initial commit, 4.5 stable

thirdparty/embree/kernels/geometry/cone.h

@@ -0,0 +1,321 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct Cone
+    {
+      const Vec3fa p0; //!< start position of cone
+      const Vec3fa p1; //!< end position of cone
+      const float r0;  //!< start radius of cone
+      const float r1;  //!< end radius of cone
+
+      __forceinline Cone(const Vec3fa& p0, const float r0, const Vec3fa& p1, const float r1) 
+        : p0(p0), p1(p1), r0(r0), r1(r1) {}
+
+      __forceinline bool intersect(const Vec3fa& org, const Vec3fa& dir, 
+                                   BBox1f& t_o, 
+                                   float& u0_o, Vec3fa& Ng0_o, 
+                                   float& u1_o, Vec3fa& Ng1_o) const 
+      {
+        /* calculate quadratic equation to solve */
+        const Vec3fa v0 = p0-org;
+        const Vec3fa v1 = p1-org;
+        
+        const float rl = rcp_length(v1-v0);
+        const Vec3fa P0 = v0, dP = (v1-v0)*rl;
+        const float dr = (r1-r0)*rl;
+        const Vec3fa O = -P0, dO = dir;
+        
+        const float dOdO = dot(dO,dO);
+        const float OdO = dot(dO,O);
+        const float OO = dot(O,O);
+        const float dOz = dot(dP,dO);
+        const float Oz = dot(dP,O);
+
+        const float R = r0 + Oz*dr;          
+        const float A = dOdO - sqr(dOz) * (1.0f+sqr(dr));
+        const float B = 2.0f * (OdO - dOz*(Oz + R*dr));
+        const float C = OO - (sqr(Oz) + sqr(R));
+
+        /* we miss the cone if determinant is smaller than zero */
+        const float D = B*B - 4.0f*A*C;
+        if (D < 0.0f) return false;
+
+        /* special case for rays that are "parallel" to the cone */
+        const float eps = float(1<<8)*float(ulp)*max(abs(dOdO),abs(sqr(dOz)));
+        if (unlikely(abs(A) < eps))
+        {
+          /* cylinder case */
+          if (abs(dr) < 16.0f*float(ulp)) {
+            if (C <= 0.0f) { t_o = BBox1f(neg_inf,pos_inf); return true; } 
+            else           { t_o = BBox1f(pos_inf,neg_inf); return false; }
+          }
+
+          /* cone case */
+          else 
+          {
+            /* if we hit the negative cone there cannot be a hit */
+            const float t = -C/B;
+            const float z0 = Oz+t*dOz;
+            const float z0r = r0+z0*dr;
+            if (z0r < 0.0f) return false;
+
+            /* test if we start inside or outside the cone */
+            if (dOz*dr > 0.0f) t_o = BBox1f(t,pos_inf);
+            else               t_o = BBox1f(neg_inf,t);
+          }
+        }
+
+        /* standard case for "non-parallel" rays */
+        else
+        {
+          const float Q = sqrt(D);
+          const float rcp_2A = rcp(2.0f*A);
+          t_o.lower = (-B-Q)*rcp_2A;
+          t_o.upper = (-B+Q)*rcp_2A;
+          
+          /* standard case where both hits are on same cone */
+          if (likely(A > 0.0f)) {
+            const float z0 = Oz+t_o.lower*dOz;
+            const float z0r = r0+z0*dr;
+            if (z0r < 0.0f) return false;
+          } 
+
+          /* special case where the hits are on the positive and negative cone */
+          else 
+          {
+            /* depending on the ray direction and the open direction
+             * of the cone we have a hit from inside or outside the
+             * cone */
+            if (dOz*dr > 0) t_o.upper = pos_inf;
+            else            t_o.lower = neg_inf;
+          }
+        }
+
+        /* calculates u and Ng for near hit */
+        {
+          u0_o = (Oz+t_o.lower*dOz)*rl;
+          const Vec3fa Pr = t_o.lower*dir;
+          const Vec3fa Pl = v0 + u0_o*(v1-v0);
+          const Vec3fa R = normalize(Pr-Pl);
+          const Vec3fa U = (p1-p0)+(r1-r0)*R;
+          const Vec3fa V = cross(p1-p0,R);
+          Ng0_o = cross(V,U);
+        }
+
+        /* calculates u and Ng for far hit */
+        {
+          u1_o = (Oz+t_o.upper*dOz)*rl;
+          const Vec3fa Pr = t_o.upper*dir;
+          const Vec3fa Pl = v0 + u1_o*(v1-v0);
+          const Vec3fa R = normalize(Pr-Pl);
+          const Vec3fa U = (p1-p0)+(r1-r0)*R;
+          const Vec3fa V = cross(p1-p0,R);
+          Ng1_o = cross(V,U);
+        }
+        return true;
+      }
+
+      __forceinline bool intersect(const Vec3fa& org, const Vec3fa& dir, BBox1f& t_o) const 
+      {
+        float u0_o; Vec3fa Ng0_o; float u1_o; Vec3fa Ng1_o;
+        return intersect(org,dir,t_o,u0_o,Ng0_o,u1_o,Ng1_o);
+      }
+
+      static bool verify(const size_t id, const Cone& cone, const Ray& ray, bool shouldhit, const float t0, const float t1)
+      {
+        float eps = 0.001f;
+        BBox1f t; bool hit;
+        hit = cone.intersect(ray.org,ray.dir,t);
+
+        bool failed = hit != shouldhit;
+        if (shouldhit) failed |= std::isinf(t0) ? t0 != t.lower : (t0 == -1E6) ? t.lower > -1E6f : abs(t0-t.lower) > eps;
+        if (shouldhit) failed |= std::isinf(t1) ? t1 != t.upper : (t1 == +1E6) ? t.upper < +1E6f : abs(t1-t.upper) > eps;
+        if (!failed) return true;
+        embree_cout << "Cone test " << id << " failed: cone = " << cone << ", ray = " << ray << ", hit = " << hit << ", t = " << t << embree_endl; 
+        return false;
+      }
+
+      /* verify cone class */
+      static bool verify()
+      {
+        bool passed = true;
+        const Cone cone0(Vec3fa(0.0f,0.0f,0.0f),0.0f,Vec3fa(1.0f,0.0f,0.0f),1.0f);
+        passed &= verify(0,cone0,Ray(Vec3fa(-2.0f,1.0f,0.0f),Vec3fa(+1.0f,+0.0f,+0.0f),0.0f,float(inf)),true,3.0f,pos_inf);
+        passed &= verify(1,cone0,Ray(Vec3fa(+2.0f,1.0f,0.0f),Vec3fa(-1.0f,+0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,1.0f);
+        passed &= verify(2,cone0,Ray(Vec3fa(-1.0f,0.0f,2.0f),Vec3fa(+0.0f,+0.0f,-1.0f),0.0f,float(inf)),false,0.0f,0.0f);
+        passed &= verify(3,cone0,Ray(Vec3fa(+1.0f,0.0f,2.0f),Vec3fa(+0.0f,+0.0f,-1.0f),0.0f,float(inf)),true,1.0f,3.0f);
+        passed &= verify(4,cone0,Ray(Vec3fa(-1.0f,0.0f,0.0f),Vec3fa(+1.0f,+0.0f,+0.0f),0.0f,float(inf)),true,1.0f,pos_inf);
+        passed &= verify(5,cone0,Ray(Vec3fa(+1.0f,0.0f,0.0f),Vec3fa(-1.0f,+0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,1.0f);
+        passed &= verify(6,cone0,Ray(Vec3fa(+0.0f,0.0f,1.0f),Vec3fa(+0.0f,+0.0f,-1.0f),0.0f,float(inf)),true,1.0f,1.0f);
+        passed &= verify(7,cone0,Ray(Vec3fa(+0.0f,1.0f,0.0f),Vec3fa(-1.0f,-1.0f,+0.0f),0.0f,float(inf)),false,0.0f,0.0f);
+        passed &= verify(8,cone0,Ray(Vec3fa(+0.0f,1.0f,0.0f),Vec3fa(+1.0f,-1.0f,+0.0f),0.0f,float(inf)),true,0.5f,+1E6);
+        passed &= verify(9,cone0,Ray(Vec3fa(+0.0f,1.0f,0.0f),Vec3fa(-1.0f,+1.0f,+0.0f),0.0f,float(inf)),true,-1E6,-0.5f);
+        const Cone cone1(Vec3fa(0.0f,0.0f,0.0f),1.0f,Vec3fa(1.0f,0.0f,0.0f),0.0f);
+        passed &= verify(10,cone1,Ray(Vec3fa(-2.0f,1.0f,0.0f),Vec3fa(+1.0f,+0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,2.0f);
+        passed &= verify(11,cone1,Ray(Vec3fa(-1.0f,0.0f,2.0f),Vec3fa(+0.0f,+0.0f,-1.0f),0.0f,float(inf)),true,0.0f,4.0f);
+        const Cone cylinder(Vec3fa(0.0f,0.0f,0.0f),1.0f,Vec3fa(1.0f,0.0f,0.0f),1.0f);
+        passed &= verify(12,cylinder,Ray(Vec3fa(-2.0f,1.0f,0.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),true,0.0f,2.0f);
+        passed &= verify(13,cylinder,Ray(Vec3fa(+2.0f,1.0f,0.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),true,0.0f,2.0f);
+        passed &= verify(14,cylinder,Ray(Vec3fa(+2.0f,1.0f,2.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),false,0.0f,0.0f);
+        passed &= verify(15,cylinder,Ray(Vec3fa(+0.0f,0.0f,0.0f),Vec3fa( 1.0f, 0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,pos_inf);
+        passed &= verify(16,cylinder,Ray(Vec3fa(+0.0f,0.0f,0.0f),Vec3fa(-1.0f, 0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,pos_inf);
+        passed &= verify(17,cylinder,Ray(Vec3fa(+0.0f,2.0f,0.0f),Vec3fa( 1.0f, 0.0f,+0.0f),0.0f,float(inf)),false,pos_inf,neg_inf);
+        passed &= verify(18,cylinder,Ray(Vec3fa(+0.0f,2.0f,0.0f),Vec3fa(-1.0f, 0.0f,+0.0f),0.0f,float(inf)),false,pos_inf,neg_inf);
+        return passed;
+      }
+
+      /*! output operator */
+      friend __forceinline embree_ostream operator<<(embree_ostream cout, const Cone& c) {
+        return cout << "Cone { p0 = " << c.p0 << ", r0 = " << c.r0 << ", p1 = " << c.p1 << ", r1 = " << c.r1 << "}";
+      }
+    };
+
+    template<int N>
+      struct ConeN
+    {
+      typedef Vec3<vfloat<N>> Vec3vfN;
+      
+      const Vec3vfN p0;     //!< start position of cone
+      const Vec3vfN p1;     //!< end position of cone
+      const vfloat<N> r0;   //!< start radius of cone
+      const vfloat<N> r1;   //!< end radius of cone
+
+      __forceinline ConeN(const Vec3vfN& p0, const vfloat<N>& r0, const Vec3vfN& p1, const vfloat<N>& r1) 
+        : p0(p0), p1(p1), r0(r0), r1(r1) {}
+
+      __forceinline Cone operator[] (const size_t i) const
+      {
+        assert(i<N);
+        return Cone(Vec3fa(p0.x[i],p0.y[i],p0.z[i]),r0[i],Vec3fa(p1.x[i],p1.y[i],p1.z[i]),r1[i]);
+      }
+
+      __forceinline vbool<N> intersect(const Vec3fa& org, const Vec3fa& dir, 
+                                       BBox<vfloat<N>>& t_o, 
+                                       vfloat<N>& u0_o, Vec3vfN& Ng0_o, 
+                                       vfloat<N>& u1_o, Vec3vfN& Ng1_o) const
+      {
+        /* calculate quadratic equation to solve */
+        const Vec3vfN v0 = p0-Vec3vfN(org);
+        const Vec3vfN v1 = p1-Vec3vfN(org);
+
+        const vfloat<N> rl = rcp_length(v1-v0);
+        const Vec3vfN P0 = v0, dP = (v1-v0)*rl;
+        const vfloat<N> dr = (r1-r0)*rl;
+        const Vec3vfN O = -P0, dO = dir;
+       
+        const vfloat<N> dOdO = dot(dO,dO);
+        const vfloat<N> OdO = dot(dO,O);
+        const vfloat<N> OO = dot(O,O);
+        const vfloat<N> dOz = dot(dP,dO);
+        const vfloat<N> Oz = dot(dP,O);
+        
+        const vfloat<N> R = r0 + Oz*dr;          
+        const vfloat<N> A = dOdO - sqr(dOz) * (vfloat<N>(1.0f)+sqr(dr));
+        const vfloat<N> B = 2.0f * (OdO - dOz*(Oz + R*dr));
+        const vfloat<N> C = OO - (sqr(Oz) + sqr(R));
+
+        /* we miss the cone if determinant is smaller than zero */
+        const vfloat<N> D = B*B - 4.0f*A*C;
+        vbool<N> valid = D >= 0.0f;
+        if (none(valid)) return valid;
+
+        /* special case for rays that are "parallel" to the cone */
+        const vfloat<N> eps = float(1<<8)*float(ulp)*max(abs(dOdO),abs(sqr(dOz)));
+        const vbool<N> validt = valid &  (abs(A) < eps);
+        const vbool<N> validf = valid & !(abs(A) < eps);
+        if (unlikely(any(validt)))
+        {
+          const vboolx validtt = validt & (abs(dr) <  16.0f*float(ulp));
+          const vboolx validtf = validt & (abs(dr) >= 16.0f*float(ulp));
+          
+          /* cylinder case */
+          if (unlikely(any(validtt))) 
+          {
+            t_o.lower = select(validtt, select(C <= 0.0f, vfloat<N>(neg_inf), vfloat<N>(pos_inf)), t_o.lower);
+            t_o.upper = select(validtt, select(C <= 0.0f, vfloat<N>(pos_inf), vfloat<N>(neg_inf)), t_o.upper);
+            valid &= !validtt | C <= 0.0f;
+          }
+
+          /* cone case */
+          if (any(validtf)) 
+          {
+            /* if we hit the negative cone there cannot be a hit */
+            const vfloat<N> t = -C/B;
+            const vfloat<N> z0 = Oz+t*dOz;
+            const vfloat<N> z0r = r0+z0*dr;
+            valid &= !validtf | z0r >= 0.0f;
+
+            /* test if we start inside or outside the cone */
+            t_o.lower = select(validtf, select(dOz*dr > 0.0f, t, vfloat<N>(neg_inf)), t_o.lower);
+            t_o.upper = select(validtf, select(dOz*dr > 0.0f, vfloat<N>(pos_inf), t), t_o.upper);
+          }
+        }
+
+        /* standard case for "non-parallel" rays */
+        if (likely(any(validf)))
+        {
+          const vfloat<N> Q = sqrt(D);
+          const vfloat<N> rcp_2A = 0.5f*rcp(A);
+          t_o.lower = select(validf, (-B-Q)*rcp_2A, t_o.lower);
+          t_o.upper = select(validf, (-B+Q)*rcp_2A, t_o.upper);
+          
+          /* standard case where both hits are on same cone */
+          const vbool<N> validft = validf &   A>0.0f;
+          const vbool<N> validff = validf & !(A>0.0f);
+          if (any(validft)) {
+            const vfloat<N> z0 = Oz+t_o.lower*dOz;
+            const vfloat<N> z0r = r0+z0*dr;
+            valid &= !validft | z0r >= 0.0f;
+          } 
+
+          /* special case where the hits are on the positive and negative cone */
+          if (any(validff)) {
+            /* depending on the ray direction and the open direction
+             * of the cone we have a hit from inside or outside the
+             * cone */
+            t_o.lower = select(validff, select(dOz*dr > 0.0f, t_o.lower, float(neg_inf)), t_o.lower);
+            t_o.upper = select(validff, select(dOz*dr > 0.0f, float(pos_inf), t_o.upper), t_o.upper);
+          }
+        }
+
+        /* calculates u and Ng for near hit */
+        {
+          u0_o = (Oz+t_o.lower*dOz)*rl;
+          const Vec3vfN Pr = t_o.lower*Vec3vfN(dir);
+          const Vec3vfN Pl = v0 + u0_o*(v1-v0);
+          const Vec3vfN R = normalize(Pr-Pl);
+          const Vec3vfN U = (p1-p0)+(r1-r0)*R;
+          const Vec3vfN V = cross(p1-p0,R);
+          Ng0_o = cross(V,U);
+        }
+
+        /* calculates u and Ng for far hit */
+        {
+          u1_o = (Oz+t_o.upper*dOz)*rl;
+          const Vec3vfN Pr = t_o.lower*Vec3vfN(dir);
+          const Vec3vfN Pl = v0 + u1_o*(v1-v0);
+          const Vec3vfN R = normalize(Pr-Pl);
+          const Vec3vfN U = (p1-p0)+(r1-r0)*R;
+          const Vec3vfN V = cross(p1-p0,R);
+          Ng1_o = cross(V,U);
+        }
+        return valid;
+      }
+ 
+      __forceinline vbool<N> intersect(const Vec3fa& org, const Vec3fa& dir, BBox<vfloat<N>>& t_o) const
+      {
+        vfloat<N> u0_o; Vec3vfN Ng0_o; vfloat<N> u1_o; Vec3vfN Ng1_o;
+        return intersect(org,dir,t_o,u0_o,Ng0_o,u1_o,Ng1_o);
+      }
+    };
+  }
+}
+

thirdparty/embree/kernels/geometry/coneline_intersector.h

@@ -0,0 +1,209 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    namespace __coneline_internal 
+    {
+      template<int M, typename Epilog, typename ray_tfar_func>
+        static __forceinline bool intersectCone(const vbool<M>& valid_i,
+                                                const Vec3vf<M>& ray_org_in, const Vec3vf<M>& ray_dir, 
+                                                const vfloat<M>& ray_tnear, const ray_tfar_func& ray_tfar,
+                                                const Vec4vf<M>& v0, const Vec4vf<M>& v1,
+                                                const vbool<M>& cL, const vbool<M>& cR,
+                                                const Epilog& epilog)
+      {   
+        vbool<M> valid = valid_i;
+
+        /* move ray origin closer to make calculations numerically stable */
+        const vfloat<M> dOdO = sqr(ray_dir);
+        const vfloat<M> rcp_dOdO = rcp(dOdO);
+        const Vec3vf<M> center = vfloat<M>(0.5f)*(v0.xyz()+v1.xyz());
+        const vfloat<M> dt = dot(center-ray_org_in,ray_dir)*rcp_dOdO;
+        const Vec3vf<M> ray_org = ray_org_in + dt*ray_dir;
+
+        const Vec3vf<M> dP = v1.xyz() - v0.xyz();
+        const Vec3vf<M> p0 = ray_org - v0.xyz();
+        const Vec3vf<M> p1 = ray_org - v1.xyz();
+        
+        const vfloat<M> dPdP  = sqr(dP);
+        const vfloat<M> dP0   = dot(p0,dP);
+        const vfloat<M> dP1   = dot(p1,dP); 
+        const vfloat<M> dOdP  = dot(ray_dir,dP);
+
+        // intersect cone body
+        const vfloat<M> dr  = v0.w - v1.w;
+        const vfloat<M> hy  = dPdP + sqr(dr);
+        const vfloat<M> dO0 = dot(ray_dir,p0);
+        const vfloat<M> OO  = sqr(p0);
+        const vfloat<M> dPdP2 = sqr(dPdP);
+        const vfloat<M> dPdPr0 = dPdP*v0.w;
+        
+        const vfloat<M> A = dPdP2     - sqr(dOdP)*hy;
+        const vfloat<M> B = dPdP2*dO0 - dP0*dOdP*hy   + dPdPr0*(dr*dOdP);
+        const vfloat<M> C = dPdP2*OO  - sqr(dP0)*hy   + dPdPr0*(2.0f*dr*dP0 - dPdPr0);
+        
+        const vfloat<M> D = B*B - A*C;
+        valid &= D >= 0.0f;
+        if (unlikely(none(valid))) {
+          return false;
+        }
+
+        /* standard case for "non-parallel" rays */
+        const vfloat<M> Q = sqrt(D);
+        const vfloat<M> rcp_A = rcp(A);
+        /* special case for rays that are "parallel" to the cone - assume miss */
+        const vbool<M> isParallel = abs(A) <= min_rcp_input;
+
+        vfloat<M> t_cone_lower = select (isParallel, neg_inf, (-B-Q)*rcp_A);
+        vfloat<M> t_cone_upper = select (isParallel, pos_inf, (-B+Q)*rcp_A);
+        const vfloat<M> y_lower = dP0 + t_cone_lower*dOdP;
+        const vfloat<M> y_upper = dP0 + t_cone_upper*dOdP;
+        t_cone_lower = select(valid & y_lower > 0.0f & y_lower < dPdP, t_cone_lower, pos_inf);
+        t_cone_upper = select(valid & y_upper > 0.0f & y_upper < dPdP, t_cone_upper, neg_inf);
+
+        const vbool<M> hitDisk0 = valid & cL;
+        const vbool<M> hitDisk1 = valid & cR;
+        const vfloat<M> rcp_dOdP = rcp(dOdP);
+        const vfloat<M> t_disk0 = select (hitDisk0, select (sqr(p0*dOdP-ray_dir*dP0)<(sqr(v0.w)*sqr(dOdP)), -dP0*rcp_dOdP, pos_inf), pos_inf);
+        const vfloat<M> t_disk1 = select (hitDisk1, select (sqr(p1*dOdP-ray_dir*dP1)<(sqr(v1.w)*sqr(dOdP)), -dP1*rcp_dOdP, pos_inf), pos_inf);
+        const vfloat<M> t_disk_lower = min(t_disk0, t_disk1);
+        const vfloat<M> t_disk_upper = max(t_disk0, t_disk1);
+
+        const vfloat<M> t_lower = min(t_cone_lower, t_disk_lower);
+        const vfloat<M> t_upper = max(t_cone_upper, select(t_lower==t_disk_lower, 
+                                                      select(t_disk_upper==vfloat<M>(pos_inf),neg_inf,t_disk_upper), 
+                                                      select(t_disk_lower==vfloat<M>(pos_inf),neg_inf,t_disk_lower)));
+
+        const vbool<M> valid_lower = valid & ray_tnear <= dt+t_lower & dt+t_lower <= ray_tfar() & t_lower != vfloat<M>(pos_inf);
+        const vbool<M> valid_upper = valid & ray_tnear <= dt+t_upper & dt+t_upper <= ray_tfar() & t_upper != vfloat<M>(neg_inf);
+
+        const vbool<M> valid_first = valid_lower | valid_upper;
+        if (unlikely(none(valid_first)))
+          return false;
+
+        const vfloat<M> t_first = select(valid_lower, t_lower, t_upper);
+        const vfloat<M> y_first = select(valid_lower, y_lower, y_upper);
+
+        const vfloat<M> rcp_dPdP = rcp(dPdP);
+        const Vec3vf<M> dP2drr0dP = dPdP*dr*v0.w*dP;
+        const Vec3vf<M> dPhy = dP*hy;
+        const vbool<M> cone_hit_first = valid & (t_first == t_cone_lower | t_first == t_cone_upper);
+        const vbool<M> disk0_hit_first = valid & (t_first == t_disk0);
+        const Vec3vf<M> Ng_first = select(cone_hit_first, dPdP2*(p0+t_first*ray_dir)+dP2drr0dP-dPhy*y_first, select(disk0_hit_first, -dP, dP));
+        const vfloat<M> u_first = select(cone_hit_first, y_first*rcp_dPdP, select(disk0_hit_first, vfloat<M>(zero), vfloat<M>(one)));
+
+        /* invoke intersection filter for first hit */
+        RoundLineIntersectorHitM<M> hit(u_first,zero,dt+t_first,Ng_first);
+        const bool is_hit_first = epilog(valid_first, hit);
+
+        /* check for possible second hits before potentially accepted hit */
+        const vfloat<M> t_second = t_upper;
+        const vfloat<M> y_second = y_upper;
+        const vbool<M> valid_second = valid_lower & valid_upper & (dt+t_upper <= ray_tfar());
+        if (unlikely(none(valid_second)))
+          return is_hit_first;
+        
+        /* invoke intersection filter for second hit */
+        const vbool<M> cone_hit_second = t_second == t_cone_lower | t_second == t_cone_upper;
+        const vbool<M> disk0_hit_second = t_second == t_disk0;
+        const Vec3vf<M> Ng_second = select(cone_hit_second, dPdP2*(p0+t_second*ray_dir)+dP2drr0dP-dPhy*y_second, select(disk0_hit_second, -dP, dP));
+        const vfloat<M> u_second = select(cone_hit_second, y_second*rcp_dPdP, select(disk0_hit_first, vfloat<M>(zero), vfloat<M>(one)));
+
+        hit = RoundLineIntersectorHitM<M>(u_second,zero,dt+t_second,Ng_second);
+        const bool is_hit_second = epilog(valid_second, hit);
+        
+        return is_hit_first | is_hit_second;
+      }
+    }
+
+    template<int M>
+      struct ConeLineIntersectorHitM
+      {
+        __forceinline ConeLineIntersectorHitM() {}
+        
+        __forceinline ConeLineIntersectorHitM(const vfloat<M>& u, const vfloat<M>& v, const vfloat<M>& t, const Vec3vf<M>& Ng)
+          : vu(u), vv(v), vt(t), vNg(Ng) {}
+	
+        __forceinline void finalize() {}
+	
+        __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+        __forceinline float t  (const size_t i) const { return vt[i]; }
+        __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+	
+      public:
+        vfloat<M> vu;
+        vfloat<M> vv;
+        vfloat<M> vt;
+        Vec3vf<M> vNg;
+      };
+    
+    template<int M>
+      struct ConeCurveIntersector1
+      {
+        typedef CurvePrecalculations1 Precalculations;
+        
+        struct ray_tfar {
+          Ray& ray;
+          __forceinline ray_tfar(Ray& ray) : ray(ray) {}
+          __forceinline vfloat<M> operator() () const { return ray.tfar; };
+        };
+
+        template<typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            Ray& ray,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const vbool<M>& cL, const vbool<M>& cR,
+                                            const Epilog& epilog)
+        {
+          const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+          const Vec3vf<M> ray_dir(ray.dir.x, ray.dir.y, ray.dir.z);
+          const vfloat<M> ray_tnear(ray.tnear());
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          return  __coneline_internal::intersectCone<M>(valid_i,ray_org,ray_dir,ray_tnear,ray_tfar(ray),v0,v1,cL,cR,epilog);
+        }
+      };
+    
+    template<int M, int K>
+      struct ConeCurveIntersectorK
+      {
+        typedef CurvePrecalculationsK<K> Precalculations;
+        
+        struct ray_tfar {
+          RayK<K>& ray;
+          size_t k;
+          __forceinline ray_tfar(RayK<K>& ray, size_t k) : ray(ray), k(k) {}
+          __forceinline vfloat<M> operator() () const { return ray.tfar[k]; };
+        };
+        
+        template<typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            RayK<K>& ray, size_t k,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const vbool<M>& cL, const vbool<M>& cR,
+                                            const Epilog& epilog)
+        {
+          const Vec3vf<M> ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const Vec3vf<M> ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]);
+          const vfloat<M> ray_tnear = ray.tnear()[k];
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          return __coneline_internal::intersectCone<M>(valid_i,ray_org,ray_dir,ray_tnear,ray_tfar(ray,k),v0,v1,cL,cR,epilog);
+        }
+      };
+  }
+}

thirdparty/embree/kernels/geometry/conelinei_intersector.h

@@ -0,0 +1,141 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "coneline_intersector.h"
+#include "intersector_epilog.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, bool filter>
+    struct ConeCurveMiIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom);
+        const vbool<M> valid = line.valid();
+        ConeCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,cL,cR,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom);
+        const vbool<M> valid = line.valid();
+        return ConeCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,cL,cR,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+        return false;
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, bool filter>
+    struct ConeCurveMiMBIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        ConeCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,cL,cR,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        return ConeCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,cL,cR,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+        return false;
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct ConeCurveMiIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom);
+        const vbool<M> valid = line.valid();
+        ConeCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,cL,cR,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom);
+        const vbool<M> valid = line.valid();
+        return ConeCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,cL,cR,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct ConeCurveMiMBIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context,  const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        ConeCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,cL,cR,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; 
+        vbool<M> cL,cR;
+        line.gather(v0,v1,cL,cR,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        return ConeCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,cL,cR,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curveNi.h

