// ======================================================================== // // Copyright 2009-2017 Intel Corporation // // // // Licensed under the Apache License, Version 2.0 (the "License"); // // you may not use this file except in compliance with the License. // // You may obtain a copy of the License at // // // // http://www.apache.org/licenses/LICENSE-2.0 // // // // Unless required by applicable law or agreed to in writing, software // // distributed under the License is distributed on an "AS IS" BASIS, // // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // // See the License for the specific language governing permissions and // // limitations under the License. // // ======================================================================== // #pragma once #include "quad_intersector_moeller.h" /*! Modified Pluecker ray/triangle intersector. The test first shifts * the ray origin into the origin of the coordinate system and then * uses Pluecker coordinates for the intersection. Due to the shift, * the Pluecker coordinate calculation simplifies and the tests get * numerically stable. The edge equations are watertight along the * edge for neighboring triangles. */ namespace embree { namespace isa { struct PlueckerIntersectorTriangle1 { template static __forceinline bool intersect(Ray& ray, const Vec3>& tri_v0, const Vec3>& tri_v1, const Vec3>& tri_v2, const vbool& flags, const Epilog& epilog) { /* calculate vertices relative to ray origin */ typedef Vec3> Vec3vfM; const Vec3vfM O = Vec3vfM(ray.org); const Vec3vfM D = Vec3vfM(ray.dir); const Vec3vfM v0 = tri_v0-O; const Vec3vfM v1 = tri_v1-O; const Vec3vfM v2 = tri_v2-O; /* calculate triangle edges */ const Vec3vfM e0 = v2-v0; const Vec3vfM e1 = v0-v1; const Vec3vfM e2 = v1-v2; /* perform edge tests */ const vfloat U = dot(cross(v2+v0,e0),D); const vfloat V = dot(cross(v0+v1,e1),D); const vfloat W = dot(cross(v1+v2,e2),D); #if defined(EMBREE_BACKFACE_CULLING) const vfloat maxUVW = max(U,V,W); vbool valid = maxUVW <= 0.0f; #else const vfloat minUVW = min(U,V,W); const vfloat maxUVW = max(U,V,W); vbool valid = (maxUVW <= 0.0f) | (minUVW >= 0.0f); #endif if (unlikely(none(valid))) return false; /* calculate geometry normal and denominator */ const Vec3vfM Ng = stable_triangle_normal(e2,e1,e0); const vfloat den = twice(dot(Ng,D)); const vfloat absDen = abs(den); const vfloat sgnDen = signmsk(den); /* perform depth test */ const vfloat T = twice(dot(v0,Ng)); valid &= ((T^sgnDen) >= absDen*vfloat(ray.tnear)); valid &=(absDen*vfloat(ray.tfar) >= (T^sgnDen)); if (unlikely(none(valid))) return false; /* avoid division by 0 */ valid &= den != vfloat(zero); if (unlikely(none(valid))) return false; /* update hit information */ QuadHitM hit(valid,U,V,T,den,Ng,flags); return epilog(valid,hit); } }; /*! Intersects M quads with 1 ray */ template struct QuadMIntersector1Pluecker { __forceinline QuadMIntersector1Pluecker() {} __forceinline QuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {} __forceinline void intersect(Ray& ray, IntersectContext* context, const Vec3>& v0, const Vec3>& v1, const Vec3>& v2, const Vec3>& v3, const vint& geomID, const vint& primID) const { Intersect1EpilogM epilog(ray,context,geomID,primID); PlueckerIntersectorTriangle1::intersect(ray,v0,v1,v3,vbool(false),epilog); PlueckerIntersectorTriangle1::intersect(ray,v2,v3,v1,vbool(true),epilog); } __forceinline bool occluded(Ray& ray, IntersectContext* context, const Vec3>& v0, const Vec3>& v1, const Vec3>& v2, const Vec3>& v3, const vint& geomID, const vint& primID) const { Occluded1EpilogM epilog(ray,context,geomID,primID); if (PlueckerIntersectorTriangle1::intersect(ray,v0,v1,v3,vbool(false),epilog)) return