| 1 | // Copyright 2009-2021 Intel Corporation |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| 4 | #pragma once |
| 5 | |
| 6 | #include "subgrid.h" |
| 7 | #include "quad_intersector_moeller.h" |
| 8 | |
| 9 | namespace embree |
| 10 | { |
| 11 | namespace isa |
| 12 | { |
| 13 | |
| 14 | /* ----------------------------- */ |
| 15 | /* -- single ray intersectors -- */ |
| 16 | /* ----------------------------- */ |
| 17 | |
| 18 | template<int M> |
| 19 | __forceinline void interpolateUV(MoellerTrumboreHitM<M,UVIdentity<M>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const vint<M> &stepX, const vint<M> &stepY) |
| 20 | { |
| 21 | /* correct U,V interpolation across the entire grid */ |
| 22 | const vint<M> sx((int)subgrid.x()); |
| 23 | const vint<M> sy((int)subgrid.y()); |
| 24 | const vint<M> sxM(sx + stepX); |
| 25 | const vint<M> syM(sy + stepY); |
| 26 | const float inv_resX = rcp((float)((int)g.resX-1)); |
| 27 | const float inv_resY = rcp((float)((int)g.resY-1)); |
| 28 | hit.U = (hit.U + (vfloat<M>)sxM * hit.absDen) * inv_resX; |
| 29 | hit.V = (hit.V + (vfloat<M>)syM * hit.absDen) * inv_resY; |
| 30 | } |
| 31 | |
| 32 | template<int M, bool filter> |
| 33 | struct SubGridQuadMIntersector1MoellerTrumbore; |
| 34 | |
| 35 | template<int M, bool filter> |
| 36 | struct SubGridQuadMIntersector1MoellerTrumbore |
| 37 | { |
| 38 | __forceinline SubGridQuadMIntersector1MoellerTrumbore() {} |
| 39 | |
| 40 | __forceinline SubGridQuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {} |
| 41 | |
| 42 | __forceinline void intersect(RayHit& ray, IntersectContext* context, |
| 43 | const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, |
| 44 | const GridMesh::Grid &g, const SubGrid& subgrid) const |
| 45 | { |
| 46 | UVIdentity<M> mapUV; |
| 47 | MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV); |
| 48 | MoellerTrumboreIntersector1<M> intersector(ray,nullptr); |
| 49 | Intersect1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID()); |
| 50 | |
| 51 | /* intersect first triangle */ |
| 52 | if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) |
| 53 | { |
| 54 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 55 | epilog(hit.valid,hit); |
| 56 | } |
| 57 | |
| 58 | /* intersect second triangle */ |
| 59 | if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) |
| 60 | { |
| 61 | hit.U = hit.absDen - hit.U; |
| 62 | hit.V = hit.absDen - hit.V; |
| 63 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 64 | epilog(hit.valid,hit); |
| 65 | } |
| 66 | } |
| 67 | |
| 68 | __forceinline bool occluded(Ray& ray, IntersectContext* context, |
| 69 | const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, |
| 70 | const GridMesh::Grid &g, const SubGrid& subgrid) const |
| 71 | { |
| 72 | UVIdentity<M> mapUV; |
| 73 | MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV); |
| 74 | MoellerTrumboreIntersector1<M> intersector(ray,nullptr); |
| 75 | Occluded1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID()); |
| 76 | |
| 77 | /* intersect first triangle */ |
| 78 | if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) |
| 79 | { |
| 80 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 81 | if (epilog(hit.valid,hit)) |
| 82 | return true; |
| 83 | } |
| 84 | |
| 85 | /* intersect second triangle */ |
| 86 | if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) |
| 87 | { |
| 88 | hit.U = hit.absDen - hit.U; |
| 89 | hit.V = hit.absDen - hit.V; |
| 90 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 91 | if (epilog(hit.valid,hit)) |
| 92 | return true; |
| 93 | } |
| 94 | return false; |
| 95 | } |
| 96 | }; |
| 97 | |
| 98 | #if defined (__AVX__) |
| 99 | |
| 100 | /*! Intersects 4 quads with 1 ray using AVX */ |
| 101 | template<bool filter> |
| 102 | struct SubGridQuadMIntersector1MoellerTrumbore<4,filter> |
| 103 | { |
| 104 | __forceinline SubGridQuadMIntersector1MoellerTrumbore() {} |
| 105 | |
| 106 | __forceinline SubGridQuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {} |