@@ -0,0 +1,222 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  template<int M>
+    struct CurveNi
+  {
+    struct Type : public PrimitiveType {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+M-1)/M; }
+
+    static __forceinline size_t bytes(size_t N)
+    {
+      const size_t f = N/M, r = N%M;
+      static_assert(sizeof(CurveNi) == 22+25*M, "internal data layout issue");
+      return f*sizeof(CurveNi) + (r!=0)*(22 + 25*r);
+    }
+
+  public:
+
+    /*! Default constructor. */
+    __forceinline CurveNi () {}
+
+    /*! fill curve from curve list */
+    __forceinline void fill(const PrimRef* prims, size_t& begin, size_t _end, Scene* scene)
+    {  
+      size_t end = min(begin+M,_end);
+      N = (unsigned char)(end-begin);
+      const unsigned int geomID0 = prims[begin].geomID();
+      this->geomID(N) = geomID0;
+      ty = (unsigned char) scene->get(geomID0)->getType();
+
+      /* encode all primitives */
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<N; i++)
+      {
+        const PrimRef& prim = prims[begin+i];
+        const unsigned int geomID = prim.geomID(); assert(geomID == geomID0);
+        const unsigned int primID = prim.primID();
+        bounds.extend(scene->get(geomID)->vbounds(primID));
+      }
+
+      /* calculate offset and scale */
+      Vec3fa loffset = bounds.lower;
+      float lscale = reduce_min(256.0f/(bounds.size()*sqrt(3.0f)));
+      if (bounds.size() == Vec3fa(zero)) lscale = 0.0f;
+      *this->offset(N) = loffset;
+      *this->scale(N) = lscale;
+      
+      /* encode all primitives */
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+        const PrimRef& prim = prims[begin];
+        const unsigned int geomID = prim.geomID();
+        const unsigned int primID = prim.primID();
+        const LinearSpace3fa space2 = scene->get(geomID)->computeAlignedSpace(primID);
+        
+        const LinearSpace3fa space3(trunc(126.0f*space2.vx),trunc(126.0f*space2.vy),trunc(126.0f*space2.vz));
+        const BBox3fa bounds = scene->get(geomID)->vbounds(loffset,lscale,max(length(space3.vx),length(space3.vy),length(space3.vz)),space3.transposed(),primID);
+        
+        bounds_vx_x(N)[i] = (char) space3.vx.x;
+        bounds_vx_y(N)[i] = (char) space3.vx.y;
+        bounds_vx_z(N)[i] = (char) space3.vx.z;
+        bounds_vx_lower(N)[i] = (short) clamp(floor(bounds.lower.x),-32767.0f,32767.0f);
+        bounds_vx_upper(N)[i] = (short) clamp(ceil (bounds.upper.x),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.lower.x) && floor(bounds.lower.x) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.upper.x) && ceil (bounds.upper.x) <= 32767.0f);
+
+        bounds_vy_x(N)[i] = (char) space3.vy.x;
+        bounds_vy_y(N)[i] = (char) space3.vy.y;
+        bounds_vy_z(N)[i] = (char) space3.vy.z;
+        bounds_vy_lower(N)[i] = (short) clamp(floor(bounds.lower.y),-32767.0f,32767.0f);
+        bounds_vy_upper(N)[i] = (short) clamp(ceil (bounds.upper.y),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.lower.y) && floor(bounds.lower.y) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.upper.y) && ceil (bounds.upper.y) <= 32767.0f);
+
+        bounds_vz_x(N)[i] = (char) space3.vz.x;
+        bounds_vz_y(N)[i] = (char) space3.vz.y;
+        bounds_vz_z(N)[i] = (char) space3.vz.z;
+        bounds_vz_lower(N)[i] = (short) clamp(floor(bounds.lower.z),-32767.0f,32767.0f);
+        bounds_vz_upper(N)[i] = (short) clamp(ceil (bounds.upper.z),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.lower.z) && floor(bounds.lower.z) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.upper.z) && ceil (bounds.upper.z) <= 32767.0f);
+               
+        this->primID(N)[i] = primID;
+      }
+    }
+
+    template<typename BVH, typename Allocator>
+      __forceinline static typename BVH::NodeRef createLeaf (BVH* bvh, const PrimRef* prims, const range<size_t>& set, const Allocator& alloc)
+    {
+      size_t start = set.begin();
+      size_t items = CurveNi::blocks(set.size());
+      size_t numbytes = CurveNi::bytes(set.size());
+      CurveNi* accel = (CurveNi*) alloc.malloc1(numbytes,BVH::byteAlignment);
+      for (size_t i=0; i<items; i++) {
+        accel[i].fill(prims,start,set.end(),bvh->scene);
+      }
+      return bvh->encodeLeaf((char*)accel,items);
+    };
+    
+  public:
+    
+    // 27.6 - 46 bytes per primitive
+    unsigned char ty;
+    unsigned char N;
+    unsigned char data[4+25*M+16];
+
+    /*
+    struct Layout
+    {
+      unsigned int geomID;
+      unsigned int primID[N];
+      
+      char bounds_vx_x[N];
+      char bounds_vx_y[N];
+      char bounds_vx_z[N];
+      short bounds_vx_lower[N];
+      short bounds_vx_upper[N];
+      
+      char bounds_vy_x[N];
+      char bounds_vy_y[N];
+      char bounds_vy_z[N];
+      short bounds_vy_lower[N];
+      short bounds_vy_upper[N];
+      
+      char bounds_vz_x[N];
+      char bounds_vz_y[N];
+      char bounds_vz_z[N];
+      short bounds_vz_lower[N];
+      short bounds_vz_upper[N];
+      
+      Vec3f offset;
+      float scale;
+    };
+    */
+    
+    __forceinline       unsigned int& geomID(size_t N)       { return *(unsigned int*)((char*)this+2); }
+    __forceinline const unsigned int& geomID(size_t N) const { return *(unsigned int*)((char*)this+2); }
+    
+    __forceinline       unsigned int* primID(size_t N)       { return (unsigned int*)((char*)this+6); }
+    __forceinline const unsigned int* primID(size_t N) const { return (unsigned int*)((char*)this+6); }
+    
+    __forceinline       char* bounds_vx_x(size_t N)       { return (char*)((char*)this+6+4*N); }
+    __forceinline const char* bounds_vx_x(size_t N) const { return (char*)((char*)this+6+4*N); }
+    
+    __forceinline       char* bounds_vx_y(size_t N)       { return (char*)((char*)this+6+5*N); }
+    __forceinline const char* bounds_vx_y(size_t N) const { return (char*)((char*)this+6+5*N); }
+    
+    __forceinline       char* bounds_vx_z(size_t N)       { return (char*)((char*)this+6+6*N); }
+    __forceinline const char* bounds_vx_z(size_t N) const { return (char*)((char*)this+6+6*N); }
+    
+    __forceinline       short* bounds_vx_lower(size_t N)       { return (short*)((char*)this+6+7*N); }
+    __forceinline const short* bounds_vx_lower(size_t N) const { return (short*)((char*)this+6+7*N); }
+    
+    __forceinline       short* bounds_vx_upper(size_t N)       { return (short*)((char*)this+6+9*N); }
+    __forceinline const short* bounds_vx_upper(size_t N) const { return (short*)((char*)this+6+9*N); }
+    
+    __forceinline       char* bounds_vy_x(size_t N)       { return (char*)((char*)this+6+11*N); }
+    __forceinline const char* bounds_vy_x(size_t N) const { return (char*)((char*)this+6+11*N); }
+    
+    __forceinline       char* bounds_vy_y(size_t N)       { return (char*)((char*)this+6+12*N); }
+    __forceinline const char* bounds_vy_y(size_t N) const { return (char*)((char*)this+6+12*N); }
+    
+    __forceinline       char* bounds_vy_z(size_t N)       { return (char*)((char*)this+6+13*N); }
+    __forceinline const char* bounds_vy_z(size_t N) const { return (char*)((char*)this+6+13*N); }
+    
+    __forceinline       short* bounds_vy_lower(size_t N)       { return (short*)((char*)this+6+14*N); }
+    __forceinline const short* bounds_vy_lower(size_t N) const { return (short*)((char*)this+6+14*N); }
+    
+    __forceinline       short* bounds_vy_upper(size_t N)       { return (short*)((char*)this+6+16*N); }
+    __forceinline const short* bounds_vy_upper(size_t N) const { return (short*)((char*)this+6+16*N); }
+    
+    __forceinline       char* bounds_vz_x(size_t N)       { return (char*)((char*)this+6+18*N); }
+    __forceinline const char* bounds_vz_x(size_t N) const { return (char*)((char*)this+6+18*N); }
+    
+    __forceinline       char* bounds_vz_y(size_t N)       { return (char*)((char*)this+6+19*N); }
+    __forceinline const char* bounds_vz_y(size_t N) const { return (char*)((char*)this+6+19*N); }
+    
+    __forceinline       char* bounds_vz_z(size_t N)       { return (char*)((char*)this+6+20*N); }
+    __forceinline const char* bounds_vz_z(size_t N) const { return (char*)((char*)this+6+20*N); }
+    
+    __forceinline       short* bounds_vz_lower(size_t N)       { return (short*)((char*)this+6+21*N); }
+    __forceinline const short* bounds_vz_lower(size_t N) const { return (short*)((char*)this+6+21*N); }
+    
+    __forceinline       short* bounds_vz_upper(size_t N)       { return (short*)((char*)this+6+23*N); }
+    __forceinline const short* bounds_vz_upper(size_t N) const { return (short*)((char*)this+6+23*N); }
+    
+    __forceinline       Vec3f* offset(size_t N)       { return (Vec3f*)((char*)this+6+25*N); }
+    __forceinline const Vec3f* offset(size_t N) const { return (Vec3f*)((char*)this+6+25*N); }
+    
+    __forceinline       float* scale(size_t N)       { return (float*)((char*)this+6+25*N+12); }
+    __forceinline const float* scale(size_t N) const { return (float*)((char*)this+6+25*N+12); }
+
+    __forceinline       char* end(size_t N)       { return (char*)this+6+25*N+16; }
+    __forceinline const char* end(size_t N) const { return (char*)this+6+25*N+16; }
+  };
+
+  template<int M>
+    typename CurveNi<M>::Type CurveNi<M>::type;
+
+  typedef CurveNi<4> Curve4i;
+  typedef CurveNi<8> Curve8i;
+}

thirdparty/embree/kernels/geometry/curveNi_intersector.h

@@ -0,0 +1,648 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "curveNi.h"
+
+#include "roundline_intersector.h"
+#include "coneline_intersector.h"
+#include "curve_intersector_ribbon.h"
+#include "curve_intersector_oriented.h"
+#include "curve_intersector_sweep.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+    struct CurveNiIntersector1
+    {
+      typedef CurveNi<M> Primitive;
+      typedef Vec3vf<M> Vec3vfM;
+      typedef LinearSpace3<Vec3vfM>LinearSpace3vfM;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline vbool<M> intersect(Ray& ray, const Primitive& prim, vfloat<M>& tNear_o)
+      {
+        const size_t N = prim.N;
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+        const Vec3fa offset = *prim.offset(N);
+        const float scale  = *prim.scale(N);
+#else
+        const vfloat4 offset_scale = vfloat4::loadu(prim.offset(N));
+        const Vec3fa offset = Vec3fa(offset_scale);
+        const Vec3fa scale = Vec3fa(shuffle<3,3,3,3>(offset_scale));
+#endif
+        const Vec3fa org1 = (ray.org-offset)*scale;
+        const Vec3fa dir1 = ray.dir*scale;
+        
+        const LinearSpace3vfM space(vfloat<M>::load(prim.bounds_vx_x(N)), vfloat<M>::load(prim.bounds_vx_y(N)), vfloat<M>::load(prim.bounds_vx_z(N)),
+                                    vfloat<M>::load(prim.bounds_vy_x(N)), vfloat<M>::load(prim.bounds_vy_y(N)), vfloat<M>::load(prim.bounds_vy_z(N)),
+                                    vfloat<M>::load(prim.bounds_vz_x(N)), vfloat<M>::load(prim.bounds_vz_y(N)), vfloat<M>::load(prim.bounds_vz_z(N)));
+
+        const Vec3vfM dir2 = xfmVector(space,Vec3vfM(dir1));
+        const Vec3vfM org2 = xfmPoint (space,Vec3vfM(org1));
+        const Vec3vfM rcp_dir2 = rcp_safe(dir2);
+       
+        const vfloat<M> t_lower_x = (vfloat<M>::load(prim.bounds_vx_lower(N))-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_upper_x = (vfloat<M>::load(prim.bounds_vx_upper(N))-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_lower_y = (vfloat<M>::load(prim.bounds_vy_lower(N))-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_upper_y = (vfloat<M>::load(prim.bounds_vy_upper(N))-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_lower_z = (vfloat<M>::load(prim.bounds_vz_lower(N))-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+        const vfloat<M> t_upper_z = (vfloat<M>::load(prim.bounds_vz_upper(N))-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+
+        const vfloat<M> round_up  (1.0f+3.0f*float(ulp));
+        const vfloat<M> round_down(1.0f-3.0f*float(ulp));
+        const vfloat<M> tNear = round_down*max(mini(t_lower_x,t_upper_x),mini(t_lower_y,t_upper_y),mini(t_lower_z,t_upper_z),vfloat<M>(ray.tnear()));
+        const vfloat<M> tFar  = round_up  *min(maxi(t_lower_x,t_upper_x),maxi(t_lower_y,t_upper_y),maxi(t_lower_z,t_upper_z),vfloat<M>(ray.tfar));
+        tNear_o = tNear;
+        return (vint<M>(step) < vint<M>(prim.N)) & (tNear <= tFar);
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID));
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+          
+          Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID));
+         
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+
+          if (Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_n(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          
+          unsigned int vertexID = geom->curve(primID);
+          Vec3ff a0,a1,a2,a3; Vec3fa n0,n1,n2,n3; geom->gather(a0,a1,a2,a3,n0,n1,n2,n3,vertexID);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+          
+          Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,n0,n1,n2,n3,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_n(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+
+          unsigned int vertexID = geom->curve(primID);
+          Vec3ff a0,a1,a2,a3; Vec3fa n0,n1,n2,n3; geom->gather(a0,a1,a2,a3,n0,n1,n2,n3,vertexID);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+
+          if (Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,n0,n1,n2,n3,Epilog(ray,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_h(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID));
+          Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_h(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID));
+          if (Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,Epilog(ray,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_hn(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; Vec3fa n0,dn0,n1,dn1; geom->gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,geom->curve(primID));
+          Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,n0,dn0,n1,dn1,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_hn(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; Vec3fa n0,dn0,n1,dn1; geom->gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,geom->curve(primID));
+          if (Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,n0,dn0,n1,dn1,Epilog(ray,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+    };
+
+    template<int M, int K>
+      struct CurveNiIntersectorK
+    {
+      typedef CurveNi<M> Primitive;
+      typedef Vec3vf<M> Vec3vfM;
+      typedef LinearSpace3<Vec3vfM>LinearSpace3vfM;
+      typedef CurvePrecalculationsK<K> Precalculations;
+      
+      static __forceinline vbool<M> intersect(RayK<K>& ray, const size_t k, const Primitive& prim, vfloat<M>& tNear_o)
+      {
+        const size_t N = prim.N;
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+        const Vec3fa offset = *prim.offset(N);
+        const float scale  = *prim.scale(N);
+#else
+        const vfloat4 offset_scale = vfloat4::loadu(prim.offset(N));
+        const Vec3fa offset = Vec3fa(offset_scale);
+        const Vec3fa scale = Vec3fa(shuffle<3,3,3,3>(offset_scale));
+#endif
+        const Vec3fa ray_org(ray.org.x[k],ray.org.y[k],ray.org.z[k]);
+        const Vec3fa ray_dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]);
+        const Vec3fa org1 = (ray_org-offset)*scale;
+        const Vec3fa dir1 = ray_dir*scale;
+        
+        const LinearSpace3vfM space(vfloat<M>::load(prim.bounds_vx_x(N)), vfloat<M>::load(prim.bounds_vx_y(N)), vfloat<M>::load(prim.bounds_vx_z(N)),
+                                    vfloat<M>::load(prim.bounds_vy_x(N)), vfloat<M>::load(prim.bounds_vy_y(N)), vfloat<M>::load(prim.bounds_vy_z(N)),
+                                    vfloat<M>::load(prim.bounds_vz_x(N)), vfloat<M>::load(prim.bounds_vz_y(N)), vfloat<M>::load(prim.bounds_vz_z(N)));
+
+        const Vec3vfM dir2 = xfmVector(space,Vec3vfM(dir1));
+        const Vec3vfM org2 = xfmPoint (space,Vec3vfM(org1));
+        const Vec3vfM rcp_dir2 = rcp_safe(dir2);
+       
+        const vfloat<M> t_lower_x = (vfloat<M>::load(prim.bounds_vx_lower(N))-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_upper_x = (vfloat<M>::load(prim.bounds_vx_upper(N))-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_lower_y = (vfloat<M>::load(prim.bounds_vy_lower(N))-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_upper_y = (vfloat<M>::load(prim.bounds_vy_upper(N))-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_lower_z = (vfloat<M>::load(prim.bounds_vz_lower(N))-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+        const vfloat<M> t_upper_z = (vfloat<M>::load(prim.bounds_vz_upper(N))-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+
+        const vfloat<M> round_up  (1.0f+3.0f*float(ulp));
+        const vfloat<M> round_down(1.0f-3.0f*float(ulp));
+        const vfloat<M> tNear = round_down*max(mini(t_lower_x,t_upper_x),mini(t_lower_y,t_upper_y),mini(t_lower_z,t_upper_z),vfloat<M>(ray.tnear()[k]));
+        const vfloat<M> tFar  = round_up  *min(maxi(t_lower_x,t_upper_x),maxi(t_lower_y,t_upper_y),maxi(t_lower_z,t_upper_z),vfloat<M>(ray.tfar[k]));
+        tNear_o = tNear;
+        return (vint<M>(step) < vint<M>(prim.N)) & (tNear <= tFar);
+      }
+      
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID));
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+
+          Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+      
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID));
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+          
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_n(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+
+          unsigned int vertexID = geom->curve(primID);
+          Vec3ff a0,a1,a2,a3; Vec3fa n0,n1,n2,n3; geom->gather(a0,a1,a2,a3,n0,n1,n2,n3,vertexID);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+
+          Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,n0,n1,n2,n3,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+      
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_n(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+
+          unsigned int vertexID = geom->curve(primID);
+          Vec3ff a0,a1,a2,a3; Vec3fa n0,n1,n2,n3; geom->gather(a0,a1,a2,a3,n0,n1,n2,n3,vertexID);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            const unsigned int primID1 = prim.primID(N)[i1];
+            geom->prefetchL1_vertices(geom->curve(primID1));
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              const unsigned int primID2 = prim.primID(N)[i2];
+              geom->prefetchL2_vertices(geom->curve(primID2));
+            }
+          }
+
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,n0,n1,n2,n3,Epilog(ray,k,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_h(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID));
+          Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+      
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_h(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID));
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,Epilog(ray,k,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_hn(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; Vec3fa n0,dn0,n1,dn1; geom->gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,geom->curve(primID));
+          Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,n0,dn0,n1,dn1,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+      
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_hn(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; Vec3fa n0,dn0,n1,dn1; geom->gather_hermite(p0,t0,n0,dn0,p1,t1,n1,dn1,geom->curve(primID));
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,n0,dn0,n1,dn1,Epilog(ray,k,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+    };
+
+     __forceinline void convert_to_bezier(const Geometry::GType gtype,
+                                         Vec3ff& v0, Vec3ff& v1, Vec3ff& v2, Vec3ff& v3,
+                                         Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3)
+    {
+      const Geometry::GType basis = (Geometry::GType)(gtype & Geometry::GTY_BASIS_MASK);
+      const Geometry::GType stype = (Geometry::GType)(gtype & Geometry::GTY_SUBTYPE_MASK);
+      
+      if (basis == Geometry::GTY_BASIS_BSPLINE) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(BSplineCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+      else if (basis == Geometry::GTY_BASIS_HERMITE) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(HermiteCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+      else if (basis == Geometry::GTY_BASIS_CATMULL_ROM) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(CatmullRomCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+
+      if (stype == Geometry::GTY_SUBTYPE_ORIENTED_CURVE)
+      {
+        if (basis == Geometry::GTY_BASIS_BSPLINE) {
+          BezierCurveT<Vec3fa> bezier;
+          convert(BSplineCurveT<Vec3fa>(n0,n1,n2,n3),bezier);
+          n0 = bezier.v0; n1 = bezier.v1; n2 = bezier.v2; n3 = bezier.v3;
+        }
+        else if (basis == Geometry::GTY_BASIS_HERMITE) {
+          BezierCurveT<Vec3fa> bezier;
+          convert(HermiteCurveT<Vec3fa>(n0,n1,n2,n3),bezier);
+          n0 = bezier.v0; n1 = bezier.v1; n2 = bezier.v2; n3 = bezier.v3;
+        }
+        else if (basis == Geometry::GTY_BASIS_CATMULL_ROM) {
+          BezierCurveT<Vec3fa> bezier;
+          convert(CatmullRomCurveT<Vec3fa>(n0,n1,n2,n3),bezier);
+          n0 = bezier.v0; n1 = bezier.v1; n2 = bezier.v2; n3 = bezier.v3;
+        }
+      }
+    }
+
+    __forceinline void convert_to_bezier(const Geometry::GType gtype, Vec3ff& v0, Vec3ff& v1, Vec3ff& v2, Vec3ff& v3)
+    {
+      const Geometry::GType basis = (Geometry::GType)(gtype & Geometry::GTY_BASIS_MASK);
+      
+      if (basis == Geometry::GTY_BASIS_BSPLINE) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(BSplineCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+      else if (basis == Geometry::GTY_BASIS_HERMITE) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(HermiteCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+      else if (basis == Geometry::GTY_BASIS_CATMULL_ROM) {
+        BezierCurveT<Vec3ff> bezier;
+        convert(CatmullRomCurveT<Vec3ff>(v0,v1,v2,v3),bezier);
+        v0 = bezier.v0; v1 = bezier.v1; v2 = bezier.v2; v3 = bezier.v3;
+      }
+    }
+  }
+}

