begin(); j<r.end(); j++) |
| 325 | { |
| 326 | if (!valid(i: j, itime_range: timeSegmentRange(range: t0t1))) continue; |
| 327 | const PrimRefMB prim(linearBounds(primID: j,dt: t0t1),this->numTimeSegments(),this->time_range,this->numTimeSegments(),geomID,unsigned(j)); |
| 328 | pinfo.add_primref(prim); |
| 329 | prims[k++] = prim; |
| 330 | } |
| 331 | return pinfo; |
| 332 | } |
| 333 | }; |
| 334 | } |
| 335 | |
| 336 | DECLARE_ISA_FUNCTION(QuadMesh*, createQuadMesh, Device*); |
| 337 | } |
| 338 | |