| 107 | |
| 108 | template<typename Epilog> |
| 109 | __forceinline bool intersect(Ray& ray, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid& subgrid, const Epilog& epilog) const |
| 110 | { |
| 111 | const Vec3vf8 vtx0(vfloat8(v0.x,v2.x),vfloat8(v0.y,v2.y),vfloat8(v0.z,v2.z)); |
| 112 | #if !defined(EMBREE_BACKFACE_CULLING) |
| 113 | const Vec3vf8 vtx1(vfloat8(v1.x),vfloat8(v1.y),vfloat8(v1.z)); |
| 114 | const Vec3vf8 vtx2(vfloat8(v3.x),vfloat8(v3.y),vfloat8(v3.z)); |
| 115 | #else |
| 116 | const Vec3vf8 vtx1(vfloat8(v1.x,v3.x),vfloat8(v1.y,v3.y),vfloat8(v1.z,v3.z)); |
| 117 | const Vec3vf8 vtx2(vfloat8(v3.x,v1.x),vfloat8(v3.y,v1.y),vfloat8(v3.z,v1.z)); |
| 118 | #endif |
| 119 | UVIdentity<8> mapUV; |
| 120 | MoellerTrumboreHitM<8,UVIdentity<8>> hit(mapUV); |
| 121 | MoellerTrumboreIntersector1<8> intersector(ray,nullptr); |
| 122 | const vbool8 flags(0,0,0,0,1,1,1,1); |
| 123 | if (unlikely(intersector.intersect(ray,vtx0,vtx1,vtx2,mapUV,hit))) |
| 124 | { |
| 125 | /* correct U,V interpolation across the entire grid */ |
| 126 | const vfloat8 U = select(flags,hit.absDen - hit.V,hit.U); |
| 127 | const vfloat8 V = select(flags,hit.absDen - hit.U,hit.V); |
| 128 | hit.U = U; |
| 129 | hit.V = V; |
| 130 | hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); |
| 131 | interpolateUV<8>(hit,g,subgrid,vint<8>(0,1,1,0,0,1,1,0),vint<8>(0,0,1,1,0,0,1,1)); |
| 132 | if (unlikely(epilog(hit.valid,hit))) |
| 133 | return true; |
| 134 | } |
| 135 | return false; |
| 136 | } |
| 137 | |
| 138 | __forceinline bool intersect(RayHit& ray, IntersectContext* context, |
| 139 | const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, |
| 140 | const GridMesh::Grid &g, const SubGrid& subgrid) const |
| 141 | { |
| 142 | return intersect(ray,v0,v1,v2,v3,g,subgrid,Intersect1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID())); |
| 143 | } |
| 144 | |
| 145 | __forceinline bool occluded(Ray& ray, IntersectContext* context, |
| 146 | const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, |
| 147 | const GridMesh::Grid &g, const SubGrid& subgrid) const |
| 148 | { |
| 149 | return intersect(ray,v0,v1,v2,v3,g,subgrid,Occluded1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID())); |
| 150 | } |
| 151 | }; |
| 152 | |
| 153 | #endif |
| 154 | |
| 155 | // ============================================================================================================================ |
| 156 | // ============================================================================================================================ |
| 157 | // ============================================================================================================================ |
| 158 | |
| 159 | |
| 160 | /* ----------------------------- */ |
| 161 | /* -- ray packet intersectors -- */ |
| 162 | /* ----------------------------- */ |
| 163 | |
| 164 | template<int K> |
| 165 | __forceinline void interpolateUV(const vbool<K>& valid, MoellerTrumboreHitK<K,UVIdentity<K>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const unsigned int i) |
| 166 | { |
| 167 | /* correct U,V interpolation across the entire grid */ |
| 168 | const unsigned int sx = subgrid.x() + (unsigned int)(i % 2); |
| 169 | const unsigned int sy = subgrid.y() + (unsigned int)(i >>1); |
| 170 | const float inv_resX = rcp((float)(int)(g.resX-1)); |
| 171 | const float inv_resY = rcp((float)(int)(g.resY-1)); |
| 172 | hit.U = select(valid,(hit.U + vfloat<K>((float)sx) * hit.absDen) * inv_resX,hit.U); |
| 173 | hit.V = select(valid,(hit.V + vfloat<K>((float)sy) * hit.absDen) * inv_resY,hit.V); |
| 174 | } |
| 175 | |
| 176 | template<int M, int K, bool filter> |
| 177 | struct SubGridQuadMIntersectorKMoellerTrumboreBase |
| 178 | { |
| 179 | __forceinline SubGridQuadMIntersectorKMoellerTrumboreBase(const vbool<K>& valid, const RayK<K>& ray) {} |
| 180 | |
| 181 | template<typename Epilog> |