thirdparty/embree/kernels/geometry/curveNi_mb.h

@@ -0,0 +1,278 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  template<int M>
+    struct CurveNiMB
+  {
+    struct Type : public PrimitiveType {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+M-1)/M; }
+
+    static __forceinline size_t bytes(size_t N)
+    {
+      const size_t f = N/M, r = N%M;
+      static_assert(sizeof(CurveNiMB) == 6+37*M+24, "internal data layout issue");
+      return f*sizeof(CurveNiMB) + (r!=0)*(6+37*r+24);
+    }
+
+  public:
+
+    /*! Default constructor. */
+    __forceinline CurveNiMB () {}
+
+    /*! fill curve from curve list */
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& begin, size_t _end, Scene* scene, const BBox1f time_range)
+    {
+      size_t end = min(begin+M,_end);
+      N = (unsigned char)(end-begin);
+      const unsigned int geomID0 = prims[begin].geomID();
+      this->geomID(N) = geomID0;
+      ty = (unsigned char) scene->get(geomID0)->getType();
+
+      /* encode all primitives */
+      LBBox3fa lbounds = empty;
+      for (size_t i=0; i<N; i++)
+      {
+        const PrimRefMB& prim = prims[begin+i];
+        const unsigned int geomID = prim.geomID(); assert(geomID == geomID0);
+        const unsigned int primID = prim.primID();
+        lbounds.extend(scene->get(geomID)->vlinearBounds(primID,time_range));
+      }
+      BBox3fa bounds = lbounds.bounds();
+
+      /* calculate offset and scale */
+      Vec3fa loffset = bounds.lower;
+      float lscale = reduce_min(256.0f/(bounds.size()*sqrt(3.0f)));
+      if (bounds.size() == Vec3fa(zero)) lscale = 0.0f;
+      *this->offset(N) = loffset;
+      *this->scale(N) = lscale;
+      this->time_offset(N) = time_range.lower;
+      this->time_scale(N) = 1.0f/time_range.size();
+      
+      /* encode all primitives */
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+        const PrimRefMB& prim = prims[begin];
+        const unsigned int geomID = prim.geomID();
+        const unsigned int primID = prim.primID();
+        const LinearSpace3fa space2 = scene->get(geomID)->computeAlignedSpaceMB(primID,time_range);
+        
+        const LinearSpace3fa space3(trunc(126.0f*space2.vx),trunc(126.0f*space2.vy),trunc(126.0f*space2.vz));
+        const LBBox3fa bounds = scene->get(geomID)->vlinearBounds(loffset,lscale,max(length(space3.vx),length(space3.vy),length(space3.vz)),space3.transposed(),primID,time_range);
+        
+        // NOTE: this weird (char) (short) cast works around VS2015 Win32 compiler bug
+        bounds_vx_x(N)[i] = (char) (short) space3.vx.x;
+        bounds_vx_y(N)[i] = (char) (short) space3.vx.y;
+        bounds_vx_z(N)[i] = (char) (short) space3.vx.z;
+        bounds_vx_lower0(N)[i] = (short) clamp(floor(bounds.bounds0.lower.x),-32767.0f,32767.0f);
+        bounds_vx_upper0(N)[i] = (short) clamp(ceil (bounds.bounds0.upper.x),-32767.0f,32767.0f);
+        bounds_vx_lower1(N)[i] = (short) clamp(floor(bounds.bounds1.lower.x),-32767.0f,32767.0f);
+        bounds_vx_upper1(N)[i] = (short) clamp(ceil (bounds.bounds1.upper.x),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds0.lower.x) && floor(bounds.bounds0.lower.x) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds0.upper.x) && ceil (bounds.bounds0.upper.x) <= 32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds1.lower.x) && floor(bounds.bounds1.lower.x) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds1.upper.x) && ceil (bounds.bounds1.upper.x) <= 32767.0f);
+        
+        bounds_vy_x(N)[i] = (char) (short) space3.vy.x;
+        bounds_vy_y(N)[i] = (char) (short) space3.vy.y;
+        bounds_vy_z(N)[i] = (char) (short) space3.vy.z;
+        bounds_vy_lower0(N)[i] = (short) clamp(floor(bounds.bounds0.lower.y),-32767.0f,32767.0f);
+        bounds_vy_upper0(N)[i] = (short) clamp(ceil (bounds.bounds0.upper.y),-32767.0f,32767.0f);
+        bounds_vy_lower1(N)[i] = (short) clamp(floor(bounds.bounds1.lower.y),-32767.0f,32767.0f);
+        bounds_vy_upper1(N)[i] = (short) clamp(ceil (bounds.bounds1.upper.y),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds0.lower.y) && floor(bounds.bounds0.lower.y) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds0.upper.y) && ceil (bounds.bounds0.upper.y) <= 32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds1.lower.y) && floor(bounds.bounds1.lower.y) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds1.upper.y) && ceil (bounds.bounds1.upper.y) <= 32767.0f);
+
+        bounds_vz_x(N)[i] = (char) (short) space3.vz.x;
+        bounds_vz_y(N)[i] = (char) (short) space3.vz.y;
+        bounds_vz_z(N)[i] = (char) (short) space3.vz.z;
+        bounds_vz_lower0(N)[i] = (short) clamp(floor(bounds.bounds0.lower.z),-32767.0f,32767.0f);
+        bounds_vz_upper0(N)[i] = (short) clamp(ceil (bounds.bounds0.upper.z),-32767.0f,32767.0f);
+        bounds_vz_lower1(N)[i] = (short) clamp(floor(bounds.bounds1.lower.z),-32767.0f,32767.0f);
+        bounds_vz_upper1(N)[i] = (short) clamp(ceil (bounds.bounds1.upper.z),-32767.0f,32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds0.lower.z) && floor(bounds.bounds0.lower.z) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds0.upper.z) && ceil (bounds.bounds0.upper.z) <= 32767.0f);
+        assert(-32767.0f <= floor(bounds.bounds1.lower.z) && floor(bounds.bounds1.lower.z) <= 32767.0f);
+        assert(-32767.0f <= ceil (bounds.bounds1.upper.z) && ceil (bounds.bounds1.upper.z) <= 32767.0f);
+               
+        this->primID(N)[i] = primID;
+      }
+      
+      return lbounds;
+    }
+
+    template<typename BVH, typename SetMB, typename Allocator>
+    __forceinline static typename BVH::NodeRecordMB4D createLeafMB(BVH* bvh, const SetMB& prims, const Allocator& alloc)
+    {
+      size_t start = prims.begin();
+      size_t end   = prims.end();
+      size_t items = CurveNiMB::blocks(prims.size());
+      size_t numbytes = CurveNiMB::bytes(prims.size());
+      CurveNiMB* accel = (CurveNiMB*) alloc.malloc1(numbytes,BVH::byteAlignment);
+      const typename BVH::NodeRef node = bvh->encodeLeaf((char*)accel,items);
+      
+      LBBox3fa bounds = empty;
+      for (size_t i=0; i<items; i++)
+        bounds.extend(accel[i].fillMB(prims.prims->data(),start,end,bvh->scene,prims.time_range));
+      
+      return typename BVH::NodeRecordMB4D(node,bounds,prims.time_range);
+    };
+
+    
+  public:
+    
+    // 27.6 - 46 bytes per primitive
+    unsigned char ty;
+    unsigned char N;
+    unsigned char data[4+37*M+24];
+
+    /*
+    struct Layout
+    {
+      unsigned int geomID;
+      unsigned int primID[N];
+      
+      char bounds_vx_x[N];
+      char bounds_vx_y[N];
+      char bounds_vx_z[N];
+      short bounds_vx_lower0[N];
+      short bounds_vx_upper0[N];
+      short bounds_vx_lower1[N];
+      short bounds_vx_upper1[N];
+      
+      char bounds_vy_x[N];
+      char bounds_vy_y[N];
+      char bounds_vy_z[N];
+      short bounds_vy_lower0[N];
+      short bounds_vy_upper0[N];
+      short bounds_vy_lower1[N];
+      short bounds_vy_upper1[N];
+      
+      char bounds_vz_x[N];
+      char bounds_vz_y[N];
+      char bounds_vz_z[N];
+      short bounds_vz_lower0[N];
+      short bounds_vz_upper0[N];
+      short bounds_vz_lower1[N];
+      short bounds_vz_upper1[N];
+      
+      Vec3f offset;
+      float scale;
+
+      float time_offset;
+      float time_scale;
+    };
+    */
+    
+    __forceinline       unsigned int& geomID(size_t N)       { return *(unsigned int*)((char*)this+2); }
+    __forceinline const unsigned int& geomID(size_t N) const { return *(unsigned int*)((char*)this+2); }
+    
+    __forceinline       unsigned int* primID(size_t N)       { return (unsigned int*)((char*)this+6); }
+    __forceinline const unsigned int* primID(size_t N) const { return (unsigned int*)((char*)this+6); }
+    
+    __forceinline       char* bounds_vx_x(size_t N)       { return (char*)((char*)this+6+4*N); }
+    __forceinline const char* bounds_vx_x(size_t N) const { return (char*)((char*)this+6+4*N); }
+    
+    __forceinline       char* bounds_vx_y(size_t N)       { return (char*)((char*)this+6+5*N); }
+    __forceinline const char* bounds_vx_y(size_t N) const { return (char*)((char*)this+6+5*N); }
+    
+    __forceinline       char* bounds_vx_z(size_t N)       { return (char*)((char*)this+6+6*N); }
+    __forceinline const char* bounds_vx_z(size_t N) const { return (char*)((char*)this+6+6*N); }
+    
+    __forceinline       short* bounds_vx_lower0(size_t N)       { return (short*)((char*)this+6+7*N); }
+    __forceinline const short* bounds_vx_lower0(size_t N) const { return (short*)((char*)this+6+7*N); }
+    
+    __forceinline       short* bounds_vx_upper0(size_t N)       { return (short*)((char*)this+6+9*N); }
+    __forceinline const short* bounds_vx_upper0(size_t N) const { return (short*)((char*)this+6+9*N); }
+
+    __forceinline       short* bounds_vx_lower1(size_t N)       { return (short*)((char*)this+6+11*N); }
+    __forceinline const short* bounds_vx_lower1(size_t N) const { return (short*)((char*)this+6+11*N); }
+    
+    __forceinline       short* bounds_vx_upper1(size_t N)       { return (short*)((char*)this+6+13*N); }
+    __forceinline const short* bounds_vx_upper1(size_t N) const { return (short*)((char*)this+6+13*N); }
+
+    __forceinline       char* bounds_vy_x(size_t N)       { return (char*)((char*)this+6+15*N); }
+    __forceinline const char* bounds_vy_x(size_t N) const { return (char*)((char*)this+6+15*N); }
+    
+    __forceinline       char* bounds_vy_y(size_t N)       { return (char*)((char*)this+6+16*N); }
+    __forceinline const char* bounds_vy_y(size_t N) const { return (char*)((char*)this+6+16*N); }
+    
+    __forceinline       char* bounds_vy_z(size_t N)       { return (char*)((char*)this+6+17*N); }
+    __forceinline const char* bounds_vy_z(size_t N) const { return (char*)((char*)this+6+17*N); }
+    
+    __forceinline       short* bounds_vy_lower0(size_t N)       { return (short*)((char*)this+6+18*N); }
+    __forceinline const short* bounds_vy_lower0(size_t N) const { return (short*)((char*)this+6+18*N); }
+    
+    __forceinline       short* bounds_vy_upper0(size_t N)       { return (short*)((char*)this+6+20*N); }
+    __forceinline const short* bounds_vy_upper0(size_t N) const { return (short*)((char*)this+6+20*N); }
+
+    __forceinline       short* bounds_vy_lower1(size_t N)       { return (short*)((char*)this+6+22*N); }
+    __forceinline const short* bounds_vy_lower1(size_t N) const { return (short*)((char*)this+6+22*N); }
+    
+    __forceinline       short* bounds_vy_upper1(size_t N)       { return (short*)((char*)this+6+24*N); }
+    __forceinline const short* bounds_vy_upper1(size_t N) const { return (short*)((char*)this+6+24*N); }
+    
+    __forceinline       char* bounds_vz_x(size_t N)       { return (char*)((char*)this+6+26*N); }
+    __forceinline const char* bounds_vz_x(size_t N) const { return (char*)((char*)this+6+26*N); }
+    
+    __forceinline       char* bounds_vz_y(size_t N)       { return (char*)((char*)this+6+27*N); }
+    __forceinline const char* bounds_vz_y(size_t N) const { return (char*)((char*)this+6+27*N); }
+    
+    __forceinline       char* bounds_vz_z(size_t N)       { return (char*)((char*)this+6+28*N); }
+    __forceinline const char* bounds_vz_z(size_t N) const { return (char*)((char*)this+6+28*N); }
+    
+    __forceinline       short* bounds_vz_lower0(size_t N)       { return (short*)((char*)this+6+29*N); }
+    __forceinline const short* bounds_vz_lower0(size_t N) const { return (short*)((char*)this+6+29*N); }
+    
+    __forceinline       short* bounds_vz_upper0(size_t N)       { return (short*)((char*)this+6+31*N); }
+    __forceinline const short* bounds_vz_upper0(size_t N) const { return (short*)((char*)this+6+31*N); }
+
+    __forceinline       short* bounds_vz_lower1(size_t N)       { return (short*)((char*)this+6+33*N); }
+    __forceinline const short* bounds_vz_lower1(size_t N) const { return (short*)((char*)this+6+33*N); }
+    
+    __forceinline       short* bounds_vz_upper1(size_t N)       { return (short*)((char*)this+6+35*N); }
+    __forceinline const short* bounds_vz_upper1(size_t N) const { return (short*)((char*)this+6+35*N); }
+
+    __forceinline       Vec3f* offset(size_t N)       { return (Vec3f*)((char*)this+6+37*N); }
+    __forceinline const Vec3f* offset(size_t N) const { return (Vec3f*)((char*)this+6+37*N); }
+    
+    __forceinline       float* scale(size_t N)       { return (float*)((char*)this+6+37*N+12); }
+    __forceinline const float* scale(size_t N) const { return (float*)((char*)this+6+37*N+12); }
+
+    __forceinline       float& time_offset(size_t N)       { return *(float*)((char*)this+6+37*N+16); }
+    __forceinline const float& time_offset(size_t N) const { return *(float*)((char*)this+6+37*N+16); }
+    
+    __forceinline       float& time_scale(size_t N)       { return *(float*)((char*)this+6+37*N+20); }
+    __forceinline const float& time_scale(size_t N) const { return *(float*)((char*)this+6+37*N+20); }
+
+    __forceinline       char* end(size_t N)       { return (char*)this+6+37*N+24; }
+    __forceinline const char* end(size_t N) const { return (char*)this+6+37*N+24; }
+  };
+
+  template<int M>
+    typename CurveNiMB<M>::Type CurveNiMB<M>::type;
+
+  typedef CurveNiMB<4> Curve4iMB;
+  typedef CurveNiMB<8> Curve8iMB;
+}

thirdparty/embree/kernels/geometry/curveNi_mb_intersector.h

@@ -0,0 +1,531 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "curveNi_mb.h"
+#include "../subdiv/linear_bezier_patch.h"
+
+#include "roundline_intersector.h"
+#include "coneline_intersector.h"
+#include "curve_intersector_ribbon.h"
+#include "curve_intersector_oriented.h"
+#include "curve_intersector_sweep.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+      struct CurveNiMBIntersector1
+    {
+      typedef CurveNiMB<M> Primitive;
+      typedef Vec3vf<M> Vec3vfM;
+      typedef LinearSpace3<Vec3vfM>LinearSpace3vfM;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline vbool<M> intersect(Ray& ray, const Primitive& prim, vfloat<M>& tNear_o)
+      {
+        const size_t N = prim.N;
+#if __SYCL_DEVICE_ONLY__
+        const Vec3f offset = *prim.offset(N);
+        const float scale  = *prim.scale(N);
+#else
+        const vfloat4 offset_scale = vfloat4::loadu(prim.offset(N));
+        const Vec3fa offset = Vec3fa(offset_scale);
+        const Vec3fa scale = Vec3fa(shuffle<3,3,3,3>(offset_scale));
+#endif
+        const Vec3fa org1 = (ray.org-offset)*scale;
+        const Vec3fa dir1 = ray.dir*scale;
+        
+        const LinearSpace3vfM space(vfloat<M>::load(prim.bounds_vx_x(N)), vfloat<M>::load(prim.bounds_vx_y(N)), vfloat<M>::load(prim.bounds_vx_z(N)),
+                                    vfloat<M>::load(prim.bounds_vy_x(N)), vfloat<M>::load(prim.bounds_vy_y(N)), vfloat<M>::load(prim.bounds_vy_z(N)),
+                                    vfloat<M>::load(prim.bounds_vz_x(N)), vfloat<M>::load(prim.bounds_vz_y(N)), vfloat<M>::load(prim.bounds_vz_z(N)));
+
+        const Vec3vfM dir2 = xfmVector(space,Vec3vfM(dir1));
+        const Vec3vfM org2 = xfmPoint (space,Vec3vfM(org1));
+        const Vec3vfM rcp_dir2 = rcp_safe(dir2);
+
+        const vfloat<M> ltime = (ray.time()-prim.time_offset(N))*prim.time_scale(N);
+        const vfloat<M> vx_lower0 = vfloat<M>::load(prim.bounds_vx_lower0(N));
+        const vfloat<M> vx_lower1 = vfloat<M>::load(prim.bounds_vx_lower1(N));
+        const vfloat<M> vx_lower = madd(ltime,vx_lower1-vx_lower0,vx_lower0);
+        const vfloat<M> vx_upper0 = vfloat<M>::load(prim.bounds_vx_upper0(N));
+        const vfloat<M> vx_upper1 = vfloat<M>::load(prim.bounds_vx_upper1(N));
+        const vfloat<M> vx_upper = madd(ltime,vx_upper1-vx_upper0,vx_upper0);
+
+        const vfloat<M> vy_lower0 = vfloat<M>::load(prim.bounds_vy_lower0(N));
+        const vfloat<M> vy_lower1 = vfloat<M>::load(prim.bounds_vy_lower1(N));
+        const vfloat<M> vy_lower = madd(ltime,vy_lower1-vy_lower0,vy_lower0);
+        const vfloat<M> vy_upper0 = vfloat<M>::load(prim.bounds_vy_upper0(N));
+        const vfloat<M> vy_upper1 = vfloat<M>::load(prim.bounds_vy_upper1(N));
+        const vfloat<M> vy_upper = madd(ltime,vy_upper1-vy_upper0,vy_upper0);
+        
+        const vfloat<M> vz_lower0 = vfloat<M>::load(prim.bounds_vz_lower0(N));
+        const vfloat<M> vz_lower1 = vfloat<M>::load(prim.bounds_vz_lower1(N));
+        const vfloat<M> vz_lower = madd(ltime,vz_lower1-vz_lower0,vz_lower0);
+        const vfloat<M> vz_upper0 = vfloat<M>::load(prim.bounds_vz_upper0(N));
+        const vfloat<M> vz_upper1 = vfloat<M>::load(prim.bounds_vz_upper1(N));
+        const vfloat<M> vz_upper = madd(ltime,vz_upper1-vz_upper0,vz_upper0);
+       
+        const vfloat<M> t_lower_x = (vx_lower-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_upper_x = (vx_upper-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_lower_y = (vy_lower-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_upper_y = (vy_upper-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_lower_z = (vz_lower-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+        const vfloat<M> t_upper_z = (vz_upper-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+
+        const vfloat<M> round_up  (1.0f+3.0f*float(ulp));
+        const vfloat<M> round_down(1.0f-3.0f*float(ulp));
+        const vfloat<M> tNear = round_down*max(mini(t_lower_x,t_upper_x),mini(t_lower_y,t_upper_y),mini(t_lower_z,t_upper_z),vfloat<M>(ray.tnear()));
+        const vfloat<M> tFar  = round_up  *min(maxi(t_lower_x,t_upper_x),maxi(t_lower_y,t_upper_y),maxi(t_lower_z,t_upper_z),vfloat<M>(ray.tfar));
+        tNear_o = tNear;
+        return (vint<M>(step) < vint<M>(prim.N)) & (tNear <= tFar);
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID),ray.time());
+
+          Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID),ray.time());
+
+          if (Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID)))
+              return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_n(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray.org, primID,ray.time());
+          Intersector().intersect(pre,ray,context,geom,primID,curve,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_n(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray.org, primID,ray.time());
+
+          if (Intersector().intersect(pre,ray,context,geom,primID,curve,Epilog(ray,context,geomID,primID)))
+              return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_h(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID),ray.time());
+          Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_h(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID),ray.time());
+          if (Intersector().intersect(pre,ray,context,geom,primID,p0,t0,p1,t1,Epilog(ray,context,geomID,primID)))
+              return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_hn(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedHermiteCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray.org, primID,ray.time());
+          Intersector().intersect(pre,ray,context,geom,primID,curve,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_hn(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedHermiteCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray.org, primID,ray.time());
+          if (Intersector().intersect(pre,ray,context,geom,primID,curve,Epilog(ray,context,geomID,primID)))
+              return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+    };
+
+    template<int M, int K>
+      struct CurveNiMBIntersectorK
+    {
+      typedef CurveNiMB<M> Primitive;
+      typedef Vec3vf<M> Vec3vfM;
+      typedef LinearSpace3<Vec3vfM>LinearSpace3vfM;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline vbool<M> intersect(RayK<K>& ray, const size_t k, const Primitive& prim, vfloat<M>& tNear_o)
+      {
+        const size_t N = prim.N;
+#if __SYCL_DEVICE_ONLY__
+        const Vec3f offset = *prim.offset(N);
+        const float scale  = *prim.scale(N);
+#else
+        const vfloat4 offset_scale = vfloat4::loadu(prim.offset(N));
+        const Vec3fa offset = Vec3fa(offset_scale);
+        const Vec3fa scale = Vec3fa(shuffle<3,3,3,3>(offset_scale));
+#endif
+        const Vec3fa ray_org(ray.org.x[k],ray.org.y[k],ray.org.z[k]);
+        const Vec3fa ray_dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]);
+        const Vec3fa org1 = (ray_org-offset)*scale;
+        const Vec3fa dir1 = ray_dir*scale;
+        
+        const LinearSpace3vfM space(vfloat<M>::load(prim.bounds_vx_x(N)), vfloat<M>::load(prim.bounds_vx_y(N)), vfloat<M>::load(prim.bounds_vx_z(N)),
+                                    vfloat<M>::load(prim.bounds_vy_x(N)), vfloat<M>::load(prim.bounds_vy_y(N)), vfloat<M>::load(prim.bounds_vy_z(N)),
+                                    vfloat<M>::load(prim.bounds_vz_x(N)), vfloat<M>::load(prim.bounds_vz_y(N)), vfloat<M>::load(prim.bounds_vz_z(N)));
+
+        const Vec3vfM dir2 = xfmVector(space,Vec3vfM(dir1));
+        const Vec3vfM org2 = xfmPoint (space,Vec3vfM(org1));
+        const Vec3vfM rcp_dir2 = rcp_safe(dir2);
+
+        const vfloat<M> ltime = (ray.time()[k]-prim.time_offset(N))*prim.time_scale(N);
+        const vfloat<M> vx_lower0 = vfloat<M>::load(prim.bounds_vx_lower0(N));
+        const vfloat<M> vx_lower1 = vfloat<M>::load(prim.bounds_vx_lower1(N));
+        const vfloat<M> vx_lower = madd(ltime,vx_lower1-vx_lower0,vx_lower0);
+        const vfloat<M> vx_upper0 = vfloat<M>::load(prim.bounds_vx_upper0(N));
+        const vfloat<M> vx_upper1 = vfloat<M>::load(prim.bounds_vx_upper1(N));
+        const vfloat<M> vx_upper = madd(ltime,vx_upper1-vx_upper0,vx_upper0);
+
+        const vfloat<M> vy_lower0 = vfloat<M>::load(prim.bounds_vy_lower0(N));
+        const vfloat<M> vy_lower1 = vfloat<M>::load(prim.bounds_vy_lower1(N));
+        const vfloat<M> vy_lower = madd(ltime,vy_lower1-vy_lower0,vy_lower0);
+        const vfloat<M> vy_upper0 = vfloat<M>::load(prim.bounds_vy_upper0(N));
+        const vfloat<M> vy_upper1 = vfloat<M>::load(prim.bounds_vy_upper1(N));
+        const vfloat<M> vy_upper = madd(ltime,vy_upper1-vy_upper0,vy_upper0);
+        
+        const vfloat<M> vz_lower0 = vfloat<M>::load(prim.bounds_vz_lower0(N));
+        const vfloat<M> vz_lower1 = vfloat<M>::load(prim.bounds_vz_lower1(N));
+        const vfloat<M> vz_lower = madd(ltime,vz_lower1-vz_lower0,vz_lower0);
+        const vfloat<M> vz_upper0 = vfloat<M>::load(prim.bounds_vz_upper0(N));
+        const vfloat<M> vz_upper1 = vfloat<M>::load(prim.bounds_vz_upper1(N));
+        const vfloat<M> vz_upper = madd(ltime,vz_upper1-vz_upper0,vz_upper0);
+       
+        const vfloat<M> t_lower_x = (vx_lower-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_upper_x = (vx_upper-vfloat<M>(org2.x))*vfloat<M>(rcp_dir2.x);
+        const vfloat<M> t_lower_y = (vy_lower-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_upper_y = (vy_upper-vfloat<M>(org2.y))*vfloat<M>(rcp_dir2.y);
+        const vfloat<M> t_lower_z = (vz_lower-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+        const vfloat<M> t_upper_z = (vz_upper-vfloat<M>(org2.z))*vfloat<M>(rcp_dir2.z);
+
+        const vfloat<M> round_up  (1.0f+3.0f*float(ulp));
+        const vfloat<M> round_down(1.0f-3.0f*float(ulp));
+        const vfloat<M> tNear = round_down*max(mini(t_lower_x,t_upper_x),mini(t_lower_y,t_upper_y),mini(t_lower_z,t_upper_z),vfloat<M>(ray.tnear()[k]));
+        const vfloat<M> tFar  = round_up  *min(maxi(t_lower_x,t_upper_x),maxi(t_lower_y,t_upper_y),maxi(t_lower_z,t_upper_z),vfloat<M>(ray.tfar[k]));
+        tNear_o = tNear;
+        return (vint<M>(step) < vint<M>(prim.N)) & (tNear <= tFar);
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID),ray.time()[k]);
+
+          Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff a0,a1,a2,a3; geom->gather(a0,a1,a2,a3,geom->curve(primID),ray.time()[k]);
+
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID)))
+            return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_n(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const Vec3fa ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,ray.time()[k]);
+          Intersector().intersect(pre,ray,k,context,geom,primID,curve,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_n(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const Vec3fa ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,ray.time()[k]);
+          
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,curve,Epilog(ray,k,context,geomID,primID)))
+            return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_h(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID),ray.time()[k]);
+          Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_h(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          Vec3ff p0,t0,p1,t1; geom->gather_hermite(p0,t0,p1,t1,geom->curve(primID),ray.time()[k]);
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,p0,t0,p1,t1,Epilog(ray,k,context,geomID,primID)))
+            return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_hn(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const Vec3fa ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedHermiteCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,ray.time()[k]);
+          Intersector().intersect(pre,ray,k,context,geom,primID,curve,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_hn(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = context->scene->get<CurveGeometry>(geomID);
+          const Vec3fa ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const TensorLinearCubicBezierSurface3fa curve = geom->getNormalOrientedHermiteCurve<typename Intersector::SourceCurve3ff, typename Intersector::SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,ray.time()[k]);
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,curve,Epilog(ray,k,context,geomID,primID)))
+            return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curveNv.h