true; if (PlueckerIntersectorTriangle1::intersect(ray,v2,v3,v1,vbool(true ),epilog)) return true; return false; } }; #if defined(__AVX512F__) /*! Intersects 4 quads with 1 ray using AVX512 */ template struct QuadMIntersector1Pluecker<4,filter> { __forceinline QuadMIntersector1Pluecker() {} __forceinline QuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {} template __forceinline bool intersect(Ray& ray, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const Epilog& epilog) const { const Vec3vf16 vtx0(select(0x0f0f,vfloat16(v0.x),vfloat16(v2.x)), select(0x0f0f,vfloat16(v0.y),vfloat16(v2.y)), select(0x0f0f,vfloat16(v0.z),vfloat16(v2.z))); #if !defined(EMBREE_BACKFACE_CULLING) const Vec3vf16 vtx1(vfloat16(v1.x),vfloat16(v1.y),vfloat16(v1.z)); const Vec3vf16 vtx2(vfloat16(v3.x),vfloat16(v3.y),vfloat16(v3.z)); #else const Vec3vf16 vtx1(select(0x0f0f,vfloat16(v1.x),vfloat16(v3.x)), select(0x0f0f,vfloat16(v1.y),vfloat16(v3.y)), select(0x0f0f,vfloat16(v1.z),vfloat16(v3.z))); const Vec3vf16 vtx2(select(0x0f0f,vfloat16(v3.x),vfloat16(v1.x)), select(0x0f0f,vfloat16(v3.y),vfloat16(v1.y)), select(0x0f0f,vfloat16(v3.z),vfloat16(v1.z))); #endif const vbool16 flags(0xf0f0); return PlueckerIntersectorTriangle1::intersect(ray,vtx0,vtx1,vtx2,flags,epilog); } __forceinline bool intersect(Ray& ray, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect(ray,v0,v1,v2,v3,Intersect1EpilogM<8,16,filter>(ray,context,vint8(geomID),vint8(primID))); } __forceinline bool occluded(Ray& ray, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect(ray,v0,v1,v2,v3,Occluded1EpilogM<8,16,filter>(ray,context,vint8(geomID),vint8(primID))); } }; #elif defined (__AVX__) /*! Intersects 4 quads with 1 ray using AVX */ template struct QuadMIntersector1Pluecker<4,filter> { __forceinline QuadMIntersector1Pluecker() {} __forceinline QuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {} template __forceinline bool intersect(Ray& ray, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const Epilog& epilog) const { const Vec3vf8 vtx0(vfloat8(v0.x,v2.x),vfloat8(v0.y,v2.y),vfloat8(v0.z,v2.z)); #if !defined(EMBREE_BACKFACE_CULLING) const Vec3vf8 vtx1(vfloat8(v1.x),vfloat8(v1.y),vfloat8(v1.z)); const Vec3vf8 vtx2(vfloat8(v3.x),vfloat8(v3.y),vfloat8(v3.z)); #else const Vec3vf8 vtx1(vfloat8(v1.x,v3.x),vfloat8(v1.y,v3.y),vfloat8(v1.z,v3.z)); const Vec3vf8 vtx2(vfloat8(v3.x,v1.x),vfloat8(v3.y,v1.y),vfloat8(v3.z,v1.z)); #endif const vbool8 flags(0,0,0,0,1,1,1,1); return PlueckerIntersectorTriangle1::intersect(ray,vtx0,vtx1,vtx2,flags,epilog); } __forceinline bool intersect(Ray& ray, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect(ray,v0,v1,v2,v3,Intersect1EpilogM<8,8,filter>(ray,context,vint8(geomID),vint8(primID))); } __forceinline bool occluded(Ray& ray, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect(ray,v0,v1,v2,v3,Occluded1EpilogM<8,8,filter>(ray,context,vint8(geomID),vint8(primID))); } }; #endif /* ----------------------------- */ /* -- ray packet intersectors -- */ /* ----------------------------- */ struct PlueckerIntersector1KTriangleM { /*! Intersect k'th ray from ray packet of size K with M triangles. */ template static __forceinline bool intersect1(RayK& ray, size_t k, const Vec3>& tri_v0, const Vec3>& tri_v1, const Vec3>& tri_v2, const vbool& flags, const Epilog& epilog) { /* calculate vertices relative to ray origin */ typedef Vec3> Vec3vfM; const Vec3vfM O = broadcast>(ray.org,k); const Vec3vfM D = broadcast>(ray.dir,k); const Vec3vfM v0 = tri_v0-O; const Vec3vfM v1 = tri_v1-O; const Vec3vfM v2 = tri_v2-O; /* calculate triangle edges */ const Vec3vfM e0 = v2-v0; const Vec3vfM e1 = v0-v1; const Vec3vfM e2 = v1-v2; /* perform edge tests */ const vfloat U = dot(cross(v2+v0,e0),D); const vfloat V = dot(cross(v0+v1,e1),D); const vfloat W = dot(cross(v1+v2,e2),D); const vfloat minUVW MAYBE_UNUSED = min(U,V,W); const vfloat maxUVW = max(U,V,W); #if defined(EMBREE_BACKFACE_CULLING) vbool valid = maxUVW <= 0.0f; #else vbool valid = (minUVW >= 0.0f) | (maxUVW <= 0.0f); #endif if (unlikely(none(valid))) return false; /* calculate geometry normal and denominator */ const Vec3vfM Ng = stable_triangle_normal(e2,e1,e0); const vfloat den = twice(dot(Ng,D)); const vfloat absDen = abs(den); const vfloat sgnDen = signmsk(den); /* perform depth test */ const vfloat T = twice(dot(v0,Ng)); valid &= ((T^sgnDen) >= absDen*vfloat(ray.tnear[k])); valid &= (absDen*vfloat(ray.tfar[k]) >= (T^sgnDen)); if (unlikely(none(valid))) return false; /* avoid division by 0 */ valid &= den != vfloat(zero); if (unlikely(none(valid))) return false; /* calculate hit information */ QuadHitM hit(valid,U,V,T,den,Ng,flags); return epilog(valid,hit); } }; template struct QuadMIntersectorKPlueckerBase { __forceinline QuadMIntersectorKPlueckerBase(const vbool& valid, const RayK& ray) {} /*! Intersects K rays with one of M triangles. */ template __forceinline vbool intersectK(const vbool& valid0, RayK& ray, const Vec3>& tri_v0, const Vec3>& tri_v1, const Vec3>& tri_v2, const vbool& flags, const Epilog& epilog) const { /* calculate vertices relative to ray origin */ typedef Vec3> Vec3vfK; vbool valid = valid0; const Vec3vfK O = ray.org; const Vec3vfK D = ray.dir; const Vec3vfK v0 = tri_v0-O; const Vec3vfK v1 = tri_v1-O; const Vec3vfK v2 = tri_v2-O; /* calculate triangle edges */ const Vec3vfK e0 = v2-v0; const Vec3vfK e1 = v0-v1; const Vec3vfK e2 = v1-v2; /* perform edge tests */ const vfloat U = dot(Vec3vfK(cross(v2+v0,e0)),D); const vfloat V = dot(Vec3vfK(cross(v0+v1,e1)),D); const vfloat W = dot(Vec3vfK(cross(v1+v2,e2)),D); #if defined(EMBREE_BACKFACE_CULLING) const vfloat maxUVW = max(U,V,W); valid &= maxUVW <= 0.0f; #else const vfloat minUVW = min(U,V,W); const vfloat maxUVW = max(U,V,W); valid &= (maxUVW <= 0.0f) | (minUVW >= 0.0f); #endif if (unlikely(none(valid))) return false; /* calculate geometry normal and denominator */ const Vec3vfK Ng = stable_triangle_normal(e2,e1,e0); const vfloat den = twice(dot(Vec3vfK(Ng),D)); const vfloat absDen = abs(den); const vfloat sgnDen = signmsk(den); /* perform depth test */ const vfloat T = twice(dot(v0,Vec3vfK(Ng))); valid &= ((T^sgnDen) >= absDen*ray.tnear); valid &= (absDen*ray.tfar >= (T^sgnDen)); if (unlikely(none(valid))) return false; /* avoid division by 0 */ valid &= den != vfloat(zero); if (unlikely(none(valid))) return false; /* calculate hit information */ QuadHitK hit(U,V,T,den,Ng,flags); return epilog(valid,hit); } /*! Intersects K rays with one of M quads. */ template __forceinline bool intersectK(const vbool& valid0, RayK& ray, const Vec3>& v0, const Vec3>& v1, const Vec3>& v2, const Vec3>& v3, const Epilog& epilog) const { intersectK(valid0,ray,v0,v1,v3,vbool(false),epilog); if (none(valid0)) return true; intersectK(valid0,ray,v2,v3,v1,vbool(true ),epilog); return none(valid0); } }; template struct QuadMIntersectorKPluecker : public QuadMIntersectorKPlueckerBase { __forceinline QuadMIntersectorKPluecker(const vbool& valid, const RayK& ray) : QuadMIntersectorKPlueckerBase(valid,ray) {} __forceinline void intersect1(RayK& ray, size_t k, IntersectContext* context, const Vec3>& v0, const Vec3>& v1, const Vec3>& v2, const Vec3>& v3, const vint& geomID, const vint& primID) const { Intersect1KEpilogM epilog(ray,k,context,geomID,primID); PlueckerIntersector1KTriangleM::intersect1(ray,k,v0,v1,v3,vbool(false),epilog); PlueckerIntersector1KTriangleM::intersect1(ray,k,v2,v3,v1,vbool(true ),epilog); } __forceinline bool occluded1(RayK& ray, size_t k, IntersectContext* context, const Vec3>& v0, const Vec3>& v1, const Vec3>& v2, const Vec3>& v3, const vint& geomID, const vint& primID) const { Occluded1KEpilogM epilog(ray,k,context,geomID,primID); if (PlueckerIntersector1KTriangleM::intersect1(ray,k,v0,v1,v3,vbool(false),epilog)) return true; if (PlueckerIntersector1KTriangleM::intersect1(ray,k,v2,v3,v1,vbool(true ),epilog)) return true; return false; } }; #if defined(__AVX512F__) /*! Intersects 4 quads with 1 ray using AVX512 */ template struct QuadMIntersectorKPluecker<4,K,filter> : public QuadMIntersectorKPlueckerBase<4,K,filter> { __forceinline QuadMIntersectorKPluecker(const vbool& valid, const RayK& ray) : QuadMIntersectorKPlueckerBase<4,K,filter>(valid,ray) {} template __forceinline bool intersect1(RayK& ray, size_t k, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const Epilog& epilog) const { const Vec3vf16 vtx0(select(0x0f0f,vfloat16(v0.x),vfloat16(v2.x)), select(0x0f0f,vfloat16(v0.y),vfloat16(v2.y)), select(0x0f0f,vfloat16(v0.z),vfloat16(v2.z))); #if !defined(EMBREE_BACKFACE_CULLING) const Vec3vf16 vtx1(vfloat16(v1.x),vfloat16(v1.y),vfloat16(v1.z)); const Vec3vf16 vtx2(vfloat16(v3.x),vfloat16(v3.y),vfloat16(v3.z)); #else const Vec3vf16 vtx1(select(0x0f0f,vfloat16(v1.x),vfloat16(v3.x)), select(0x0f0f,vfloat16(v1.y),vfloat16(v3.y)), select(0x0f0f,vfloat16(v1.z),vfloat16(v3.z))); const Vec3vf16 vtx2(select(0x0f0f,vfloat16(v3.x),vfloat16(v1.x)), select(0x0f0f,vfloat16(v3.y),vfloat16(v1.y)), select(0x0f0f,vfloat16(v3.z),vfloat16(v1.z))); #endif const vbool16 flags(0xf0f0); return PlueckerIntersector1KTriangleM::intersect1(ray,k,vtx0,vtx1,vtx2,flags,epilog); } __forceinline bool intersect1(RayK& ray, size_t k, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect1(ray,k,v0,v1,v2,v3,Intersect1KEpilogM<8,16,K,filter>(ray,k,context,vint8(geomID),vint8(primID))); } __forceinline bool occluded1(RayK& ray, size_t k, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect1(ray,k,v0,v1,v2,v3,Occluded1KEpilogM<8,16,K,filter>(ray,k,context,vint8(geomID),vint8(primID))); } }; #elif defined (__AVX__) /*! Intersects 4 quads with 1 ray using AVX */ template struct QuadMIntersectorKPluecker<4,K,filter> : public QuadMIntersectorKPlueckerBase<4,K,filter> { __forceinline QuadMIntersectorKPluecker(const vbool& valid, const RayK& ray) : QuadMIntersectorKPlueckerBase<4,K,filter>(valid,ray) {} template __forceinline bool intersect1(RayK& ray, size_t k, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const Epilog& epilog) const { const Vec3vf8 vtx0(vfloat8(v0.x,v2.x),vfloat8(v0.y,v2.y),vfloat8(v0.z,v2.z)); const vbool8 flags(0,0,0,0,1,1,1,1); #if !defined(EMBREE_BACKFACE_CULLING) const Vec3vf8 vtx1(vfloat8(v1.x),vfloat8(v1.y),vfloat8(v1.z)); const Vec3vf8 vtx2(vfloat8(v3.x),vfloat8(v3.y),vfloat8(v3.z)); #else const Vec3vf8 vtx1(vfloat8(v1.x,v3.x),vfloat8(v1.y,v3.y),vfloat8(v1.z,v3.z)); const Vec3vf8 vtx2(vfloat8(v3.x,v1.x),vfloat8(v3.y,v1.y),vfloat8(v3.z,v1.z)); #endif return PlueckerIntersector1KTriangleM::intersect1(ray,k,vtx0,vtx1,vtx2,flags,epilog); } __forceinline bool intersect1(RayK& ray, size_t k, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect1(ray,k,v0,v1,v2,v3,Intersect1KEpilogM<8,8,K,filter>(ray,k,context,vint8(geomID),vint8(primID))); } __forceinline bool occluded1(RayK& ray, size_t k, IntersectContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const vint4& geomID, const vint4& primID) const { return intersect1(ray,k,v0,v1,v2,v3,Occluded1KEpilogM<8,8,K,filter>(ray,k,context,vint8(geomID),vint8(primID))); } }; #endif } }