| 182 | __forceinline bool intersectK(const vbool<K>& valid, |
| 183 | RayK<K>& ray, |
| 184 | const Vec3vf<K>& v0, |
| 185 | const Vec3vf<K>& v1, |
| 186 | const Vec3vf<K>& v2, |
| 187 | const Vec3vf<K>& v3, |
| 188 | const GridMesh::Grid &g, |
| 189 | const SubGrid &subgrid, |
| 190 | const unsigned int i, |
| 191 | const Epilog& epilog) const |
| 192 | { |
| 193 | UVIdentity<K> mapUV; |
| 194 | MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV); |
| 195 | MoellerTrumboreIntersectorK<M,K> intersector; |
| 196 | |
| 197 | const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit); |
| 198 | if (any(valid0)) |
| 199 | { |
| 200 | interpolateUV(valid0,hit,g,subgrid,i); |
| 201 | epilog(valid0,hit); |
| 202 | } |
| 203 | const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit); |
| 204 | if (any(valid1)) |
| 205 | { |
| 206 | hit.U = hit.absDen - hit.U; |
| 207 | hit.V = hit.absDen - hit.V; |
| 208 | interpolateUV(valid1,hit,g,subgrid,i); |
| 209 | epilog(valid1,hit); |
| 210 | } |
| 211 | return any(valid0|valid1); |
| 212 | } |
| 213 | |
| 214 | template<typename Epilog> |
| 215 | __forceinline bool occludedK(const vbool<K>& valid, |
| 216 | RayK<K>& ray, |
| 217 | const Vec3vf<K>& v0, |
| 218 | const Vec3vf<K>& v1, |
| 219 | const Vec3vf<K>& v2, |
| 220 | const Vec3vf<K>& v3, |
| 221 | const GridMesh::Grid &g, |
| 222 | const SubGrid &subgrid, |
| 223 | const unsigned int i, |
| 224 | const Epilog& epilog) const |
| 225 | { |
| 226 | UVIdentity<K> mapUV; |
| 227 | MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV); |
| 228 | MoellerTrumboreIntersectorK<M,K> intersector; |
| 229 | |
| 230 | vbool<K> valid_final = valid; |
| 231 | const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit); |
| 232 | if (any(valid0)) |
| 233 | { |
| 234 | interpolateUV(valid0,hit,g,subgrid,i); |
| 235 | epilog(valid0,hit); |
| 236 | valid_final &= !valid0; |
| 237 | } |
| 238 | if (none(valid_final)) return true; |
| 239 | const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit); |
| 240 | if (any(valid1)) |
| 241 | { |
| 242 | hit.U = hit.absDen - hit.U; |
| 243 | hit.V = hit.absDen - hit.V; |
| 244 | interpolateUV(valid1,hit,g,subgrid,i); |
| 245 | epilog(valid1,hit); |
| 246 | valid_final &= !valid1; |
| 247 | } |
| 248 | return none(valid_final); |
| 249 | } |
| 250 | |
| 251 | static __forceinline bool intersect1(RayK<K>& ray, |
| 252 | size_t k, |
| 253 | const Vec3vf<M>& v0, |
| 254 | const Vec3vf<M>& v1, |
| 255 | const Vec3vf<M>& v2, |
| 256 | MoellerTrumboreHitM<M,UVIdentity<M>> &hit) |
| 257 | { |
| 258 | const Vec3vf<M> e1 = v0-v1; |
| 259 | const Vec3vf<M> e2 = v2-v0; |
| 260 | MoellerTrumboreIntersectorK<8,K> intersector; |
| 261 | UVIdentity<M> mapUV; |
| 262 | return intersector.intersectEdge(ray,k,v0,e1,e2,mapUV,hit); |
| 263 | } |
| 264 | |
| 265 | }; |
| 266 | |
| 267 | template<int M, int K, bool filter> |
| 268 | struct SubGridQuadMIntersectorKMoellerTrumbore : public SubGridQuadMIntersectorKMoellerTrumboreBase<M,K,filter> |
| 269 | { |
| 270 | __forceinline SubGridQuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray) |
| 271 | : SubGridQuadMIntersectorKMoellerTrumboreBase<M,K,filter>(valid,ray) {} |
| 272 | |
| 273 | __forceinline void intersect1(RayHitK<K>& ray, size_t k, IntersectContext* context, |
| 274 | const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const |
| 275 | { |
| 276 | UVIdentity<M> mapUV; |
| 277 | MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV); |
| 278 | Intersect1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID()); |
| 279 | MoellerTrumboreIntersectorK<M,K> intersector; |
| 280 | /* intersect first triangle */ |
| 281 | if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) |
| 282 | { |
| 283 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 284 | epilog(hit.valid,hit); |
| 285 | } |
| 286 | |
| 287 | /* intersect second triangle */ |
| 288 | if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) |
| 289 | { |
| 290 | hit.U = hit.absDen - hit.U; |