@@ -0,0 +1,101 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "curveNi.h"
+
+namespace embree
+{
+  template<int M>
+    struct CurveNv : public CurveNi<M>
+  {
+    using CurveNi<M>::N;
+      
+    struct Type : public PrimitiveType {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+M-1)/M; }
+
+    static __forceinline size_t bytes(size_t N)
+    {
+      const size_t f = N/M, r = N%M;
+      static_assert(sizeof(CurveNv) == 22+25*M+4*16*M, "internal data layout issue");
+      return f*sizeof(CurveNv) + (r!=0)*(22 + 25*r + 4*16*r);
+    }
+
+  public:
+
+    /*! Default constructor. */
+    __forceinline CurveNv () {}
+
+    /*! fill curve from curve list */
+    __forceinline void fill(const PrimRef* prims, size_t& begin, size_t _end, Scene* scene)
+    {
+      size_t end = min(begin+M,_end);
+      size_t N = end-begin;
+
+      /* encode all primitives */
+      for (size_t i=0; i<N; i++)
+      {
+        const PrimRef& prim = prims[begin+i];
+        const unsigned int geomID = prim.geomID();
+        const unsigned int primID = prim.primID();
+        CurveGeometry* mesh = (CurveGeometry*) scene->get(geomID);
+        const unsigned vtxID = mesh->curve(primID);
+        Vec3fa::storeu(&this->vertices(i,N)[0],mesh->vertex(vtxID+0));
+        Vec3fa::storeu(&this->vertices(i,N)[1],mesh->vertex(vtxID+1));
+        Vec3fa::storeu(&this->vertices(i,N)[2],mesh->vertex(vtxID+2));
+        Vec3fa::storeu(&this->vertices(i,N)[3],mesh->vertex(vtxID+3));
+      }
+    }
+
+    template<typename BVH, typename Allocator>
+      __forceinline static typename BVH::NodeRef createLeaf (BVH* bvh, const PrimRef* prims, const range<size_t>& set, const Allocator& alloc)
+    {
+      if (set.size() == 0)
+        return BVH::emptyNode;
+      
+      /* fall back to CurveNi for oriented curves */
+      unsigned int geomID = prims[set.begin()].geomID();
+      if (bvh->scene->get(geomID)->getCurveType() == Geometry::GTY_SUBTYPE_ORIENTED_CURVE) {
+        return CurveNi<M>::createLeaf(bvh,prims,set,alloc);
+      }
+      if (bvh->scene->get(geomID)->getCurveBasis() == Geometry::GTY_BASIS_HERMITE) {
+        return CurveNi<M>::createLeaf(bvh,prims,set,alloc);
+      }
+      
+      size_t start = set.begin();
+      size_t items = CurveNv::blocks(set.size());
+      size_t numbytes = CurveNv::bytes(set.size());
+      CurveNv* accel = (CurveNv*) alloc.malloc1(numbytes,BVH::byteAlignment);
+      for (size_t i=0; i<items; i++) {
+        accel[i].CurveNv<M>::fill(prims,start,set.end(),bvh->scene);
+        accel[i].CurveNi<M>::fill(prims,start,set.end(),bvh->scene);
+      }
+      return bvh->encodeLeaf((char*)accel,items);
+    };
+    
+  public:
+    unsigned char data[4*16*M];
+    __forceinline       Vec3fa* vertices(size_t i, size_t N)       { return (Vec3fa*)CurveNi<M>::end(N)+4*i; }
+    __forceinline const Vec3fa* vertices(size_t i, size_t N) const { return (Vec3fa*)CurveNi<M>::end(N)+4*i; }
+  };
+
+  template<int M>
+    typename CurveNv<M>::Type CurveNv<M>::type;
+
+  typedef CurveNv<4> Curve4v;
+  typedef CurveNv<8> Curve8v;
+}

thirdparty/embree/kernels/geometry/curveNv_intersector.h

@@ -0,0 +1,181 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "curveNv.h"
+#include "curveNi_intersector.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+      struct CurveNvIntersector1 : public CurveNiIntersector1<M>
+    {
+      typedef CurveNv<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = CurveNiIntersector1<M>::intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = (CurveGeometry*) context->scene->get(geomID);
+          const Vec3ff a0 = Vec3ff::loadu(&prim.vertices(i,N)[0]);
+          const Vec3ff a1 = Vec3ff::loadu(&prim.vertices(i,N)[1]);
+          const Vec3ff a2 = Vec3ff::loadu(&prim.vertices(i,N)[2]);
+          const Vec3ff a3 = Vec3ff::loadu(&prim.vertices(i,N)[3]);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            prefetchL1(&prim.vertices(i1,N)[0]);
+            prefetchL1(&prim.vertices(i1,N)[4]);
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              prefetchL2(&prim.vertices(i2,N)[0]);
+              prefetchL2(&prim.vertices(i2,N)[4]);
+            }
+          }
+
+          Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = CurveNiIntersector1<M>::intersect(ray,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = (CurveGeometry*) context->scene->get(geomID);
+          const Vec3ff a0 = Vec3ff::loadu(&prim.vertices(i,N)[0]);
+          const Vec3ff a1 = Vec3ff::loadu(&prim.vertices(i,N)[1]);
+          const Vec3ff a2 = Vec3ff::loadu(&prim.vertices(i,N)[2]);
+          const Vec3ff a3 = Vec3ff::loadu(&prim.vertices(i,N)[3]);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            prefetchL1(&prim.vertices(i1,N)[0]);
+            prefetchL1(&prim.vertices(i1,N)[4]);
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              prefetchL2(&prim.vertices(i2,N)[0]);
+              prefetchL2(&prim.vertices(i2,N)[4]);
+            }
+          }
+          
+          if (Intersector().intersect(pre,ray,context,geom,primID,a0,a1,a2,a3,Epilog(ray,context,geomID,primID)))
+            return true;
+          
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar));
+        }
+        return false;
+      }
+    };
+
+    template<int M, int K>
+      struct CurveNvIntersectorK : public CurveNiIntersectorK<M,K>
+    {
+      typedef CurveNv<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline void intersect_t(Precalculations& pre, RayHitK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = CurveNiIntersectorK<M,K>::intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(normal.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = (CurveGeometry*) context->scene->get(geomID);
+          const Vec3ff a0 = Vec3ff::loadu(&prim.vertices(i,N)[0]);
+          const Vec3ff a1 = Vec3ff::loadu(&prim.vertices(i,N)[1]);
+          const Vec3ff a2 = Vec3ff::loadu(&prim.vertices(i,N)[2]);
+          const Vec3ff a3 = Vec3ff::loadu(&prim.vertices(i,N)[3]);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            prefetchL1(&prim.vertices(i1,N)[0]);
+            prefetchL1(&prim.vertices(i1,N)[4]);
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              prefetchL2(&prim.vertices(i2,N)[0]);
+              prefetchL2(&prim.vertices(i2,N)[4]);
+            }
+          }
+
+          Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID));
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+      }
+
+      template<typename Intersector, typename Epilog>
+        static __forceinline bool occluded_t(Precalculations& pre, RayK<K>& ray, const size_t k, RayQueryContext* context, const Primitive& prim)
+      {
+        vfloat<M> tNear;
+        vbool<M> valid = CurveNiIntersectorK<M,K>::intersect(ray,k,prim,tNear);
+
+        const size_t N = prim.N;
+        size_t mask = movemask(valid);
+        while (mask)
+        {
+          const size_t i = bscf(mask);
+          STAT3(shadow.trav_prims,1,1,1);
+          const unsigned int geomID = prim.geomID(N);
+          const unsigned int primID = prim.primID(N)[i];
+          const CurveGeometry* geom = (CurveGeometry*) context->scene->get(geomID);
+          const Vec3ff a0 = Vec3ff::loadu(&prim.vertices(i,N)[0]);
+          const Vec3ff a1 = Vec3ff::loadu(&prim.vertices(i,N)[1]);
+          const Vec3ff a2 = Vec3ff::loadu(&prim.vertices(i,N)[2]);
+          const Vec3ff a3 = Vec3ff::loadu(&prim.vertices(i,N)[3]);
+
+          size_t mask1 = mask;
+          const size_t i1 = bscf(mask1);
+          if (mask) {
+            prefetchL1(&prim.vertices(i1,N)[0]);
+            prefetchL1(&prim.vertices(i1,N)[4]);
+            if (mask1) {
+              const size_t i2 = bsf(mask1);
+              prefetchL2(&prim.vertices(i2,N)[0]);
+              prefetchL2(&prim.vertices(i2,N)[4]);
+            }
+          }
+
+          if (Intersector().intersect(pre,ray,k,context,geom,primID,a0,a1,a2,a3,Epilog(ray,k,context,geomID,primID)))
+            return true;
+
+          mask &= movemask(tNear <= vfloat<M>(ray.tfar[k]));
+        }
+        return false;
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector.h

@@ -0,0 +1,98 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../subdiv/bezier_curve.h"
+#include "../common/primref.h"
+#include "bezier_hair_intersector.h"
+#include "bezier_ribbon_intersector.h"
+#include "bezier_curve_intersector.h"
+#include "oriented_curve_intersector.h"
+#include "../bvh/node_intersector1.h"
+
+// FIXME: this file seems replicate of curve_intersector_virtual.h
+
+namespace embree
+{
+  namespace isa
+  {
+    struct VirtualCurveIntersector1
+    {
+      typedef unsigned char Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+      
+      template<int N, bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        assert(num == 1);
+        RTCGeometryType ty = (RTCGeometryType)(*prim);
+        assert(This->leafIntersector);
+        VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+        leafIntersector.intersect<1>(&pre,&ray,context,prim);
+      }
+      
+      template<int N, bool robust>        
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        assert(num == 1);
+        RTCGeometryType ty = (RTCGeometryType)(*prim);
+        assert(This->leafIntersector);
+        VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+        return leafIntersector.occluded<1>(&pre,&ray,context,prim);
+      }
+    };
+
+    template<int K>
+      struct VirtualCurveIntersectorK 
+      {
+        typedef unsigned char Primitive;
+        typedef CurvePrecalculationsK<K> Precalculations;
+        
+        static __forceinline void intersect(const vbool<K>& valid_i, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+          size_t mask = movemask(valid_i);
+          while (mask) leafIntersector.intersect<K>(&pre,&ray,bscf(mask),context,prim);
+        }
+        
+        static __forceinline vbool<K> occluded(const vbool<K>& valid_i, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+          vbool<K> valid_o = false;
+          size_t mask = movemask(valid_i);
+          while (mask) {
+            size_t k = bscf(mask);
+            if (leafIntersector.occluded<K>(&pre,&ray,k,context,prim))
+              set(valid_o, k);
+          }
+          return valid_o;
+        }
+        
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+          leafIntersector.intersect<K>(&pre,&ray,k,context,prim);
+        }
+        
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurvePrimitive::Intersectors& leafIntersector = ((VirtualCurvePrimitive*) This->leafIntersector)->vtbl[ty];
+          return leafIntersector.occluded<K>(&pre,&ray,k,context,prim);
+        }
+      };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_distance.h

@@ -0,0 +1,170 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename NativeCurve3fa, int M>
+    struct DistanceCurveHit
+    {
+      __forceinline DistanceCurveHit() {}
+
+      __forceinline DistanceCurveHit(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& T, const int i, const int N,
+                                     const NativeCurve3fa& curve3D)
+        : U(U), V(V), T(T), i(i), N(N), curve3D(curve3D), valid(valid) {}
+      
+      __forceinline void finalize() 
+      {
+        vu = (vfloat<M>(step)+U+vfloat<M>(float(i)))*(1.0f/float(N));
+        vv = V;
+        vt = T;
+      }
+      
+      __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+      __forceinline float t  (const size_t i) const { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) const { 
+        return curve3D.eval_du(vu[i]);
+      }
+      
+    public:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> T;
+      int i, N;
+      NativeCurve3fa curve3D;
+      
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+    };
+
+    template<typename NativeCurve3fa>
+    struct DistanceCurveHit<NativeCurve3fa,1>
+    {
+      enum { M = 1 };
+      
+      __forceinline DistanceCurveHit() {}
+
+      __forceinline DistanceCurveHit(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& T, const int i, const int N,
+                                     const NativeCurve3fa& curve3D)
+        : U(U), V(V), T(T), i(i), N(N), curve3D(curve3D), valid(valid) {}
+      
+      __forceinline void finalize() 
+      {
+        vu = (vfloat<M>(step)+U+vfloat<M>(float(i)))*(1.0f/float(N));
+        vv = V;
+        vt = T;
+      }
+      
+      __forceinline Vec2f uv () const { return Vec2f(vu,vv); }
+      __forceinline float t  () const { return vt; }
+      __forceinline Vec3fa Ng() const { return curve3D.eval_du(vu); }
+      
+    public:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> T;
+      int i, N;
+      NativeCurve3fa curve3D;
+      
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+    };
+    
+    template<typename NativeCurve3fa, int W = VSIZEX>
+    struct DistanceCurve1Intersector1
+    {
+      using vboolx = vbool<W>;
+      using vintx = vint<W>;
+      using vfloatx = vfloat<W>;
+      using Vec4vfx = Vec4vf<W>;
+      
+      template<typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3,
+                                   const Epilog& epilog)
+      {
+        const int N = geom->tessellationRate;
+        
+        /* transform control points into ray space */
+        const NativeCurve3fa curve3Di(v0,v1,v2,v3);
+        const NativeCurve3fa curve3D = enlargeRadiusToMinWidth(context,geom,ray.org,curve3Di);
+        const NativeCurve3fa curve2D = curve3D.xfm_pr(pre.ray_space,ray.org);
+      
+        /* evaluate the bezier curve */
+        vboolx valid = vfloatx(step) < vfloatx(float(N));
+        const Vec4vfx p0 = curve2D.template eval0<W>(0,N);
+        const Vec4vfx p1 = curve2D.template eval1<W>(0,N);
+
+        /* approximative intersection with cone */
+        const Vec4vfx v = p1-p0;
+        const Vec4vfx w = -p0;
+        const vfloatx d0 = madd(w.x,v.x,w.y*v.y);
+        const vfloatx d1 = madd(v.x,v.x,v.y*v.y);
+        const vfloatx u = clamp(d0*rcp(d1),vfloatx(zero),vfloatx(one));
+        const Vec4vfx p = madd(u,v,p0);
+        const vfloatx t = p.z*pre.depth_scale;
+        const vfloatx d2 = madd(p.x,p.x,p.y*p.y); 
+        const vfloatx r = p.w;
+        const vfloatx r2 = r*r;
+        valid &= (d2 <= r2) & (vfloatx(ray.tnear()) <= t) & (t <= vfloatx(ray.tfar));
+        if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) 
+          valid &= t > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*pre.depth_scale; // ignore self intersections
+
+        /* update hit information */
+        bool ishit = false;
+        if (unlikely(any(valid))) {
+          DistanceCurveHit<NativeCurve3fa,W> hit(valid,u,0.0f,t,0,N,curve3D);
+          ishit = ishit | epilog(valid,hit);
+        }
+
+        if (unlikely(W < N)) 
+        {
+          /* process SIMD-size many segments per iteration */
+          for (int i=W; i<N; i+=W)
+          {
+            /* evaluate the bezier curve */
+            vboolx valid = vintx(i)+vintx(step) < vintx(N);
+            const Vec4vfx p0 = curve2D.template eval0<W>(i,N);
+            const Vec4vfx p1 = curve2D.template eval1<W>(i,N);
+            
+            /* approximative intersection with cone */
+            const Vec4vfx v = p1-p0;
+            const Vec4vfx w = -p0;
+            const vfloatx d0 = madd(w.x,v.x,w.y*v.y);
+            const vfloatx d1 = madd(v.x,v.x,v.y*v.y);
+            const vfloatx u = clamp(d0*rcp(d1),vfloatx(zero),vfloatx(one));
+            const Vec4vfx p = madd(u,v,p0);
+            const vfloatx t = p.z*pre.depth_scale;
+            const vfloatx d2 = madd(p.x,p.x,p.y*p.y); 
+            const vfloatx r = p.w;
+            const vfloatx r2 = r*r;
+            valid &= (d2 <= r2) & (vfloatx(ray.tnear()) <= t) & (t <= vfloatx(ray.tfar));
+            if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f)
+              valid &= t > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*pre.depth_scale; // ignore self intersections
+
+             /* update hit information */
+            if (unlikely(any(valid))) {
+              DistanceCurveHit<NativeCurve3fa,W> hit(valid,u,0.0f,t,i,N,curve3D);
+              ishit = ishit | epilog(valid,hit);
+            }
+          }
+        }
+        return ishit;
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_oriented.h

@@ -0,0 +1,425 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "curve_intersector_precalculations.h"
+#include "curve_intersector_sweep.h"
+#include "../subdiv/linear_bezier_patch.h"
+
+#define DBG(x)
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename Ray, typename Epilog, int N = VSIZEX-1, int V = VSIZEX>
+      struct TensorLinearCubicBezierSurfaceIntersector
+      {
+        const LinearSpace3fa& ray_space;
+        Ray& ray;
+        TensorLinearCubicBezierSurface3fa curve3d;
+        TensorLinearCubicBezierSurface2fa curve2d;
+        float eps;
+        const Epilog& epilog;
+        bool isHit;
+
+        __forceinline TensorLinearCubicBezierSurfaceIntersector (const LinearSpace3fa& ray_space, Ray& ray, const TensorLinearCubicBezierSurface3fa& curve3d, const Epilog& epilog)
+          : ray_space(ray_space), ray(ray), curve3d(curve3d), epilog(epilog), isHit(false)
+        {
+          const TensorLinearCubicBezierSurface3fa curve3dray = curve3d.xfm(ray_space,ray.org);
+          curve2d = TensorLinearCubicBezierSurface2fa(CubicBezierCurve2fa(curve3dray.L),CubicBezierCurve2fa(curve3dray.R));
+          const BBox2fa b2 = curve2d.bounds();
+          eps = 8.0f*float(ulp)*reduce_max(max(abs(b2.lower),abs(b2.upper)));
+        }
+        
+        __forceinline Interval1f solve_linear(const float u0, const float u1, const float& p0, const float& p1)
+        {
+          if (p1 == p0) {
+            if (p0 == 0.0f) return Interval1f(u0,u1);
+            else return Interval1f(empty);
+          }
+          const float t = -p0/(p1-p0);
+          const float tt = lerp(u0,u1,t);
+          return Interval1f(tt);
+        }
+
+        __forceinline void solve_linear(const float u0, const float u1, const Interval1f& p0, const Interval1f& p1, Interval1f& u)
+        {
+          if (sign(p0.lower) != sign(p0.upper)) u.extend(u0);
+          if (sign(p0.lower) != sign(p1.lower)) u.extend(solve_linear(u0,u1,p0.lower,p1.lower));
+          if (sign(p0.upper) != sign(p1.upper)) u.extend(solve_linear(u0,u1,p0.upper,p1.upper));
+          if (sign(p1.lower) != sign(p1.upper)) u.extend(u1);
+        }
+
+        __forceinline Interval1f bezier_clipping(const CubicBezierCurve<Interval1f>& curve)
+        {
+          Interval1f u = empty;
+          solve_linear(0.0f/3.0f,1.0f/3.0f,curve.v0,curve.v1,u);
+          solve_linear(0.0f/3.0f,2.0f/3.0f,curve.v0,curve.v2,u);
+          solve_linear(0.0f/3.0f,3.0f/3.0f,curve.v0,curve.v3,u);
+          solve_linear(1.0f/3.0f,2.0f/3.0f,curve.v1,curve.v2,u);
+          solve_linear(1.0f/3.0f,3.0f/3.0f,curve.v1,curve.v3,u);
+          solve_linear(2.0f/3.0f,3.0f/3.0f,curve.v2,curve.v3,u);
+          return intersect(u,Interval1f(0.0f,1.0f));
+        }
+        
+        __forceinline Interval1f bezier_clipping(const LinearBezierCurve<Interval1f>& curve)
+        {
+          Interval1f v = empty;
+          solve_linear(0.0f,1.0f,curve.v0,curve.v1,v);
+          return intersect(v,Interval1f(0.0f,1.0f));
+        }
+
+        __forceinline void solve_bezier_clipping(BBox1f cu, BBox1f cv, const TensorLinearCubicBezierSurface2fa& curve2)
+        {
+          BBox2fa bounds = curve2.bounds();
+          if (bounds.upper.x < 0.0f) return;
+          if (bounds.upper.y < 0.0f) return;
+          if (bounds.lower.x > 0.0f) return;
+          if (bounds.lower.y > 0.0f) return;
+          
+          if (max(cu.size(),cv.size()) < 1E-4f)
+          {
+            const float u = cu.center();
+            const float v = cv.center();
+            TensorLinearCubicBezierSurface1f curve_z = curve3d.xfm(ray_space.row2(),ray.org);
+            const float t = curve_z.eval(u,v);
+            if (ray.tnear() <= t && t <= ray.tfar) {
+              const Vec3fa Ng = cross(curve3d.eval_du(u,v),curve3d.eval_dv(u,v));
+              BezierCurveHit hit(t,u,v,Ng);
+              isHit |= epilog(hit);
+            }
+            return;
+          }
+          
+          const Vec2fa dv = curve2.axis_v();
+          const TensorLinearCubicBezierSurface1f curve1v = curve2.xfm(dv);
+          LinearBezierCurve<Interval1f> curve0v = curve1v.reduce_u();
+          if (!curve0v.hasRoot()) return;
+          
+          const Interval1f v = bezier_clipping(curve0v);
+          if (isEmpty(v)) return;
+          TensorLinearCubicBezierSurface2fa curve2a = curve2.clip_v(v);
+          cv = BBox1f(lerp(cv.lower,cv.upper,v.lower),lerp(cv.lower,cv.upper,v.upper));
+
+          const Vec2fa du = curve2.axis_u();
+          const TensorLinearCubicBezierSurface1f curve1u = curve2a.xfm(du);
+          CubicBezierCurve<Interval1f> curve0u = curve1u.reduce_v();         
+          int roots = curve0u.maxRoots();
+          if (roots == 0) return;
+          
+          if (roots == 1)
+          {
+            const Interval1f u = bezier_clipping(curve0u);
+            if (isEmpty(u)) return;
+            TensorLinearCubicBezierSurface2fa curve2b = curve2a.clip_u(u);
+            cu = BBox1f(lerp(cu.lower,cu.upper,u.lower),lerp(cu.lower,cu.upper,u.upper));
+            solve_bezier_clipping(cu,cv,curve2b);
+            return;
+          }
+
+          TensorLinearCubicBezierSurface2fa curve2l, curve2r;
+          curve2a.split_u(curve2l,curve2r);
+          solve_bezier_clipping(BBox1f(cu.lower,cu.center()),cv,curve2l);
+          solve_bezier_clipping(BBox1f(cu.center(),cu.upper),cv,curve2r);
+        }
+        
+        __forceinline bool solve_bezier_clipping()
+        {
+          solve_bezier_clipping(BBox1f(0.0f,1.0f),BBox1f(0.0f,1.0f),curve2d);
+          return isHit;
+        }
+
+        __forceinline void solve_newton_raphson(BBox1f cu, BBox1f cv)
+        {
+          Vec2fa uv(cu.center(),cv.center());
+          const Vec2fa dfdu = curve2d.eval_du(uv.x,uv.y);
+          const Vec2fa dfdv = curve2d.eval_dv(uv.x,uv.y);
+          const LinearSpace2fa rcp_J = rcp(LinearSpace2fa(dfdu,dfdv));
+          solve_newton_raphson_loop(cu,cv,uv,dfdu,dfdv,rcp_J);
+        }
+
+        __forceinline void solve_newton_raphson_loop(BBox1f cu, BBox1f cv, const Vec2fa& uv_in, const Vec2fa& dfdu, const Vec2fa& dfdv, const LinearSpace2fa& rcp_J)
+        {
+          Vec2fa uv = uv_in;
+          
+          for (size_t i=0; i<200; i++)
+          {
+            const Vec2fa f = curve2d.eval(uv.x,uv.y);
+            const Vec2fa duv = rcp_J*f;
+            uv -= duv;
+
+            if (max(abs(f.x),abs(f.y)) < eps)
+            {
+              const float u = uv.x;
+              const float v = uv.y;
+              if (!(u >= 0.0f && u <= 1.0f)) return; // rejects NaNs
+              if (!(v >= 0.0f && v <= 1.0f)) return; // rejects NaNs
+              const TensorLinearCubicBezierSurface1f curve_z = curve3d.xfm(ray_space.row2(),ray.org);
+              const float t = curve_z.eval(u,v);
+              if (!(ray.tnear() <= t && t <= ray.tfar)) return; // rejects NaNs
+              const Vec3fa Ng = cross(curve3d.eval_du(u,v),curve3d.eval_dv(u,v));
+              BezierCurveHit hit(t,u,v,Ng);
+              isHit |= epilog(hit);
+              return;
+            }
+          }       
+        }
+
+        __forceinline bool clip_v(BBox1f& cu, BBox1f& cv)
+        {
+          const Vec2fa dv = curve2d.eval_dv(cu.lower,cv.lower);
+          const TensorLinearCubicBezierSurface1f curve1v = curve2d.xfm(dv).clip(cu,cv);
+          LinearBezierCurve<Interval1f> curve0v = curve1v.reduce_u();
+          if (!curve0v.hasRoot()) return false;
+          Interval1f v = bezier_clipping(curve0v);
+          if (isEmpty(v)) return false;
+          v = intersect(v + Interval1f(-0.1f,+0.1f),Interval1f(0.0f,1.0f));
+          cv = BBox1f(lerp(cv.lower,cv.upper,v.lower),lerp(cv.lower,cv.upper,v.upper));
+          return true;
+        }
+
+        __forceinline bool solve_krawczyk(bool very_small, BBox1f& cu, BBox1f& cv)
+        {
+          /* perform bezier clipping in v-direction to get tight v-bounds */
+          TensorLinearCubicBezierSurface2fa curve2 = curve2d.clip(cu,cv);
+          const Vec2fa dv = curve2.axis_v();
+          const TensorLinearCubicBezierSurface1f curve1v = curve2.xfm(dv);
+          LinearBezierCurve<Interval1f> curve0v = curve1v.reduce_u();
+          if (unlikely(!curve0v.hasRoot())) return true;
+          Interval1f v = bezier_clipping(curve0v);
+          if (unlikely(isEmpty(v))) return true;
+          v = intersect(v + Interval1f(-0.1f,+0.1f),Interval1f(0.0f,1.0f));
+          curve2 = curve2.clip_v(v);
+          cv = BBox1f(lerp(cv.lower,cv.upper,v.lower),lerp(cv.lower,cv.upper,v.upper));
+
+          /* perform one newton raphson iteration */
+          Vec2fa c(cu.center(),cv.center());
+          Vec2fa f,dfdu,dfdv; curve2d.eval(c.x,c.y,f,dfdu,dfdv);
+          const LinearSpace2fa rcp_J = rcp(LinearSpace2fa(dfdu,dfdv));
+          const Vec2fa c1 = c - rcp_J*f;
+          
+          /* calculate bounds of derivatives */
+          const BBox2fa bounds_du = (1.0f/cu.size())*curve2.derivative_u().bounds();
+          const BBox2fa bounds_dv = (1.0f/cv.size())*curve2.derivative_v().bounds();
+
+          /* calculate krawczyk test */
+          LinearSpace2<Vec2<Interval1f>> I(Interval1f(1.0f), Interval1f(0.0f),
+                                           Interval1f(0.0f), Interval1f(1.0f));
+
+          LinearSpace2<Vec2<Interval1f>> G(Interval1f(bounds_du.lower.x,bounds_du.upper.x), Interval1f(bounds_dv.lower.x,bounds_dv.upper.x),
+                                           Interval1f(bounds_du.lower.y,bounds_du.upper.y), Interval1f(bounds_dv.lower.y,bounds_dv.upper.y));
+
+          const LinearSpace2<Vec2f> rcp_J2(rcp_J);
+          const LinearSpace2<Vec2<Interval1f>> rcp_Ji(rcp_J2);
+          
+          const Vec2<Interval1f> x(cu,cv);
+          const Vec2<Interval1f> K = Vec2<Interval1f>(Vec2f(c1)) + (I - rcp_Ji*G)*(x-Vec2<Interval1f>(Vec2f(c)));
+
+          /* test if there is no solution */
+          const Vec2<Interval1f> KK = intersect(K,x);
+          if (unlikely(isEmpty(KK.x) || isEmpty(KK.y))) return true;
+
+          /* exit if convergence cannot get proven, but terminate if we are very small */
+          if (unlikely(!subset(K,x) && !very_small)) return false;
+
+          /* solve using newton raphson iteration of convergence is guaranteed */
+          solve_newton_raphson_loop(cu,cv,c1,dfdu,dfdv,rcp_J);
+          return true;
+        }
+
+        __forceinline void solve_newton_raphson_no_recursion(BBox1f cu, BBox1f cv)
+        {
+           if (!clip_v(cu,cv)) return;
+           return solve_newton_raphson(cu,cv);
+        }
+
+        __forceinline void solve_newton_raphson_recursion(BBox1f cu, BBox1f cv)
+        {
+          unsigned int sptr = 0;
+          const unsigned int stack_size = 4;
+          unsigned int mask_stack[stack_size];
+          BBox1f cu_stack[stack_size];
+          BBox1f cv_stack[stack_size];
+          goto entry;
+          
+          /* terminate if stack is empty */
+          while (sptr)
+          {
+            /* pop from stack */
+            {
+              sptr--;
+              size_t mask = mask_stack[sptr];
+              cu = cu_stack[sptr];
+              cv = cv_stack[sptr];
+              const size_t i = bscf(mask);
+              mask_stack[sptr] = mask;
+              if (mask) sptr++; // there are still items on the stack
+              
+              /* process next element recurse into each hit curve segment */
+              const float u0 = float(i+0)*(1.0f/(N));
+              const float u1 = float(i+1)*(1.0f/(N));
+              const BBox1f cui(lerp(cu.lower,cu.upper,u0),lerp(cu.lower,cu.upper,u1));
+              cu = cui;
+            }
+
+#if 0
+            solve_newton_raphson_no_recursion(cu,cv);
+            continue;
+            
+#else
+            /* we assume convergence for small u ranges and verify using krawczyk */
+            if (cu.size() < 1.0f/6.0f) {
+              const bool very_small = cu.size() < 0.001f || sptr >= stack_size;
+              if (solve_krawczyk(very_small,cu,cv)) {
+                continue;
+              }
+            }
+#endif
+
+          entry:
+
+            /* split the curve into N segments in u-direction */
+            unsigned int mask = 0;
+            for (int i=0; i<N;)
+            {
+              int i0 = i;
+              vbool<V> valid = true;
+              TensorLinearCubicBezierSurface<Vec2vf<V>> subcurves = curve2d.clip_v(cv).template vsplit_u<V>(valid,cu,i,N);
+              
+              /* slabs test in u-direction */
+              Vec2vf<V> ndv = cross(subcurves.axis_v());
+              BBox<vfloat<V>> boundsv = subcurves.template vxfm<V>(ndv).bounds();
+              valid &= boundsv.lower <= eps;
+              valid &= boundsv.upper >= -eps;
+              if (none(valid)) continue;
+
+              /* slabs test in v-direction */
+              Vec2vf<V> ndu = cross(subcurves.axis_u());
+              BBox<vfloat<V>> boundsu = subcurves.template vxfm<V>(ndu).bounds();
+              valid &= boundsu.lower <= eps;
+              valid &= boundsu.upper >= -eps;
+              if (none(valid)) continue;
+
+              mask |= movemask(valid) << i0;
+            }
+
+            if (!mask) continue;
+
+            /* push valid segments to stack */
+            assert(sptr < stack_size);
+            mask_stack [sptr] = mask;
+            cu_stack   [sptr] = cu;
+            cv_stack   [sptr] = cv;
+            sptr++;
+          }
+        }
+        
+        __forceinline bool solve_newton_raphson_main()
+        {
+          BBox1f vu(0.0f,1.0f);
+          BBox1f vv(0.0f,1.0f);
+          solve_newton_raphson_recursion(vu,vv);
+          return isHit;
+        }
+      };
+
+
+    template<template<typename Ty> class SourceCurve, int N = VSIZEX-1, int V = VSIZEX>
+      struct OrientedCurve1Intersector1
+    {
+      //template<typename Ty> using Curve = SourceCurve<Ty>;
+      typedef SourceCurve<Vec3ff> SourceCurve3ff;
+      typedef SourceCurve<Vec3fa> SourceCurve3fa;
+      
+      __forceinline OrientedCurve1Intersector1() {}
+      
+      __forceinline OrientedCurve1Intersector1(const Ray& ray, const void* ptr) {}
+      
+      template<typename Ray, typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray,
+                                RayQueryContext* context,
+                                const CurveGeometry* geom, const unsigned int primID, 
+                                const Vec3ff& v0i, const Vec3ff& v1i, const Vec3ff& v2i, const Vec3ff& v3i,
+                                const Vec3fa& n0i, const Vec3fa& n1i, const Vec3fa& n2i, const Vec3fa& n3i,
+                                const Epilog& epilog) const
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        SourceCurve3ff ccurve(v0i,v1i,v2i,v3i);
+        SourceCurve3fa ncurve(n0i,n1i,n2i,n3i);
+        ccurve = enlargeRadiusToMinWidth(context,geom,ray.org,ccurve);
+        TensorLinearCubicBezierSurface3fa curve = TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(ccurve,ncurve);
+        //return TensorLinearCubicBezierSurfaceIntersector<Ray,Epilog>(pre.ray_space,ray,curve,epilog).solve_bezier_clipping();
+        return TensorLinearCubicBezierSurfaceIntersector<Ray,Epilog,N,V>(pre.ray_space,ray,curve,epilog).solve_newton_raphson_main();
+      }
+
+      template<typename Ray, typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray,
+                                RayQueryContext* context,
+                                const CurveGeometry* geom, const unsigned int primID,
+                                const TensorLinearCubicBezierSurface3fa& curve, const Epilog& epilog) const
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        //return TensorLinearCubicBezierSurfaceIntersector<Ray,Epilog>(pre.ray_space,ray,curve,epilog).solve_bezier_clipping();
+        return TensorLinearCubicBezierSurfaceIntersector<Ray,Epilog,N,V>(pre.ray_space,ray,curve,epilog).solve_newton_raphson_main();
+      }
+    };
+
+    template<template<typename Ty> class SourceCurve, int K>
+      struct OrientedCurve1IntersectorK
+    {
+      //template<typename Ty> using Curve = SourceCurve<Ty>;
+      typedef SourceCurve<Vec3ff> SourceCurve3ff;
+      typedef SourceCurve<Vec3fa> SourceCurve3fa;
+      
+      struct Ray1
+      {
+        __forceinline Ray1(RayK<K>& ray, size_t k)
+          : org(ray.org.x[k],ray.org.y[k],ray.org.z[k]), dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]), _tnear(ray.tnear()[k]), tfar(ray.tfar[k]) {}
+
+        Vec3fa org;
+        Vec3fa dir;
+        float _tnear;
+        float& tfar;
+
+        __forceinline float& tnear() { return _tnear; }
+        //__forceinline float& tfar()  { return _tfar; }
+        __forceinline const float& tnear() const { return _tnear; }
+        //__forceinline const float& tfar()  const { return _tfar; }
+      };
+
+      template<typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculationsK<K>& pre, RayK<K>& vray, size_t k,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const Vec3ff& v0i, const Vec3ff& v1i, const Vec3ff& v2i, const Vec3ff& v3i,
+                                   const Vec3fa& n0i, const Vec3fa& n1i, const Vec3fa& n2i, const Vec3fa& n3i,
+                                   const Epilog& epilog)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Ray1 ray(vray,k);
+        SourceCurve3ff ccurve(v0i,v1i,v2i,v3i);
+        SourceCurve3fa ncurve(n0i,n1i,n2i,n3i);
+        ccurve = enlargeRadiusToMinWidth(context,geom,ray.org,ccurve);
+        TensorLinearCubicBezierSurface3fa curve = TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(ccurve,ncurve);
+        //return TensorLinearCubicBezierSurfaceIntersector<Ray1,Epilog>(pre.ray_space[k],ray,curve,epilog).solve_bezier_clipping();
+        return TensorLinearCubicBezierSurfaceIntersector<Ray1,Epilog>(pre.ray_space[k],ray,curve,epilog).solve_newton_raphson_main();
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculationsK<K>& pre, RayK<K>& vray, size_t k,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const TensorLinearCubicBezierSurface3fa& curve,
+                                   const Epilog& epilog)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Ray1 ray(vray,k);
+        //return TensorLinearCubicBezierSurfaceIntersector<Ray1,Epilog>(pre.ray_space[k],ray,curve,epilog).solve_bezier_clipping();
+        return TensorLinearCubicBezierSurfaceIntersector<Ray1,Epilog>(pre.ray_space[k],ray,curve,epilog).solve_newton_raphson_main();
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_precalculations.h

@@ -0,0 +1,49 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/geometry.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct CurvePrecalculations1
+    {
+      float depth_scale;
+      LinearSpace3fa ray_space;
+           
+      __forceinline CurvePrecalculations1() {}
+
+      __forceinline CurvePrecalculations1(const Ray& ray, const void* ptr)
+      {
+        depth_scale = rsqrt(dot(ray.dir,ray.dir));
+        LinearSpace3fa space = frame(depth_scale*ray.dir);
+        space.vz *= depth_scale;
+        ray_space = space.transposed();
+      }
+    };
+    
+    template<int K>
+      struct CurvePrecalculationsK
+    {
+      vfloat<K> depth_scale;
+      LinearSpace3fa ray_space[K];
+
+      __forceinline CurvePrecalculationsK(const vbool<K>& valid, const RayK<K>& ray)
+      {
+        size_t mask = movemask(valid);
+        depth_scale = rsqrt(dot(ray.dir,ray.dir));
+        while (mask) {
+          size_t k = bscf(mask);
+          Vec3fa ray_dir_k = Vec3fa(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]);
+          LinearSpace3fa ray_space_k = frame(depth_scale[k]*ray_dir_k);
+          ray_space_k.vz *= depth_scale[k];
+          ray_space[k] = ray_space_k.transposed();
+        }
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_ribbon.h

@@ -0,0 +1,225 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "quad_intersector.h"
+#include "curve_intersector_precalculations.h"
+
+#define Bezier1Intersector1 RibbonCurve1Intersector1
+#define Bezier1IntersectorK RibbonCurve1IntersectorK
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename NativeCurve3ff, int M>
+    struct RibbonHit
+    {
+      __forceinline RibbonHit() {}
+
+      __forceinline RibbonHit(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& T, const int i, const int N,
+                              const NativeCurve3ff& curve3D)
+        : U(U), V(V), T(T), i(i), N(N), curve3D(curve3D), valid(valid) {}
+      
+      __forceinline void finalize() 
+      {
+        vu = (vfloat<M>(step)+U+vfloat<M>(float(i)))*(1.0f/float(N));
+        vv = V;
+        vt = T;
+      }
+      
+      __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+      __forceinline float t  (const size_t i) const { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) const { return curve3D.eval_du(vu[i]); }
+
+      __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); }
+      __forceinline vfloat<M> t () const { return vt; }
+      __forceinline Vec3vf<M> Ng() const { return (Vec3vf<M>) curve3D.template veval_du<M>(vu); }
+      
+    public:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> T;
+      int i, N;
+      NativeCurve3ff curve3D;
+      
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+    };
+
+    /* calculate squared distance of point p0 to line p1->p2 */
+    template<int M>
+    __forceinline std::pair<vfloat<M>,vfloat<M>> sqr_point_line_distance(const Vec2vf<M>& p0, const Vec2vf<M>& p1, const Vec2vf<M>& p2)
+    {
+      const vfloat<M> num = det(p2-p1,p1-p0);
+      const vfloat<M> den2 = dot(p2-p1,p2-p1);
+      return std::make_pair(num*num,den2);
+    }
+    
+    /* performs culling against a cylinder */
+     template<int M>
+     __forceinline vbool<M> cylinder_culling_test(const Vec2vf<M>& p0, const Vec2vf<M>& p1, const Vec2vf<M>& p2, const vfloat<M>& r)
+    {
+      const std::pair<vfloat<M>,vfloat<M>> d = sqr_point_line_distance<M>(p0,p1,p2);
+      return d.first <= r*r*d.second;
+    }
+
+    template<int M = VSIZEX, typename NativeCurve3ff, typename Epilog>
+    __forceinline bool intersect_ribbon(const Vec3fa& ray_org, const Vec3fa& ray_dir, const float ray_tnear, const float& ray_tfar,
+                                        const LinearSpace3fa& ray_space, const float& depth_scale,
+                                        const NativeCurve3ff& curve3D, const int N,
+                                        const Epilog& epilog)
+    {
+      /* transform control points into ray space */
+      const NativeCurve3ff curve2D = curve3D.xfm_pr(ray_space,ray_org);
+      float eps = 4.0f*float(ulp)*reduce_max(max(abs(curve2D.v0),abs(curve2D.v1),abs(curve2D.v2),abs(curve2D.v3)));
+
+      int i=0;
+      bool ishit = false;
+      
+#if !defined(__SYCL_DEVICE_ONLY__)
+      {
+        /* evaluate the bezier curve */
+        vbool<M> valid = vfloat<M>(step) < vfloat<M>(float(N));
+        const Vec4vf<M> p0 = curve2D.template eval0<M>(0,N);
+        const Vec4vf<M> p1 = curve2D.template eval1<M>(0,N);
+        valid &= cylinder_culling_test<M>(zero,Vec2vf<M>(p0.x,p0.y),Vec2vf<M>(p1.x,p1.y),max(p0.w,p1.w));
+        
+        if (any(valid)) 
+        {
+          Vec3vf<M> dp0dt = curve2D.template derivative0<M>(0,N);
+          Vec3vf<M> dp1dt = curve2D.template derivative1<M>(0,N);
+          dp0dt = select(reduce_max(abs(dp0dt)) < vfloat<M>(eps),Vec3vf<M>(p1-p0),dp0dt);
+          dp1dt = select(reduce_max(abs(dp1dt)) < vfloat<M>(eps),Vec3vf<M>(p1-p0),dp1dt);
+          const Vec3vf<M> n0(dp0dt.y,-dp0dt.x,0.0f);
+          const Vec3vf<M> n1(dp1dt.y,-dp1dt.x,0.0f);
+          const Vec3vf<M> nn0 = normalize(n0);
+          const Vec3vf<M> nn1 = normalize(n1);
+          const Vec3vf<M> lp0 = madd(p0.w,nn0,Vec3vf<M>(p0));
+          const Vec3vf<M> lp1 = madd(p1.w,nn1,Vec3vf<M>(p1));
+          const Vec3vf<M> up0 = nmadd(p0.w,nn0,Vec3vf<M>(p0));
+          const Vec3vf<M> up1 = nmadd(p1.w,nn1,Vec3vf<M>(p1));
+          
+          vfloat<M> vu,vv,vt;
+          vbool<M> valid0 = intersect_quad_backface_culling<M>(valid,zero,Vec3fa(0,0,1),ray_tnear,ray_tfar,lp0,lp1,up1,up0,vu,vv,vt);
+          
+          if (any(valid0))
+          {
+            /* ignore self intersections */
+            if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) {
+              vfloat<M> r = lerp(p0.w, p1.w, vu);
+              valid0 &= vt > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale;
+            }
+            
+            if (any(valid0))
+            {
+              vv = madd(2.0f,vv,vfloat<M>(-1.0f));
+              RibbonHit<NativeCurve3ff,M> bhit(valid0,vu,vv,vt,0,N,curve3D);
+              ishit |= epilog(bhit.valid,bhit);
+            }
+          }
+        }
+        i += M;
+      }
+      
+      if (unlikely(i < N))
+#endif
+      {
+        /* process SIMD-size many segments per iteration */
+        for (; i<N; i+=M)
+        {
+          /* evaluate the bezier curve */
+          vbool<M> valid = vint<M>(i)+vint<M>(step) < vint<M>(N);
+          const Vec4vf<M> p0 = curve2D.template eval0<M>(i,N);
+          const Vec4vf<M> p1 = curve2D.template eval1<M>(i,N);
+          valid &= cylinder_culling_test<M>(zero,Vec2vf<M>(p0.x,p0.y),Vec2vf<M>(p1.x,p1.y),max(p0.w,p1.w));
+          if (none(valid)) continue;
+          
+          Vec3vf<M> dp0dt = curve2D.template derivative0<M>(i,N);
+          Vec3vf<M> dp1dt = curve2D.template derivative1<M>(i,N);
+          dp0dt = select(reduce_max(abs(dp0dt)) < vfloat<M>(eps),Vec3vf<M>(p1-p0),dp0dt);
+          dp1dt = select(reduce_max(abs(dp1dt)) < vfloat<M>(eps),Vec3vf<M>(p1-p0),dp1dt);
+          const Vec3vf<M> n0(dp0dt.y,-dp0dt.x,0.0f);
+          const Vec3vf<M> n1(dp1dt.y,-dp1dt.x,0.0f);
+          const Vec3vf<M> nn0 = normalize(n0);
+          const Vec3vf<M> nn1 = normalize(n1);
+          const Vec3vf<M> lp0 = madd(p0.w,nn0,Vec3vf<M>(p0));
+          const Vec3vf<M> lp1 = madd(p1.w,nn1,Vec3vf<M>(p1));
+          const Vec3vf<M> up0 = nmadd(p0.w,nn0,Vec3vf<M>(p0));
+          const Vec3vf<M> up1 = nmadd(p1.w,nn1,Vec3vf<M>(p1));
+          
+          vfloat<M> vu,vv,vt;
+          vbool<M> valid0 = intersect_quad_backface_culling<M>(valid,zero,Vec3fa(0,0,1),ray_tnear,ray_tfar,lp0,lp1,up1,up0,vu,vv,vt);
+
+          if (any(valid0))
+          {
+            /* ignore self intersections */
+            if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) {
+              vfloat<M> r = lerp(p0.w, p1.w, vu);
+              valid0 &= vt > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale;
+            }
+            
+            if (any(valid0))
+            {
+              vv = madd(2.0f,vv,vfloat<M>(-1.0f));
+              RibbonHit<NativeCurve3ff,M> bhit(valid0,vu,vv,vt,i,N,curve3D);
+              ishit |= epilog(bhit.valid,bhit);
+            }
+          }
+        }
+      }
+      return ishit;
+    }
+        
+    template<template<typename Ty> class NativeCurve, int M = VSIZEX>
+    struct RibbonCurve1Intersector1
+    {
+      typedef NativeCurve<Vec3ff> NativeCurve3ff;
+      
+      template<typename Ray, typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3,
+                                   const Epilog& epilog)
+      {
+        const int N = geom->tessellationRate;
+        NativeCurve3ff curve(v0,v1,v2,v3);
+        curve = enlargeRadiusToMinWidth(context,geom,ray.org,curve);
+        return intersect_ribbon<M,NativeCurve3ff>(ray.org,ray.dir,ray.tnear(),ray.tfar,
+                                                pre.ray_space,pre.depth_scale,
+                                                curve,N,
+                                                epilog);
+      }
+    };
+    
+    template<template<typename Ty> class NativeCurve, int K, int M = VSIZEX>
+    struct RibbonCurve1IntersectorK
+    {
+      typedef NativeCurve<Vec3ff> NativeCurve3ff;
+      
+      template<typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculationsK<K>& pre, RayK<K>& ray, size_t k,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3,
+                                   const Epilog& epilog)
+      {
+        const int N = geom->tessellationRate;
+        const Vec3fa ray_org(ray.org.x[k],ray.org.y[k],ray.org.z[k]);
+        const Vec3fa ray_dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]);
+        NativeCurve3ff curve(v0,v1,v2,v3);
+        curve = enlargeRadiusToMinWidth(context,geom,ray_org,curve);
+        return intersect_ribbon<M,NativeCurve3ff>(ray_org,ray_dir,ray.tnear()[k],ray.tfar[k],
+                                                pre.ray_space[k],pre.depth_scale[k],
+                                                curve,N,
+                                                epilog);
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_sweep.h