| 291 | hit.V = hit.absDen - hit.V; |
| 292 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 293 | epilog(hit.valid,hit); |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | __forceinline bool occluded1(RayK<K>& ray, size_t k, IntersectContext* context, |
| 298 | const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const |
| 299 | { |
| 300 | UVIdentity<M> mapUV; |
| 301 | MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV); |
| 302 | Occluded1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID()); |
| 303 | MoellerTrumboreIntersectorK<M,K> intersector; |
| 304 | /* intersect first triangle */ |
| 305 | if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) |
| 306 | { |
| 307 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 308 | if (epilog(hit.valid,hit)) return true; |
| 309 | } |
| 310 | |
| 311 | /* intersect second triangle */ |
| 312 | if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) |
| 313 | { |
| 314 | hit.U = hit.absDen - hit.U; |
| 315 | hit.V = hit.absDen - hit.V; |
| 316 | interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1)); |
| 317 | if (epilog(hit.valid,hit)) return true; |
| 318 | } |
| 319 | return false; |
| 320 | } |
| 321 | }; |
| 322 | |
| 323 | |
| 324 | #if defined (__AVX__) |
| 325 | |
| 326 | /*! Intersects 4 quads with 1 ray using AVX */ |
| 327 | template<int K, bool filter> |
| 328 | struct SubGridQuadMIntersectorKMoellerTrumbore<4,K,filter> : public SubGridQuadMIntersectorKMoellerTrumboreBase<4,K,filter> |
| 329 | { |
| 330 | __forceinline SubGridQuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray) |
| 331 | : SubGridQuadMIntersectorKMoellerTrumboreBase<4,K,filter>(valid,ray) {} |
| 332 | |
| 333 | template<typename Epilog> |
| 334 | __forceinline bool intersect1(RayK<K>& ray, size_t k,const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, |
| 335 | const GridMesh::Grid &g, const SubGrid &subgrid, const Epilog& epilog) const |
| 336 | { |
| 337 | const Vec3vf8 vtx0(vfloat8(v0.x,v2.x),vfloat8(v0.y,v2.y),vfloat8(v0.z,v2.z)); |
| 338 | #if !defined(EMBREE_BACKFACE_CULLING) |
| 339 | const Vec3vf8 vtx1(vfloat8(v1.x),vfloat8(v1.y),vfloat8(v1.z)); |
| 340 | const Vec3vf8 vtx2(vfloat8(v3.x),vfloat8(v3.y),vfloat8(v3.z)); |
| 341 | #else |
| 342 | const Vec3vf8 vtx1(vfloat8(v1.x,v3.x),vfloat8(v1.y,v3.y),vfloat8(v1.z,v3.z)); |
| 343 | const Vec3vf8 vtx2(vfloat8(v3.x,v1.x),vfloat8(v3.y,v1.y),vfloat8(v3.z,v1.z)); |
| 344 | #endif |
| 345 | const vbool8 flags(0,0,0,0,1,1,1,1); |
| 346 | |
| 347 | UVIdentity<8> mapUV; |
| 348 | MoellerTrumboreHitM<8,UVIdentity<8>> hit(mapUV); |
| 349 | if (SubGridQuadMIntersectorKMoellerTrumboreBase<8,K,filter>::intersect1(ray,k,vtx0,vtx1,vtx2,hit)) |
| 350 | { |
| 351 | const vfloat8 U = select(flags,hit.absDen - hit.V,hit.U); |
| 352 | const vfloat8 V = select(flags,hit.absDen - hit.U,hit.V); |
| 353 | hit.U = U; |
| 354 | hit.V = V; |
| 355 | hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); |
| 356 | interpolateUV<8>(hit,g,subgrid,vint<8>(0,1,1,0,0,1,1,0),vint<8>(0,0,1,1,0,0,1,1)); |
| 357 | if (unlikely(epilog(hit.valid,hit))) |
| 358 | return true; |
| 359 | |
| 360 | } |
| 361 | return false; |
| 362 | } |
| 363 | |
| 364 | __forceinline bool intersect1(RayHitK<K>& ray, size_t k, IntersectContext* context, |
| 365 | const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const |
| 366 | { |
| 367 | return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Intersect1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID())); |
| 368 | } |
| 369 | |
| 370 | __forceinline bool occluded1(RayK<K>& ray, size_t k, IntersectContext* context, |
| 371 | const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const |
| 372 | { |
| 373 | return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Occluded1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID())); |
| 374 | } |
| 375 | }; |
| 376 | |
| 377 | #endif |
| 378 | |
| 379 | |
| 380 | |
| 381 | } |
| 382 | } |
| 383 | |