@@ -0,0 +1,511 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "cylinder.h"
+#include "plane.h"
+#include "line_intersector.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    static const size_t numJacobianIterations = 5;
+#if defined(EMBREE_SYCL_SUPPORT) && defined(__SYCL_DEVICE_ONLY__)
+    static const size_t numBezierSubdivisions = 2;
+#elif defined(__AVX__)
+    static const size_t numBezierSubdivisions = 2;
+#else
+    static const size_t numBezierSubdivisions = 3;
+#endif
+
+    struct BezierCurveHit
+    {
+      __forceinline BezierCurveHit() {}
+
+      __forceinline BezierCurveHit(const float t, const float u, const Vec3fa& Ng)
+        : t(t), u(u), v(0.0f), Ng(Ng) {}
+
+      __forceinline BezierCurveHit(const float t, const float u, const float v, const Vec3fa& Ng)
+        : t(t), u(u), v(v), Ng(Ng) {}
+      
+      __forceinline void finalize() {}
+      
+    public:
+      float t;
+      float u;
+      float v; 
+      Vec3fa Ng;
+    };
+    
+    template<typename NativeCurve3ff, typename Ray, typename Epilog>
+    __forceinline bool intersect_bezier_iterative_debug(const Ray& ray, const float dt, const NativeCurve3ff& curve, size_t i,
+                                                        const vfloatx& u, const BBox<vfloatx>& tp, const BBox<vfloatx>& h0, const BBox<vfloatx>& h1, 
+                                                        const Vec3vfx& Ng, const Vec4vfx& dP0du, const Vec4vfx& dP3du,
+                                                        const Epilog& epilog)
+    {
+      if (tp.lower[i]+dt > ray.tfar) return false;
+      Vec3fa Ng_o = Vec3fa(Ng.x[i],Ng.y[i],Ng.z[i]);
+      if (h0.lower[i] == tp.lower[i]) Ng_o = -Vec3fa(dP0du.x[i],dP0du.y[i],dP0du.z[i]);
+      if (h1.lower[i] == tp.lower[i]) Ng_o = +Vec3fa(dP3du.x[i],dP3du.y[i],dP3du.z[i]);
+      BezierCurveHit hit(tp.lower[i]+dt,u[i],Ng_o);
+      return epilog(hit);
+    }
+
+    template<typename NativeCurve3ff, typename Ray, typename Epilog> 
+     __forceinline bool intersect_bezier_iterative_jacobian(const Ray& ray, const float dt, const NativeCurve3ff& curve, float u, float t, const Epilog& epilog)
+    {
+      const Vec3fa org = zero;
+      const Vec3fa dir = ray.dir;
+      const float length_ray_dir = length(dir);
+
+      /* error of curve evaluations is proportional to largest coordinate */
+      const BBox3ff box = curve.bounds();
+      const float P_err = 16.0f*float(ulp)*reduce_max(max(abs(box.lower),abs(box.upper)));
+     
+      for (size_t i=0; i<numJacobianIterations; i++) 
+      {
+        const Vec3fa Q = madd(Vec3fa(t),dir,org);
+        //const Vec3fa dQdu = zero;
+        const Vec3fa dQdt = dir;
+        const float Q_err = 16.0f*float(ulp)*length_ray_dir*t; // works as org=zero here
+           
+        Vec3ff P,dPdu,ddPdu; curve.eval(u,P,dPdu,ddPdu);
+        //const Vec3fa dPdt = zero;
+
+        const Vec3fa R = Q-P;
+        const float len_R = length(R); //reduce_max(abs(R));
+        const float R_err = max(Q_err,P_err);
+        const Vec3fa dRdu = /*dQdu*/-dPdu;
+        const Vec3fa dRdt = dQdt;//-dPdt;
+
+        const Vec3fa T = normalize(dPdu);
+        const Vec3fa dTdu = dnormalize(dPdu,ddPdu);
+        //const Vec3fa dTdt = zero;
+        const float cos_err = P_err/length(dPdu);
+
+        /* Error estimate for dot(R,T):
+
+           dot(R,T) = cos(R,T) |R| |T|
+                    = (cos(R,T) +- cos_error) * (|R| +- |R|_err) * (|T| +- |T|_err)
+                    = cos(R,T)*|R|*|T| 
+                      +- cos(R,T)*(|R|*|T|_err + |T|*|R|_err)
+                      +- cos_error*(|R| + |T|)
+                      +- lower order terms
+           with cos(R,T) being in [0,1] and |T| = 1 we get:
+             dot(R,T)_err = |R|*|T|_err + |R|_err = cos_error*(|R|+1)
+        */
+              
+        const float f = dot(R,T);
+        const float f_err = len_R*P_err + R_err + cos_err*(1.0f+len_R);
+        const float dfdu = dot(dRdu,T) + dot(R,dTdu);
+        const float dfdt = dot(dRdt,T);// + dot(R,dTdt);
+
+        const float K = dot(R,R)-sqr(f);
+        const float dKdu = /*2.0f*/(dot(R,dRdu)-f*dfdu);
+        const float dKdt = /*2.0f*/(dot(R,dRdt)-f*dfdt);
+        const float rsqrt_K = rsqrt(K);
+
+        const float g = sqrt(K)-P.w;
+        const float g_err = R_err + f_err + 16.0f*float(ulp)*box.upper.w;
+        const float dgdu = /*0.5f*/dKdu*rsqrt_K-dPdu.w;
+        const float dgdt = /*0.5f*/dKdt*rsqrt_K;//-dPdt.w;
+
+        const LinearSpace2f J = LinearSpace2f(dfdu,dfdt,dgdu,dgdt);
+        const Vec2f dut = rcp(J)*Vec2f(f,g);
+        const Vec2f ut = Vec2f(u,t) - dut;
+        u = ut.x; t = ut.y;
+
+        if (abs(f) < f_err && abs(g) < g_err)
+        {
+          t+=dt;
+          if (!(ray.tnear() <= t && t <= ray.tfar)) return false; // rejects NaNs
+          if (!(u >= 0.0f && u <= 1.0f)) return false; // rejects NaNs
+          const Vec3fa R = normalize(Q-P);
+          const Vec3fa U = madd(Vec3fa(dPdu.w),R,dPdu);
+          const Vec3fa V = cross(dPdu,R);
+          BezierCurveHit hit(t,u,cross(V,U));
+          return epilog(hit);
+        }
+      }
+      return false;
+    }
+
+#if !defined(__SYCL_DEVICE_ONLY__)
+    
+    template<typename NativeCurve3ff, typename Ray, typename Epilog>
+    __forceinline bool intersect_bezier_recursive_jacobian(const Ray& ray, const float dt, const NativeCurve3ff& curve, const Epilog& epilog)
+    {
+      float u0 = 0.0f;
+      float u1 = 1.0f;
+      unsigned int depth = 1;
+        
+#if defined(__AVX__)
+      enum { VSIZEX_ = 8 };
+      typedef vbool8 vboolx; // maximally 8-wide to work around KNL issues
+      typedef vint8 vintx; 
+      typedef vfloat8 vfloatx;
+#else
+      enum { VSIZEX_ = 4 };
+      typedef vbool4 vboolx;
+      typedef vint4 vintx; 
+      typedef vfloat4 vfloatx;
+#endif
+    
+      unsigned int maxDepth = numBezierSubdivisions;
+      bool found = false;
+      const Vec3fa org = zero;
+      const Vec3fa dir = ray.dir;
+
+      unsigned int sptr = 0;
+      const unsigned int stack_size = numBezierSubdivisions+1; // +1 because of unstable workaround below
+      struct StackEntry {
+        vboolx valid;
+        vfloatx tlower;
+        float u0;
+        float u1;
+        unsigned int depth;
+      };
+      StackEntry stack[stack_size];
+      goto entry;
+
+       /* terminate if stack is empty */
+      while (sptr)
+      {
+        /* pop from stack */
+        {
+          sptr--;
+          vboolx valid = stack[sptr].valid;
+          const vfloatx tlower = stack[sptr].tlower;
+          valid &= tlower+dt <= ray.tfar;
+          if (none(valid)) continue;
+          u0 = stack[sptr].u0;
+          u1 = stack[sptr].u1;
+          depth = stack[sptr].depth;
+          const size_t i = select_min(valid,tlower); clear(valid,i);
+          stack[sptr].valid = valid;
+          if (any(valid)) sptr++; // there are still items on the stack
+
+          /* process next segment */
+          const vfloatx vu0 = lerp(u0,u1,vfloatx(step)*(1.0f/(vfloatx::size-1)));
+          u0 = vu0[i+0];
+          u1 = vu0[i+1];
+        }
+      entry:
+
+        /* subdivide curve */
+        const float dscale = (u1-u0)*(1.0f/(3.0f*(vfloatx::size-1)));
+        const vfloatx vu0 = lerp(u0,u1,vfloatx(step)*(1.0f/(vfloatx::size-1)));
+        Vec4vfx P0, dP0du; curve.template veval<VSIZEX_>(vu0,P0,dP0du); dP0du = dP0du * Vec4vfx(dscale);
+        const Vec4vfx P3 = shift_right_1(P0);
+        const Vec4vfx dP3du = shift_right_1(dP0du); 
+        const Vec4vfx P1 = P0 + dP0du; 
+        const Vec4vfx P2 = P3 - dP3du;
+        
+        /* calculate bounding cylinders */
+        const vfloatx rr1 = sqr_point_to_line_distance(Vec3vfx(dP0du),Vec3vfx(P3-P0));
+        const vfloatx rr2 = sqr_point_to_line_distance(Vec3vfx(dP3du),Vec3vfx(P3-P0));
+        const vfloatx maxr12 = sqrt(max(rr1,rr2));
+        const vfloatx one_plus_ulp  = 1.0f+2.0f*float(ulp);
+        const vfloatx one_minus_ulp = 1.0f-2.0f*float(ulp);
+        vfloatx r_outer = max(P0.w,P1.w,P2.w,P3.w)+maxr12;
+        vfloatx r_inner = min(P0.w,P1.w,P2.w,P3.w)-maxr12;
+        r_outer = one_plus_ulp*r_outer;
+        r_inner = max(0.0f,one_minus_ulp*r_inner);
+        const CylinderN<vfloatx::size> cylinder_outer(Vec3vfx(P0),Vec3vfx(P3),r_outer);
+        const CylinderN<vfloatx::size> cylinder_inner(Vec3vfx(P0),Vec3vfx(P3),r_inner);
+        vboolx valid = true; clear(valid,vfloatx::size-1);
+        
+        /* intersect with outer cylinder */
+        BBox<vfloatx> tc_outer; vfloatx u_outer0; Vec3vfx Ng_outer0; vfloatx u_outer1; Vec3vfx Ng_outer1;
+        valid &= cylinder_outer.intersect(org,dir,tc_outer,u_outer0,Ng_outer0,u_outer1,Ng_outer1);
+        if (none(valid)) continue;
+        
+        /* intersect with cap-planes */
+        BBox<vfloatx> tp(ray.tnear()-dt,ray.tfar-dt);
+        tp = embree::intersect(tp,tc_outer);
+        BBox<vfloatx> h0 = HalfPlaneN<vfloatx::size>(Vec3vfx(P0),+Vec3vfx(dP0du)).intersect(org,dir);
+        tp = embree::intersect(tp,h0);
+        BBox<vfloatx> h1 = HalfPlaneN<vfloatx::size>(Vec3vfx(P3),-Vec3vfx(dP3du)).intersect(org,dir);
+        tp = embree::intersect(tp,h1);
+        valid &= tp.lower <= tp.upper;
+        if (none(valid)) continue;
+        
+        /* clamp and correct u parameter */
+        u_outer0 = clamp(u_outer0,vfloatx(0.0f),vfloatx(1.0f));
+        u_outer1 = clamp(u_outer1,vfloatx(0.0f),vfloatx(1.0f));
+        u_outer0 = lerp(u0,u1,(vfloatx(step)+u_outer0)*(1.0f/float(vfloatx::size)));
+        u_outer1 = lerp(u0,u1,(vfloatx(step)+u_outer1)*(1.0f/float(vfloatx::size)));
+        
+        /* intersect with inner cylinder */
+        BBox<vfloatx> tc_inner;
+        vfloatx u_inner0 = zero; Vec3vfx Ng_inner0 = zero; vfloatx u_inner1 = zero; Vec3vfx Ng_inner1 = zero;
+        const vboolx valid_inner = cylinder_inner.intersect(org,dir,tc_inner,u_inner0,Ng_inner0,u_inner1,Ng_inner1);
+        
+        /* at the unstable area we subdivide deeper */
+        const vboolx unstable0 = (!valid_inner) | (abs(dot(Vec3vfx(Vec3fa(ray.dir)),Ng_inner0)) < 0.3f);
+        const vboolx unstable1 = (!valid_inner) | (abs(dot(Vec3vfx(Vec3fa(ray.dir)),Ng_inner1)) < 0.3f);
+      
+        /* subtract the inner interval from the current hit interval */
+        BBox<vfloatx> tp0, tp1;
+        subtract(tp,tc_inner,tp0,tp1);
+        vboolx valid0 = valid & (tp0.lower <= tp0.upper);
+        vboolx valid1 = valid & (tp1.lower <= tp1.upper);
+        if (none(valid0 | valid1)) continue;
+        
+        /* iterate over all first hits front to back */
+        const vintx termDepth0 = select(unstable0,vintx(maxDepth+1),vintx(maxDepth));
+        vboolx recursion_valid0 = valid0 & (depth < termDepth0);
+        valid0 &= depth >= termDepth0;
+        
+        while (any(valid0))
+        {
+          const size_t i = select_min(valid0,tp0.lower); clear(valid0,i);
+          found = found | intersect_bezier_iterative_jacobian(ray,dt,curve,u_outer0[i],tp0.lower[i],epilog);
+          //found = found | intersect_bezier_iterative_debug   (ray,dt,curve,i,u_outer0,tp0,h0,h1,Ng_outer0,dP0du,dP3du,epilog);
+          valid0 &= tp0.lower+dt <= ray.tfar;
+        }
+        valid1 &= tp1.lower+dt <= ray.tfar;
+        
+        /* iterate over all second hits front to back */
+        const vintx termDepth1 = select(unstable1,vintx(maxDepth+1),vintx(maxDepth));
+        vboolx recursion_valid1 = valid1 & (depth < termDepth1);
+        valid1 &= depth >= termDepth1;
+        while (any(valid1))
+        {
+          const size_t i = select_min(valid1,tp1.lower); clear(valid1,i);
+          found = found | intersect_bezier_iterative_jacobian(ray,dt,curve,u_outer1[i],tp1.upper[i],epilog);
+          //found = found | intersect_bezier_iterative_debug   (ray,dt,curve,i,u_outer1,tp1,h0,h1,Ng_outer1,dP0du,dP3du,epilog);
+          valid1 &= tp1.lower+dt <= ray.tfar;
+        }
+
+        /* push valid segments to stack */
+        recursion_valid0 &= tp0.lower+dt <= ray.tfar;
+        recursion_valid1 &= tp1.lower+dt <= ray.tfar;
+        const vboolx recursion_valid = recursion_valid0 | recursion_valid1;
+        if (any(recursion_valid))
+        {
+          assert(sptr < stack_size);
+          stack[sptr].valid = recursion_valid;
+          stack[sptr].tlower = select(recursion_valid0,tp0.lower,tp1.lower);
+          stack[sptr].u0 = u0;
+          stack[sptr].u1 = u1;
+          stack[sptr].depth = depth+1;
+          sptr++;
+        }
+      }
+      return found;
+    }
+
+#else
+    
+     template<typename NativeCurve3ff, typename Ray, typename Epilog>
+     __forceinline bool intersect_bezier_recursive_jacobian(const Ray& ray, const float dt, const NativeCurve3ff& curve, const Epilog& epilog)
+    {
+      const Vec3fa org = zero;
+      const Vec3fa dir = ray.dir;
+      const unsigned int max_depth = 7;
+      
+      bool found = false;
+
+      struct ShortStack
+      {
+        /* pushes both children */
+        __forceinline void push() {
+          depth++;
+        }
+
+        /* pops next node */
+        __forceinline void pop() {
+          short_stack += (1<<(31-depth));
+          depth = 31-bsf(short_stack);
+        }
+        
+        unsigned int depth = 0;
+        unsigned int short_stack = 0;
+      };
+
+      ShortStack stack;
+
+      do
+      {
+        const float u0 = (stack.short_stack+0*(1<<(31-stack.depth)))/float(0x80000000);
+        const float u1 = (stack.short_stack+1*(1<<(31-stack.depth)))/float(0x80000000);
+      
+        /* subdivide bezier curve */
+        Vec3ff P0, dP0du; curve.eval(u0,P0,dP0du); dP0du = dP0du * (u1-u0);
+        Vec3ff P3, dP3du; curve.eval(u1,P3,dP3du); dP3du = dP3du * (u1-u0);
+        const Vec3ff P1 = P0 + dP0du*(1.0f/3.0f); 
+        const Vec3ff P2 = P3 - dP3du*(1.0f/3.0f);
+
+        /* check if curve is well behaved, by checking deviation of tangents from straight line */
+        const Vec3ff W = Vec3ff(P3-P0,0.0f);
+        const Vec3ff dQ0 = abs(3.0f*(P1-P0) - W);
+        const Vec3ff dQ1 = abs(3.0f*(P2-P1) - W);
+        const Vec3ff dQ2 = abs(3.0f*(P3-P2) - W);
+        const Vec3ff max_dQ = max(dQ0,dQ1,dQ2);
+        const float m = max(max_dQ.x,max_dQ.y,max_dQ.z); //,max_dQ.w);
+        const float l = length(Vec3f(W));
+        const bool well_behaved = m < 0.2f*l;
+
+        if (!well_behaved && stack.depth < max_depth) {
+          stack.push();
+          continue;
+        }
+        
+        /* calculate bounding cylinders */
+        const float rr1 = sqr_point_to_line_distance(Vec3f(dP0du),Vec3f(P3-P0));
+        const float rr2 = sqr_point_to_line_distance(Vec3f(dP3du),Vec3f(P3-P0));
+        const float maxr12 = sqrt(max(rr1,rr2));
+        const float one_plus_ulp  = 1.0f+2.0f*float(ulp);
+        const float one_minus_ulp = 1.0f-2.0f*float(ulp);
+        float r_outer = max(P0.w,P1.w,P2.w,P3.w)+maxr12;
+        float r_inner = min(P0.w,P1.w,P2.w,P3.w)-maxr12;
+        r_outer = one_plus_ulp*r_outer;
+        r_inner = max(0.0f,one_minus_ulp*r_inner);
+        const Cylinder cylinder_outer(Vec3f(P0),Vec3f(P3),r_outer);
+        const Cylinder cylinder_inner(Vec3f(P0),Vec3f(P3),r_inner);
+        
+        /* intersect with outer cylinder */
+        BBox<float> tc_outer; float u_outer0; Vec3fa Ng_outer0; float u_outer1; Vec3fa Ng_outer1;
+        if (!cylinder_outer.intersect(org,dir,tc_outer,u_outer0,Ng_outer0,u_outer1,Ng_outer1))
+        {
+          stack.pop();
+          continue;
+        }
+                
+        /* intersect with cap-planes */
+        BBox<float> tp(ray.tnear()-dt,ray.tfar-dt);
+        tp = embree::intersect(tp,tc_outer);
+        BBox<float> h0 = HalfPlane(Vec3f(P0),+Vec3f(dP0du)).intersect(org,dir);
+        tp = embree::intersect(tp,h0);
+        BBox<float> h1 = HalfPlane(Vec3f(P3),-Vec3f(dP3du)).intersect(org,dir);
+        tp = embree::intersect(tp,h1);
+        if (tp.lower > tp.upper)
+        {
+          stack.pop();
+          continue;
+        }
+        
+        bool valid = true;
+        
+        /* clamp and correct u parameter */
+        u_outer0 = clamp(u_outer0,float(0.0f),float(1.0f));
+        u_outer1 = clamp(u_outer1,float(0.0f),float(1.0f));
+        u_outer0 = lerp(u0,u1,u_outer0);
+        u_outer1 = lerp(u0,u1,u_outer1);
+        
+        /* intersect with inner cylinder */
+        BBox<float> tc_inner;
+        float u_inner0 = zero; Vec3fa Ng_inner0 = zero; float u_inner1 = zero; Vec3fa Ng_inner1 = zero;
+        const bool valid_inner =  cylinder_inner.intersect(org,dir,tc_inner,u_inner0,Ng_inner0,u_inner1,Ng_inner1);
+
+        /* subtract the inner interval from the current hit interval */
+        BBox<float> tp0, tp1;
+        subtract(tp,tc_inner,tp0,tp1);
+        bool valid0 = valid & (tp0.lower <= tp0.upper);
+        bool valid1 = valid & (tp1.lower <= tp1.upper);
+        if (!(valid0 | valid1))
+        {
+          stack.pop();
+          continue;
+        }
+
+        /* at the unstable area we subdivide deeper */
+        const bool unstable0 = valid0 && ((!valid_inner) | (abs(dot(Vec3fa(ray.dir),Ng_inner0)) < 0.3f));
+        const bool unstable1 = valid1 && ((!valid_inner) | (abs(dot(Vec3fa(ray.dir),Ng_inner1)) < 0.3f));
+    
+        if ((unstable0 | unstable1) && (stack.depth < max_depth)) {
+           stack.push();
+           continue;
+         }
+
+        if (valid0)
+          found |= intersect_bezier_iterative_jacobian(ray,dt,curve,u_outer0,tp0.lower,epilog);
+          
+        /* the far hit cannot be closer, thus skip if we hit entry already */
+        valid1 &= tp1.lower+dt <= ray.tfar;
+        
+        /* iterate over second hit */
+        if (valid1)
+          found |= intersect_bezier_iterative_jacobian(ray,dt,curve,u_outer1,tp1.upper,epilog);
+
+        stack.pop();
+        
+      } while (stack.short_stack != 0x80000000);
+
+      return found;
+    }
+
+#endif
+    
+    template<template<typename Ty> class NativeCurve>
+    struct SweepCurve1Intersector1
+    {
+      typedef NativeCurve<Vec3ff> NativeCurve3ff;
+      
+      template<typename Ray, typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray,
+                                RayQueryContext* context,
+                                const CurveGeometry* geom, const unsigned int primID,
+                                const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3,
+                                const Epilog& epilog)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+
+        /* move ray closer to make intersection stable */
+        NativeCurve3ff curve0(v0,v1,v2,v3);
+        curve0 = enlargeRadiusToMinWidth(context,geom,ray.org,curve0);
+        const float dt = dot(curve0.center()-ray.org,ray.dir)*rcp(dot(ray.dir,ray.dir));
+        const Vec3ff ref(madd(Vec3fa(dt),ray.dir,ray.org),0.0f);
+        const NativeCurve3ff curve1 = curve0-ref;
+        return intersect_bezier_recursive_jacobian(ray,dt,curve1,epilog);
+      }
+    };
+
+    template<template<typename Ty> class NativeCurve, int K>
+    struct SweepCurve1IntersectorK
+    {
+      typedef NativeCurve<Vec3ff> NativeCurve3ff;
+      
+      struct Ray1
+      {
+        __forceinline Ray1(RayK<K>& ray, size_t k)
+          : org(ray.org.x[k],ray.org.y[k],ray.org.z[k]), dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]), _tnear(ray.tnear()[k]), tfar(ray.tfar[k]) {}
+
+        Vec3fa org;
+        Vec3fa dir;
+        float _tnear;
+        float& tfar;
+
+        __forceinline float& tnear() { return _tnear; }
+        //__forceinline float& tfar()  { return _tfar; }
+        __forceinline const float& tnear() const { return _tnear; }
+        //__forceinline const float& tfar()  const { return _tfar; }
+        
+      };
+
+      template<typename Epilog>
+      __forceinline bool intersect(const CurvePrecalculationsK<K>& pre, RayK<K>& vray, size_t k,
+                                   RayQueryContext* context,
+                                   const CurveGeometry* geom, const unsigned int primID,
+                                   const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3,
+                                   const Epilog& epilog)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Ray1 ray(vray,k);
+
+        /* move ray closer to make intersection stable */
+        NativeCurve3ff curve0(v0,v1,v2,v3);
+        curve0 = enlargeRadiusToMinWidth(context,geom,ray.org,curve0);
+        const float dt = dot(curve0.center()-ray.org,ray.dir)*rcp(dot(ray.dir,ray.dir));
+        const Vec3ff ref(madd(Vec3fa(dt),ray.dir,ray.org),0.0f);
+        const NativeCurve3ff curve1 = curve0-ref;
+        return intersect_bezier_recursive_jacobian(ray,dt,curve1,epilog);
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/curve_intersector_virtual.h

@@ -0,0 +1,671 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../subdiv/bezier_curve.h"
+#include "../builders/primref.h"
+#include "curve_intersector_precalculations.h"
+#include "../bvh/node_intersector1.h"
+#include "../bvh/node_intersector_packet.h"
+
+#include "intersector_epilog.h"
+
+#include "../subdiv/bezier_curve.h"
+#include "../subdiv/bspline_curve.h"
+#include "../subdiv/hermite_curve.h"
+#include "../subdiv/catmullrom_curve.h"
+
+#include "spherei_intersector.h"
+#include "disci_intersector.h"
+
+#include "linei_intersector.h"
+#include "roundlinei_intersector.h"
+#include "conelinei_intersector.h"
+
+#include "curveNi_intersector.h"
+#include "curveNv_intersector.h"
+#include "curveNi_mb_intersector.h"
+
+#include "curve_intersector_distance.h"
+#include "curve_intersector_ribbon.h"
+#include "curve_intersector_oriented.h"
+#include "curve_intersector_sweep.h"
+
+namespace embree
+{
+  struct VirtualCurveIntersector
+  {
+    typedef void (*Intersect1Ty)(void* pre, void* ray, RayQueryContext* context, const void* primitive);
+    typedef bool (*Occluded1Ty )(void* pre, void* ray, RayQueryContext* context, const void* primitive);
+    
+    typedef void (*Intersect4Ty)(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+    typedef bool (*Occluded4Ty) (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+    
+    typedef void (*Intersect8Ty)(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+    typedef bool (*Occluded8Ty) (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+    
+    typedef void (*Intersect16Ty)(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+    typedef bool (*Occluded16Ty) (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+
+  public:
+    struct Intersectors
+    {
+      Intersectors() {} // WARNING: Do not zero initialize this, as we otherwise get problems with thread unsafe local static variable initialization (e.g. on VS2013) in curve_intersector_virtual.cpp.
+      
+      template<int K> void intersect(void* pre, void* ray, RayQueryContext* context, const void* primitive);
+      template<int K> bool occluded (void* pre, void* ray, RayQueryContext* context, const void* primitive);
+
+      template<int K> void intersect(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+      template<int K> bool occluded (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive);
+
+    public:
+      Intersect1Ty intersect1;
+      Occluded1Ty  occluded1;
+      Intersect4Ty intersect4;
+      Occluded4Ty  occluded4;
+      Intersect8Ty intersect8;
+      Occluded8Ty  occluded8;
+      Intersect16Ty intersect16;
+      Occluded16Ty  occluded16;
+    };
+    
+    Intersectors vtbl[Geometry::GTY_END];
+  };
+
+  template<> __forceinline void VirtualCurveIntersector::Intersectors::intersect<1> (void* pre, void* ray, RayQueryContext* context, const void* primitive) { assert(intersect1); intersect1(pre,ray,context,primitive); }
+  template<> __forceinline bool VirtualCurveIntersector::Intersectors::occluded<1>  (void* pre, void* ray, RayQueryContext* context, const void* primitive) { assert(occluded1); return occluded1(pre,ray,context,primitive); }
+      
+  template<> __forceinline void VirtualCurveIntersector::Intersectors::intersect<4>(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(intersect4); intersect4(pre,ray,k,context,primitive); }
+  template<> __forceinline bool VirtualCurveIntersector::Intersectors::occluded<4> (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(occluded4); return occluded4(pre,ray,k,context,primitive); }
+      
+#if defined(__AVX__)
+  template<> __forceinline void VirtualCurveIntersector::Intersectors::intersect<8>(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(intersect8); intersect8(pre,ray,k,context,primitive); }
+  template<> __forceinline bool VirtualCurveIntersector::Intersectors::occluded<8> (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(occluded8); return occluded8(pre,ray,k,context,primitive); }
+#endif
+  
+#if defined(__AVX512F__)
+  template<> __forceinline void VirtualCurveIntersector::Intersectors::intersect<16>(void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(intersect16); intersect16(pre,ray,k,context,primitive); }
+  template<> __forceinline bool VirtualCurveIntersector::Intersectors::occluded<16> (void* pre, void* ray, size_t k, RayQueryContext* context, const void* primitive) { assert(occluded16); return occluded16(pre,ray,k,context,primitive); }
+#endif
+  
+  namespace isa
+  {
+    struct VirtualCurveIntersector1
+    {
+      typedef unsigned char Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+      
+      template<int N, bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        assert(num == 1);
+        RTCGeometryType ty = (RTCGeometryType)(*prim);
+        assert(This->leafIntersector);
+        VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+        leafIntersector.intersect<1>(&pre,&ray,context,prim);
+      }
+
+      template<int N, bool robust>      
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        assert(num == 1);
+        RTCGeometryType ty = (RTCGeometryType)(*prim);
+        assert(This->leafIntersector);
+        VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+        return leafIntersector.occluded<1>(&pre,&ray,context,prim);
+      }
+    };
+
+    template<int K>
+      struct VirtualCurveIntersectorK 
+      {
+        typedef unsigned char Primitive;
+        typedef CurvePrecalculationsK<K> Precalculations;
+        
+        template<bool robust>        
+        static __forceinline void intersect(const vbool<K>& valid_i, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+          size_t mask = movemask(valid_i);
+          while (mask) leafIntersector.intersect<K>(&pre,&ray,bscf(mask),context,prim);
+        }
+        
+        template<bool robust>        
+        static __forceinline vbool<K> occluded(const vbool<K>& valid_i, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+          vbool<K> valid_o = false;
+          size_t mask = movemask(valid_i);
+          while (mask) {
+            size_t k = bscf(mask);
+            if (leafIntersector.occluded<K>(&pre,&ray,k,context,prim))
+              set(valid_o, k);
+          }
+          return valid_o;
+        }
+        
+        template<int N, bool robust>              
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+          leafIntersector.intersect<K>(&pre,&ray,k,context,prim);
+        }
+        
+        template<int N, bool robust>      
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          assert(num == 1);
+          RTCGeometryType ty = (RTCGeometryType)(*prim);
+          assert(This->leafIntersector);
+          VirtualCurveIntersector::Intersectors& leafIntersector = ((VirtualCurveIntersector*) This->leafIntersector)->vtbl[ty];
+          return leafIntersector.occluded<K>(&pre,&ray,k,context,prim);
+        }
+      };
+
+    template<int N>
+    static VirtualCurveIntersector::Intersectors LinearRoundConeNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &RoundLinearCurveMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &RoundLinearCurveMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &RoundLinearCurveMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &RoundLinearCurveMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&RoundLinearCurveMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &RoundLinearCurveMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&RoundLinearCurveMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &RoundLinearCurveMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+
+    template<int N>
+    static VirtualCurveIntersector::Intersectors LinearConeNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &ConeCurveMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &ConeCurveMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &ConeCurveMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &ConeCurveMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&ConeCurveMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &ConeCurveMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&ConeCurveMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &ConeCurveMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+
+    template<int N>
+    static VirtualCurveIntersector::Intersectors LinearRoundConeNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &RoundLinearCurveMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &RoundLinearCurveMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &RoundLinearCurveMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &RoundLinearCurveMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&RoundLinearCurveMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &RoundLinearCurveMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&RoundLinearCurveMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &RoundLinearCurveMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+
+    template<int N>
+    static VirtualCurveIntersector::Intersectors LinearConeNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &ConeCurveMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &ConeCurveMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &ConeCurveMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &ConeCurveMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&ConeCurveMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &ConeCurveMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&ConeCurveMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &ConeCurveMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+
+
+    template<int N>
+      static VirtualCurveIntersector::Intersectors LinearRibbonNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &FlatLinearCurveMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &FlatLinearCurveMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &FlatLinearCurveMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &FlatLinearCurveMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&FlatLinearCurveMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &FlatLinearCurveMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&FlatLinearCurveMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &FlatLinearCurveMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors LinearRibbonNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &FlatLinearCurveMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &FlatLinearCurveMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &FlatLinearCurveMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &FlatLinearCurveMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&FlatLinearCurveMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &FlatLinearCurveMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&FlatLinearCurveMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &FlatLinearCurveMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors SphereNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &SphereMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &SphereMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &SphereMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &SphereMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&SphereMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &SphereMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&SphereMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &SphereMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors SphereNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &SphereMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &SphereMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &SphereMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &SphereMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&SphereMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &SphereMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&SphereMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &SphereMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors DiscNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &DiscMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &DiscMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &DiscMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &DiscMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&DiscMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &DiscMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&DiscMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &DiscMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors DiscNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &DiscMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &DiscMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &DiscMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &DiscMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&DiscMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &DiscMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&DiscMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &DiscMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors OrientedDiscNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &OrientedDiscMiIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &OrientedDiscMiIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &OrientedDiscMiIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &OrientedDiscMiIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&OrientedDiscMiIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &OrientedDiscMiIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&OrientedDiscMiIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &OrientedDiscMiIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<int N>
+      static VirtualCurveIntersector::Intersectors OrientedDiscNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &OrientedDiscMiMBIntersector1<N,true>::intersect;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &OrientedDiscMiMBIntersector1<N,true>::occluded;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &OrientedDiscMiMBIntersectorK<N,4,true>::intersect;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &OrientedDiscMiMBIntersectorK<N,4,true>::occluded;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&OrientedDiscMiMBIntersectorK<N,8,true>::intersect;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &OrientedDiscMiMBIntersectorK<N,8,true>::occluded;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&OrientedDiscMiMBIntersectorK<N,16,true>::intersect;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &OrientedDiscMiMBIntersectorK<N,16,true>::occluded;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors RibbonNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_t<RibbonCurve1Intersector1<Curve>, Intersect1EpilogMU<VSIZEX,true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_t <RibbonCurve1Intersector1<Curve>, Occluded1EpilogMU<VSIZEX,true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &CurveNiIntersectorK<N,4>::template intersect_t<RibbonCurve1IntersectorK<Curve,4>, Intersect1KEpilogMU<VSIZEX,4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &CurveNiIntersectorK<N,4>::template occluded_t <RibbonCurve1IntersectorK<Curve,4>, Occluded1KEpilogMU<VSIZEX,4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_t<RibbonCurve1IntersectorK<Curve,8>, Intersect1KEpilogMU<VSIZEX,8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_t <RibbonCurve1IntersectorK<Curve,8>, Occluded1KEpilogMU<VSIZEX,8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_t<RibbonCurve1IntersectorK<Curve,16>, Intersect1KEpilogMU<VSIZEX,16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_t <RibbonCurve1IntersectorK<Curve,16>, Occluded1KEpilogMU<VSIZEX,16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors RibbonNvIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNvIntersector1<N>::template intersect_t<RibbonCurve1Intersector1<Curve>, Intersect1EpilogMU<VSIZEX,true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNvIntersector1<N>::template occluded_t <RibbonCurve1Intersector1<Curve>, Occluded1EpilogMU<VSIZEX,true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &CurveNvIntersectorK<N,4>::template intersect_t<RibbonCurve1IntersectorK<Curve,4>, Intersect1KEpilogMU<VSIZEX,4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &CurveNvIntersectorK<N,4>::template occluded_t <RibbonCurve1IntersectorK<Curve,4>, Occluded1KEpilogMU<VSIZEX,4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNvIntersectorK<N,8>::template intersect_t<RibbonCurve1IntersectorK<Curve,8>, Intersect1KEpilogMU<VSIZEX,8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNvIntersectorK<N,8>::template occluded_t <RibbonCurve1IntersectorK<Curve,8>, Occluded1KEpilogMU<VSIZEX,8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNvIntersectorK<N,16>::template intersect_t<RibbonCurve1IntersectorK<Curve,16>, Intersect1KEpilogMU<VSIZEX,16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNvIntersectorK<N,16>::template occluded_t <RibbonCurve1IntersectorK<Curve,16>, Occluded1KEpilogMU<VSIZEX,16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors RibbonNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_t<RibbonCurve1Intersector1<Curve>, Intersect1EpilogMU<VSIZEX,true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_t <RibbonCurve1Intersector1<Curve>, Occluded1EpilogMU<VSIZEX,true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty) &CurveNiMBIntersectorK<N,4>::template intersect_t<RibbonCurve1IntersectorK<Curve,4>, Intersect1KEpilogMU<VSIZEX,4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty)  &CurveNiMBIntersectorK<N,4>::template occluded_t <RibbonCurve1IntersectorK<Curve,4>, Occluded1KEpilogMU<VSIZEX,4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_t<RibbonCurve1IntersectorK<Curve,8>, Intersect1KEpilogMU<VSIZEX,8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_t <RibbonCurve1IntersectorK<Curve,8>, Occluded1KEpilogMU<VSIZEX,8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_t<RibbonCurve1IntersectorK<Curve,16>, Intersect1KEpilogMU<VSIZEX,16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_t <RibbonCurve1IntersectorK<Curve,16>, Occluded1KEpilogMU<VSIZEX,16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors CurveNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_t<SweepCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_t <SweepCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiIntersectorK<N,4>::template intersect_t<SweepCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiIntersectorK<N,4>::template occluded_t <SweepCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_t<SweepCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_t <SweepCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_t<SweepCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_t <SweepCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors CurveNvIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNvIntersector1<N>::template intersect_t<SweepCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNvIntersector1<N>::template occluded_t <SweepCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNvIntersectorK<N,4>::template intersect_t<SweepCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNvIntersectorK<N,4>::template occluded_t <SweepCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNvIntersectorK<N,8>::template intersect_t<SweepCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNvIntersectorK<N,8>::template occluded_t <SweepCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNvIntersectorK<N,16>::template intersect_t<SweepCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNvIntersectorK<N,16>::template occluded_t <SweepCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors CurveNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_t<SweepCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_t <SweepCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiMBIntersectorK<N,4>::template intersect_t<SweepCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiMBIntersectorK<N,4>::template occluded_t <SweepCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_t<SweepCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_t <SweepCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_t<SweepCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_t <SweepCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors OrientedCurveNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_n<OrientedCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_n <OrientedCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiIntersectorK<N,4>::template intersect_n<OrientedCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiIntersectorK<N,4>::template occluded_n <OrientedCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_n<OrientedCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_n <OrientedCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_n<OrientedCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_n <OrientedCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors OrientedCurveNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_n<OrientedCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_n <OrientedCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiMBIntersectorK<N,4>::template intersect_n<OrientedCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiMBIntersectorK<N,4>::template occluded_n <OrientedCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_n<OrientedCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_n <OrientedCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_n<OrientedCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_n <OrientedCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteRibbonNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_h<RibbonCurve1Intersector1<Curve>, Intersect1EpilogMU<VSIZEX,true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_h <RibbonCurve1Intersector1<Curve>, Occluded1EpilogMU<VSIZEX,true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiIntersectorK<N,4>::template intersect_h<RibbonCurve1IntersectorK<Curve,4>, Intersect1KEpilogMU<VSIZEX,4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiIntersectorK<N,4>::template occluded_h <RibbonCurve1IntersectorK<Curve,4>, Occluded1KEpilogMU<VSIZEX,4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_h<RibbonCurve1IntersectorK<Curve,8>, Intersect1KEpilogMU<VSIZEX,8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_h <RibbonCurve1IntersectorK<Curve,8>, Occluded1KEpilogMU<VSIZEX,8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_h<RibbonCurve1IntersectorK<Curve,16>, Intersect1KEpilogMU<VSIZEX,16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_h <RibbonCurve1IntersectorK<Curve,16>, Occluded1KEpilogMU<VSIZEX,16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteRibbonNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_h<RibbonCurve1Intersector1<Curve>, Intersect1EpilogMU<VSIZEX,true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_h <RibbonCurve1Intersector1<Curve>, Occluded1EpilogMU<VSIZEX,true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiMBIntersectorK<N,4>::template intersect_h<RibbonCurve1IntersectorK<Curve,4>, Intersect1KEpilogMU<VSIZEX,4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiMBIntersectorK<N,4>::template occluded_h <RibbonCurve1IntersectorK<Curve,4>, Occluded1KEpilogMU<VSIZEX,4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_h<RibbonCurve1IntersectorK<Curve,8>, Intersect1KEpilogMU<VSIZEX,8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_h <RibbonCurve1IntersectorK<Curve,8>, Occluded1KEpilogMU<VSIZEX,8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_h<RibbonCurve1IntersectorK<Curve,16>, Intersect1KEpilogMU<VSIZEX,16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_h <RibbonCurve1IntersectorK<Curve,16>, Occluded1KEpilogMU<VSIZEX,16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteCurveNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_h<SweepCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_h <SweepCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiIntersectorK<N,4>::template intersect_h<SweepCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiIntersectorK<N,4>::template occluded_h <SweepCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_h<SweepCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_h <SweepCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_h<SweepCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_h <SweepCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteCurveNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_h<SweepCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_h <SweepCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiMBIntersectorK<N,4>::template intersect_h<SweepCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiMBIntersectorK<N,4>::template occluded_h <SweepCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_h<SweepCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_h <SweepCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_h<SweepCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_h <SweepCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteOrientedCurveNiIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiIntersector1<N>::template intersect_hn<OrientedCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiIntersector1<N>::template occluded_hn <OrientedCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiIntersectorK<N,4>::template intersect_hn<OrientedCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiIntersectorK<N,4>::template occluded_hn <OrientedCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiIntersectorK<N,8>::template intersect_hn<OrientedCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiIntersectorK<N,8>::template occluded_hn <OrientedCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiIntersectorK<N,16>::template intersect_hn<OrientedCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiIntersectorK<N,16>::template occluded_hn <OrientedCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+    
+    template<template<typename Ty> class Curve, int N>
+      static VirtualCurveIntersector::Intersectors HermiteOrientedCurveNiMBIntersectors()
+    {
+      VirtualCurveIntersector::Intersectors intersectors;
+      intersectors.intersect1 = (VirtualCurveIntersector::Intersect1Ty) &CurveNiMBIntersector1<N>::template intersect_hn<OrientedCurve1Intersector1<Curve>, Intersect1Epilog1<true> >;
+      intersectors.occluded1  = (VirtualCurveIntersector::Occluded1Ty)  &CurveNiMBIntersector1<N>::template occluded_hn <OrientedCurve1Intersector1<Curve>, Occluded1Epilog1<true> >;
+      intersectors.intersect4 = (VirtualCurveIntersector::Intersect4Ty)&CurveNiMBIntersectorK<N,4>::template intersect_hn<OrientedCurve1IntersectorK<Curve,4>, Intersect1KEpilog1<4,true> >;
+      intersectors.occluded4  = (VirtualCurveIntersector::Occluded4Ty) &CurveNiMBIntersectorK<N,4>::template occluded_hn <OrientedCurve1IntersectorK<Curve,4>, Occluded1KEpilog1<4,true> >;
+#if defined(__AVX__)
+      intersectors.intersect8 = (VirtualCurveIntersector::Intersect8Ty)&CurveNiMBIntersectorK<N,8>::template intersect_hn<OrientedCurve1IntersectorK<Curve,8>, Intersect1KEpilog1<8,true> >;
+      intersectors.occluded8  = (VirtualCurveIntersector::Occluded8Ty) &CurveNiMBIntersectorK<N,8>::template occluded_hn <OrientedCurve1IntersectorK<Curve,8>, Occluded1KEpilog1<8,true> >;
+#endif
+#if defined(__AVX512F__)
+      intersectors.intersect16 = (VirtualCurveIntersector::Intersect16Ty)&CurveNiMBIntersectorK<N,16>::template intersect_hn<OrientedCurve1IntersectorK<Curve,16>, Intersect1KEpilog1<16,true> >;
+      intersectors.occluded16  = (VirtualCurveIntersector::Occluded16Ty) &CurveNiMBIntersectorK<N,16>::template occluded_hn <OrientedCurve1IntersectorK<Curve,16>, Occluded1KEpilog1<16,true> >;
+#endif
+      return intersectors;
+    }
+  }
+}

thirdparty/embree/kernels/geometry/cylinder.h

@@ -0,0 +1,223 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct Cylinder
+    {
+      const Vec3fa p0;  //!< start location
+      const Vec3fa p1;  //!< end position
+      const float rr;   //!< squared radius of cylinder
+
+      __forceinline Cylinder(const Vec3fa& p0, const Vec3fa& p1, const float r) 
+        : p0(p0), p1(p1), rr(sqr(r)) {}
+
+      __forceinline Cylinder(const Vec3fa& p0, const Vec3fa& p1, const float rr, bool) 
+        : p0(p0), p1(p1), rr(rr) {}
+
+      __forceinline bool intersect(const Vec3fa& org,
+                                   const Vec3fa& dir, 
+                                   BBox1f& t_o, 
+                                   float& u0_o, Vec3fa& Ng0_o,
+                                   float& u1_o, Vec3fa& Ng1_o) const
+      {
+        /* calculate quadratic equation to solve */
+        const float rl = rcp_length(p1-p0);
+        const Vec3fa P0 = p0, dP = (p1-p0)*rl;
+        const Vec3fa O = org-P0, dO = dir;
+        
+        const float dOdO = dot(dO,dO);
+        const float OdO = dot(dO,O);
+        const float OO = dot(O,O);
+        const float dOz = dot(dP,dO);
+        const float Oz = dot(dP,O);
+        
+        const float A = dOdO - sqr(dOz);
+        const float B = 2.0f * (OdO - dOz*Oz);
+        const float C = OO - sqr(Oz) - rr;
+        
+        /* we miss the cylinder if determinant is smaller than zero */
+        const float D = B*B - 4.0f*A*C;
+        if (D < 0.0f) {
+          t_o = BBox1f(pos_inf,neg_inf);
+          return false;
+        }
+        
+        /* special case for rays that are parallel to the cylinder */
+        const float eps = 16.0f*float(ulp)*max(abs(dOdO),abs(sqr(dOz)));
+        if (abs(A) < eps) 
+        {
+          if (C <= 0.0f) {
+            t_o = BBox1f(neg_inf,pos_inf);
+            return true;
+          } else {
+            t_o = BBox1f(pos_inf,neg_inf);
+            return false;
+          }
+        }
+        
+        /* standard case for rays that are not parallel to the cylinder */
+        const float Q = sqrt(D);
+        const float rcp_2A = rcp(2.0f*A);
+        const float t0 = (-B-Q)*rcp_2A;
+        const float t1 = (-B+Q)*rcp_2A;
+        
+        /* calculates u and Ng for near hit */
+        {
+          u0_o = madd(t0,dOz,Oz)*rl;
+          const Vec3fa Pr = t0*dir;
+          const Vec3fa Pl = madd(u0_o,p1-p0,p0);
+          Ng0_o = Pr-Pl;
+        }
+
+        /* calculates u and Ng for far hit */
+        {
+          u1_o = madd(t1,dOz,Oz)*rl;
+          const Vec3fa Pr = t1*dir;
+          const Vec3fa Pl = madd(u1_o,p1-p0,p0);
+          Ng1_o = Pr-Pl;
+        }
+
+        t_o.lower = t0;
+        t_o.upper = t1;
+        return true;
+      }
+
+      __forceinline bool intersect(const Vec3fa& org_i, const Vec3fa& dir, BBox1f& t_o) const
+      {
+        float u0_o; Vec3fa Ng0_o;
+        float u1_o; Vec3fa Ng1_o;
+        return intersect(org_i,dir,t_o,u0_o,Ng0_o,u1_o,Ng1_o);
+      }
+
+      static bool verify(const size_t id, const Cylinder& cylinder, const RayHit& ray, bool shouldhit, const float t0, const float t1)
+      {
+        float eps = 0.001f;
+        BBox1f t; bool hit;
+        hit = cylinder.intersect(ray.org,ray.dir,t);
+
+        bool failed = hit != shouldhit;
+        if (shouldhit) failed |= std::isinf(t0) ? t0 != t.lower : abs(t0-t.lower) > eps;
+        if (shouldhit) failed |= std::isinf(t1) ? t1 != t.upper : abs(t1-t.upper) > eps;
+        if (!failed) return true;
+        embree_cout << "Cylinder test " << id << " failed: cylinder = " << cylinder << ", ray = " << ray << ", hit = " << hit << ", t = " << t << embree_endl; 
+        return false;
+      }
+
+      /* verify cylinder class */
+      static bool verify()
+      {
+        bool passed = true;
+        const Cylinder cylinder(Vec3fa(0.0f,0.0f,0.0f),Vec3fa(1.0f,0.0f,0.0f),1.0f);
+        passed &= verify(0,cylinder,RayHit(Vec3fa(-2.0f,1.0f,0.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),true,0.0f,2.0f);
+        passed &= verify(1,cylinder,RayHit(Vec3fa(+2.0f,1.0f,0.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),true,0.0f,2.0f);
+        passed &= verify(2,cylinder,RayHit(Vec3fa(+2.0f,1.0f,2.0f),Vec3fa( 0.0f,-1.0f,+0.0f),0.0f,float(inf)),false,0.0f,0.0f);
+        passed &= verify(3,cylinder,RayHit(Vec3fa(+0.0f,0.0f,0.0f),Vec3fa( 1.0f, 0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,pos_inf);
+        passed &= verify(4,cylinder,RayHit(Vec3fa(+0.0f,0.0f,0.0f),Vec3fa(-1.0f, 0.0f,+0.0f),0.0f,float(inf)),true,neg_inf,pos_inf);
+        passed &= verify(5,cylinder,RayHit(Vec3fa(+0.0f,2.0f,0.0f),Vec3fa( 1.0f, 0.0f,+0.0f),0.0f,float(inf)),false,pos_inf,neg_inf);
+        passed &= verify(6,cylinder,RayHit(Vec3fa(+0.0f,2.0f,0.0f),Vec3fa(-1.0f, 0.0f,+0.0f),0.0f,float(inf)),false,pos_inf,neg_inf);
+        return passed;
+      }
+
+      /*! output operator */
+      friend __forceinline embree_ostream operator<<(embree_ostream cout, const Cylinder& c) {
+        return cout << "Cylinder { p0 = " << c.p0 << ", p1 = " << c.p1 << ", r = " << sqrtf(c.rr) << "}";
+      }
+    };
+
+    template<int N>
+      struct CylinderN
+    { 
+      const Vec3vf<N> p0;     //!< start location
+      const Vec3vf<N> p1;     //!< end position
+      const vfloat<N> rr;   //!< squared radius of cylinder
+
+      __forceinline CylinderN(const Vec3vf<N>& p0, const Vec3vf<N>& p1, const vfloat<N>& r)
+        : p0(p0), p1(p1), rr(sqr(r)) {}
+
+      __forceinline CylinderN(const Vec3vf<N>& p0, const Vec3vf<N>& p1, const vfloat<N>& rr, bool)
+        : p0(p0), p1(p1), rr(rr) {}
+
+     
+      __forceinline vbool<N> intersect(const Vec3fa& org, const Vec3fa& dir, 
+                                       BBox<vfloat<N>>& t_o, 
+                                       vfloat<N>& u0_o, Vec3vf<N>& Ng0_o,
+                                       vfloat<N>& u1_o, Vec3vf<N>& Ng1_o) const
+      {
+        /* calculate quadratic equation to solve */
+        const vfloat<N> rl = rcp_length(p1-p0);
+        const Vec3vf<N> P0 = p0, dP = (p1-p0)*rl;
+        const Vec3vf<N> O = Vec3vf<N>(org)-P0, dO = dir;
+        
+        const vfloat<N> dOdO = dot(dO,dO);
+        const vfloat<N> OdO = dot(dO,O);
+        const vfloat<N> OO = dot(O,O);
+        const vfloat<N> dOz = dot(dP,dO);
+        const vfloat<N> Oz = dot(dP,O);
+        
+        const vfloat<N> A = dOdO - sqr(dOz);
+        const vfloat<N> B = 2.0f * (OdO - dOz*Oz);
+        const vfloat<N> C = OO - sqr(Oz) - rr;
+        
+        /* we miss the cylinder if determinant is smaller than zero */
+        const vfloat<N> D = B*B - 4.0f*A*C;
+        vbool<N> valid = D >= 0.0f;
+        if (none(valid)) {
+          t_o = BBox<vfloat<N>>(empty);
+          return valid;
+        }
+
+        /* standard case for rays that are not parallel to the cylinder */
+        const vfloat<N> Q = sqrt(D);
+        const vfloat<N> rcp_2A = rcp(2.0f*A);
+        const vfloat<N> t0 = (-B-Q)*rcp_2A;
+        const vfloat<N> t1 = (-B+Q)*rcp_2A;
+        
+        /* calculates u and Ng for near hit */
+        {
+          u0_o = madd(t0,dOz,Oz)*rl;
+          const Vec3vf<N> Pr = t0*Vec3vf<N>(dir);
+          const Vec3vf<N> Pl = madd(u0_o,p1-p0,p0);
+          Ng0_o = Pr-Pl;
+        }
+        
+        /* calculates u and Ng for far hit */
+        {
+          u1_o = madd(t1,dOz,Oz)*rl;
+          const Vec3vf<N> Pr = t1*Vec3vf<N>(dir);
+          const Vec3vf<N> Pl = madd(u1_o,p1-p0,p0);
+          Ng1_o = Pr-Pl;
+        }
+
+        t_o.lower = select(valid, t0, vfloat<N>(pos_inf));
+        t_o.upper = select(valid, t1, vfloat<N>(neg_inf));
+
+        /* special case for rays that are parallel to the cylinder */
+        const vfloat<N> eps = 16.0f*float(ulp)*max(abs(dOdO),abs(sqr(dOz)));
+        vbool<N> validt = valid & (abs(A) < eps); 
+        if (unlikely(any(validt))) 
+        {
+          vbool<N> inside = C <= 0.0f;
+          t_o.lower = select(validt,select(inside,vfloat<N>(neg_inf),vfloat<N>(pos_inf)),t_o.lower);
+          t_o.upper = select(validt,select(inside,vfloat<N>(pos_inf),vfloat<N>(neg_inf)),t_o.upper);
+          valid &= !validt | inside;
+        }
+        return valid;
+      }
+
+      __forceinline vbool<N> intersect(const Vec3fa& org_i, const Vec3fa& dir, BBox<vfloat<N>>& t_o) const
+      {
+        vfloat<N> u0_o; Vec3vf<N> Ng0_o;
+        vfloat<N> u1_o; Vec3vf<N> Ng1_o;
+        return intersect(org_i,dir,t_o,u0_o,Ng0_o,u1_o,Ng1_o);
+      }
+    };
+  }
+}
+

thirdparty/embree/kernels/geometry/disc_intersector.h

@@ -0,0 +1,269 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/scene_points.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+    struct DiscIntersectorHitM
+    {
+      __forceinline DiscIntersectorHitM() {}
+
+      __forceinline DiscIntersectorHitM(const vfloat<M>& u, const vfloat<M>& v, const vfloat<M>& t, const Vec3vf<M>& Ng)
+          : vu(u), vv(v), vt(t), vNg(Ng)
+      {
+      }
+
+      __forceinline void finalize() {}
+
+      __forceinline Vec2f uv(const size_t i) const {
+        return Vec2f(vu[i], vv[i]);
+      }
+      __forceinline Vec2vf<M> uv() const {
+        return Vec2vf<M>(vu, vv);
+      }
+      __forceinline float t(const size_t i) const {
+        return vt[i];
+      }
+      __forceinline vfloat<M> t() const {
+        return vt;
+      }
+      __forceinline Vec3fa Ng(const size_t i) const {
+        return Vec3fa(vNg.x[i], vNg.y[i], vNg.z[i]);
+      }
+      __forceinline Vec3vf<M> Ng() const { 
+        return vNg;
+      }
+
+     public:
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+    };
+
+    template<>
+    struct DiscIntersectorHitM<1>
+    {
+      __forceinline DiscIntersectorHitM() {}
+
+      __forceinline DiscIntersectorHitM(const float& u, const float& v, const float& t, const Vec3fa& Ng)
+          : vu(u), vv(v), vt(t), vNg(Ng) {}
+
+      __forceinline void finalize() {}
+
+      __forceinline Vec2f uv() const {
+        return Vec2f(vu, vv);
+      }
+
+      __forceinline float t() const {
+        return vt;
+      }
+
+      __forceinline Vec3fa Ng() const { 
+        return vNg;
+      }
+
+     public:
+      float vu;
+      float vv;
+      float vt;
+      Vec3fa vNg;
+    };
+
+    template<int M>
+    struct DiscIntersector1
+    {
+      typedef CurvePrecalculations1 Precalculations;
+
+      template<typename Ray, typename Epilog>
+      static __forceinline bool intersect(
+          const vbool<M>& valid_i,
+          Ray& ray,
+          RayQueryContext* context,
+          const Points* geom,
+          const Precalculations& pre,
+          const Vec4vf<M>& v0i,
+          const Epilog& epilog)
+      {
+        vbool<M> valid = valid_i;
+
+        const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+        const Vec3vf<M> ray_dir(ray.dir.x, ray.dir.y, ray.dir.z);
+        const vfloat<M> rd2    = rcp(dot(ray_dir, ray_dir));
+
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+
+        /* compute ray distance projC0 to hit point with ray oriented plane */
+        const Vec3vf<M> c0     = center - ray_org;
+        const vfloat<M> projC0 = dot(c0, ray_dir) * rd2;
+
+        valid &= (vfloat<M>(ray.tnear()) <= projC0) & (projC0 <= vfloat<M>(ray.tfar));
+        if (unlikely(none(valid)))
+          return false;
+
+        /* check if hit point lies inside disc */
+        const Vec3vf<M> perp   = c0 - projC0 * ray_dir;
+        const vfloat<M> l2     = dot(perp, perp);
+        const vfloat<M> r2     = radius * radius;
+        valid &= (l2 <= r2);
+        if (unlikely(none(valid)))
+          return false;
+
+        /* We reject hits where the ray origin lies inside the ray
+         * oriented disc to avoid self intersections. */
+#if defined(EMBREE_DISC_POINT_SELF_INTERSECTION_AVOIDANCE)
+        const vfloat<M> m2 = dot(c0, c0);
+        valid &= (m2 > r2);
+        if (unlikely(none(valid)))
+          return false;
+#endif
+        
+        DiscIntersectorHitM<M> hit(zero, zero, projC0, -ray_dir);
+        return epilog(valid, hit);
+      }
+
+      template<typename Ray, typename Epilog>
+      static __forceinline bool intersect(const vbool<M>& valid_i,
+                                          Ray& ray,
+                                          RayQueryContext* context,
+                                          const Points* geom,
+                                          const Precalculations& pre,
+                                          const Vec4vf<M>& v0i,
+                                          const Vec3vf<M>& normal,
+                                          const Epilog& epilog)
+      {
+        vbool<M> valid         = valid_i;
+        const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+
+        vfloat<M> divisor       = dot(Vec3vf<M>((Vec3fa)ray.dir), normal);
+        const vbool<M> parallel = divisor == vfloat<M>(0.f);
+        valid &= !parallel;
+        divisor = select(parallel, 1.f, divisor);  // prevent divide by zero
+
+        vfloat<M> t = dot(center - Vec3vf<M>((Vec3fa)ray.org), Vec3vf<M>(normal)) / divisor;
+
+        valid &= (vfloat<M>(ray.tnear()) <= t) & (t <= vfloat<M>(ray.tfar));
+        if (unlikely(none(valid)))
+          return false;
+
+        Vec3vf<M> intersection = Vec3vf<M>((Vec3fa)ray.org) + Vec3vf<M>((Vec3fa)ray.dir) * t;
+        vfloat<M> dist2        = dot(intersection - center, intersection - center);
+        valid &= dist2 < radius * radius;
+        if (unlikely(none(valid)))
+          return false;
+
+        DiscIntersectorHitM<M> hit(zero, zero, t, normal);
+        return epilog(valid, hit);
+      }
+    };
+
+    template<int M, int K>
+    struct DiscIntersectorK
+    {
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      template<typename Epilog>
+      static __forceinline bool intersect(const vbool<M>& valid_i,
+                                          RayK<K>& ray,
+                                          size_t k,
+                                          RayQueryContext* context,
+                                          const Points* geom,
+                                          const Precalculations& pre,
+                                          const Vec4vf<M>& v0i,
+                                          const Epilog& epilog)
+      {
+        vbool<M> valid = valid_i;
+
+        const Vec3vf<M> ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+        const Vec3vf<M> ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]);
+        const vfloat<M> rd2    = rcp(dot(ray_dir, ray_dir));
+
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+
+        /* compute ray distance projC0 to hit point with ray oriented plane */
+        const Vec3vf<M> c0     = center - ray_org;
+        const vfloat<M> projC0 = dot(c0, ray_dir) * rd2;
+
+        valid &= (vfloat<M>(ray.tnear()[k]) <= projC0) & (projC0 <= vfloat<M>(ray.tfar[k]));
+        if (unlikely(none(valid)))
+          return false;
+
+        /* check if hit point lies inside disc */
+        const Vec3vf<M> perp   = c0 - projC0 * ray_dir;
+        const vfloat<M> l2     = dot(perp, perp);
+        const vfloat<M> r2     = radius * radius;
+        valid &= (l2 <= r2);
+        if (unlikely(none(valid)))
+          return false;
+
+        /* We reject hits where the ray origin lies inside the ray
+         * oriented disc to avoid self intersections. */
+#if defined(EMBREE_DISC_POINT_SELF_INTERSECTION_AVOIDANCE)
+        const vfloat<M> m2 = dot(c0, c0);
+        valid &= (m2 > r2);
+        if (unlikely(none(valid)))
+          return false;
+#endif
+
+        DiscIntersectorHitM<M> hit(zero, zero, projC0, -ray_dir);
+        return epilog(valid, hit);
+      }
+
+      template<typename Epilog>
+      static __forceinline bool intersect(const vbool<M>& valid_i,
+                                          RayK<K>& ray,
+                                          size_t k,
+                                          RayQueryContext* context,
+                                          const Points* geom,
+                                          const Precalculations& pre,
+                                          const Vec4vf<M>& v0i,
+                                          const Vec3vf<M>& normal,
+                                          const Epilog& epilog)
+      {
+        vbool<M> valid         = valid_i;
+        const Vec3vf<M> ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+        const Vec3vf<M> ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]);
+
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+        
+        vfloat<M> divisor       = dot(Vec3vf<M>(ray_dir), normal);
+        const vbool<M> parallel = divisor == vfloat<M>(0.f);
+        valid &= !parallel;
+        divisor = select(parallel, 1.f, divisor);  // prevent divide by zero
+
+        vfloat<M> t = dot(center - Vec3vf<M>(ray_org), Vec3vf<M>(normal)) / divisor;
+
+        valid &= (vfloat<M>(ray.tnear()[k]) <= t) & (t <= vfloat<M>(ray.tfar[k]));
+        if (unlikely(none(valid)))
+          return false;
+
+        Vec3vf<M> intersection = Vec3vf<M>(ray_org) + Vec3vf<M>(ray_dir) * t;
+        vfloat<M> dist2        = dot(intersection - center, intersection - center);
+        valid &= dist2 < radius * radius;
+        if (unlikely(none(valid)))
+          return false;
+
+        DiscIntersectorHitM<M> hit(zero, zero, t, normal);
+        return epilog(valid, hit);
+      }
+    };
+  }  // namespace isa
+}  // namespace embree

thirdparty/embree/kernels/geometry/disci_intersector.h

@@ -0,0 +1,277 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "disc_intersector.h"
+#include "intersector_epilog.h"
+#include "pointi.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, bool filter>
+    struct DiscMiIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Intersect1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Occluded1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, bool filter>
+    struct DiscMiMBIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom, ray.time());
+        const vbool<M> valid = Disc.valid();
+        DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Intersect1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom, ray.time());
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Occluded1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct DiscMiIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersectorK<M, K>::intersect(
+            valid, ray, k, context, geom, pre, v0,
+            Intersect1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Occluded1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct DiscMiMBIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom, ray.time()[k]);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Intersect1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Disc.gather(v0, geom, ray.time()[k]);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0, Occluded1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, bool filter>
+    struct OrientedDiscMiIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, n0, Intersect1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, n0, Occluded1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, bool filter>
+    struct OrientedDiscMiMBIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom, ray.time());
+        const vbool<M> valid = Disc.valid();
+        DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, n0, Intersect1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom, ray.time());
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, n0, Occluded1EpilogM<M, filter>(ray, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct OrientedDiscMiIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersectorK<M, K>::intersect(
+            valid, ray, k, context, geom, pre, v0, n0,
+            Intersect1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersectorK<M, K>::intersect(
+            valid, ray, k, context, geom, pre, v0, n0,
+            Occluded1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct OrientedDiscMiMBIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom, ray.time()[k]);
+        const vbool<M> valid = Disc.valid();
+        DiscIntersectorK<M, K>::intersect(
+            valid, ray, k, context, geom, pre, v0, n0,
+            Intersect1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& Disc)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(Disc.geomID());
+        Vec4vf<M> v0; Vec3vf<M> n0;
+        Disc.gather(v0, n0, geom, ray.time()[k]);
+        const vbool<M> valid = Disc.valid();
+        return DiscIntersectorK<M, K>::intersect(
+            valid, ray, k, context, geom, pre, v0, n0,
+            Occluded1KEpilogM<M, K, filter>(ray, k, context, Disc.geomID(), Disc.primID()));
+      }
+    };
+  }  // namespace isa
+}  // namespace embree

thirdparty/embree/kernels/geometry/filter.h

@@ -0,0 +1,157 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/geometry.h"
+#include "../common/ray.h"
+#include "../common/hit.h"
+#include "../common/context.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    __forceinline bool runIntersectionFilter1Helper(RTCFilterFunctionNArguments* args, const Geometry* const geometry, RayQueryContext* context)
+    {
+      if (geometry->intersectionFilterN)
+      {
+        geometry->intersectionFilterN(args);
+
+        if (args->valid[0] == 0)
+          return false;
+      }
+
+      if (context->getFilter())
+      {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          context->getFilter()(args);
+
+        if (args->valid[0] == 0)
+          return false;
+      }
+      
+      copyHitToRay(*(RayHit*)args->ray,*(Hit*)args->hit);
+      return true;
+    }
+    
+    __forceinline bool runIntersectionFilter1(const Geometry* const geometry, RayHit& ray, RayQueryContext* context, Hit& hit)
+    {
+      RTCFilterFunctionNArguments args;
+      int mask = -1;
+      args.valid = &mask;
+      args.geometryUserPtr = geometry->userPtr;
+      args.context = context->user;
+      args.ray = (RTCRayN*)&ray;
+      args.hit = (RTCHitN*)&hit;
+      args.N = 1;
+      return runIntersectionFilter1Helper(&args,geometry,context);
+    }
+
+    __forceinline bool runOcclusionFilter1Helper(RTCFilterFunctionNArguments* args, const Geometry* const geometry, RayQueryContext* context)
+    {
+      if (geometry->occlusionFilterN)
+      {
+        geometry->occlusionFilterN(args);
+
+        if (args->valid[0] == 0)
+          return false;
+      }
+
+      if (context->getFilter())
+      {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          context->getFilter()(args);
+
+        if (args->valid[0] == 0)
+          return false;
+      }
+
+      return true;
+    }
+
+    __forceinline bool runOcclusionFilter1(const Geometry* const geometry, Ray& ray, RayQueryContext* context, Hit& hit)
+    {
+      RTCFilterFunctionNArguments args;
+      int mask = -1;
+      args.valid = &mask;
+      args.geometryUserPtr = geometry->userPtr;
+      args.context = context->user;
+      args.ray = (RTCRayN*)&ray;
+      args.hit = (RTCHitN*)&hit;
+      args.N = 1;
+      return runOcclusionFilter1Helper(&args,geometry,context);
+    }
+
+    template<int K>
+      __forceinline vbool<K> runIntersectionFilterHelper(RTCFilterFunctionNArguments* args, const Geometry* const geometry, RayQueryContext* context)
+    {
+      vint<K>* mask = (vint<K>*) args->valid;
+      if (geometry->intersectionFilterN)
+        geometry->intersectionFilterN(args);
+      
+      vbool<K> valid_o = *mask != vint<K>(zero);
+      if (none(valid_o)) return valid_o;
+
+      if (context->getFilter()) {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          context->getFilter()(args);
+      }
+
+      valid_o = *mask != vint<K>(zero);
+      if (none(valid_o)) return valid_o;
+      
+      copyHitToRay(valid_o,*(RayHitK<K>*)args->ray,*(HitK<K>*)args->hit);
+      return valid_o;
+    }
+    
+    template<int K>
+    __forceinline vbool<K> runIntersectionFilter(const vbool<K>& valid, const Geometry* const geometry, RayHitK<K>& ray, RayQueryContext* context, HitK<K>& hit)
+    {
+      RTCFilterFunctionNArguments args;
+      vint<K> mask = valid.mask32();
+      args.valid = (int*)&mask;
+      args.geometryUserPtr = geometry->userPtr;
+      args.context = context->user;
+      args.ray = (RTCRayN*)&ray;
+      args.hit = (RTCHitN*)&hit;
+      args.N = K;
+      return runIntersectionFilterHelper<K>(&args,geometry,context);
+    }
+
+    template<int K>
+      __forceinline vbool<K> runOcclusionFilterHelper(RTCFilterFunctionNArguments* args, const Geometry* const geometry, RayQueryContext* context)
+    {
+      vint<K>* mask = (vint<K>*) args->valid;
+      if (geometry->occlusionFilterN)
+        geometry->occlusionFilterN(args);
+      
+      vbool<K> valid_o = *mask != vint<K>(zero);
+      if (none(valid_o)) return valid_o;
+
+      if (context->getFilter()) {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          context->getFilter()(args);
+      }
+      valid_o = *mask != vint<K>(zero);
+
+      RayK<K>* ray = (RayK<K>*) args->ray;
+      ray->tfar = select(valid_o, vfloat<K>(neg_inf), ray->tfar);
+      return valid_o;
+    }
+
+    template<int K>
+      __forceinline vbool<K> runOcclusionFilter(const vbool<K>& valid, const Geometry* const geometry, RayK<K>& ray, RayQueryContext* context, HitK<K>& hit)
+    {
+      RTCFilterFunctionNArguments args;
+      vint<K> mask = valid.mask32();
+      args.valid = (int*)&mask;
+      args.geometryUserPtr = geometry->userPtr;
+      args.context = context->user;
+      args.ray = (RTCRayN*)&ray;
+      args.hit = (RTCHitN*)&hit;
+      args.N = K;
+      return runOcclusionFilterHelper<K>(&args,geometry,context);
+    }
+  }
+}

thirdparty/embree/kernels/geometry/filter_sycl.h

@@ -0,0 +1,109 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/geometry.h"
+#include "../common/ray.h"
+#include "../common/hit.h"
+#include "../common/context.h"
+
+namespace embree
+{
+  __forceinline bool runIntersectionFilter1Helper(RTCFilterFunctionNArguments* args, int& mask, const Geometry* const geometry, RayQueryContext* context)
+  {
+    typedef void (*RTCFilterFunctionSYCL)(const void* args);
+    const RTCFeatureFlags feature_mask MAYBE_UNUSED = context->args->feature_mask;
+    
+#if EMBREE_SYCL_GEOMETRY_CALLBACK
+    if (feature_mask & RTC_FEATURE_FLAG_FILTER_FUNCTION_IN_GEOMETRY)
+    {
+      RTCFilterFunctionSYCL gfilter = (RTCFilterFunctionSYCL) geometry->intersectionFilterN;
+      if (gfilter)
+      {
+        gfilter(args);
+        
+        if (mask == 0)
+          return false;
+      }
+    }
+#endif
+
+    if (feature_mask & RTC_FEATURE_FLAG_FILTER_FUNCTION_IN_ARGUMENTS)
+    {
+      RTCFilterFunctionSYCL cfilter = (RTCFilterFunctionSYCL) context->args->filter;
+      if (cfilter)
+      {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          cfilter(args);
+        
+        if (mask == 0)
+          return false;
+      }
+    }
+    
+    return true;
+  }
+
+  __forceinline bool runOcclusionFilter1Helper(RTCFilterFunctionNArguments* args, int& mask, const Geometry* const geometry, RayQueryContext* context)
+  {
+    typedef void (*RTCFilterFunctionSYCL)(const void* args);
+    const RTCFeatureFlags feature_mask MAYBE_UNUSED = context->args->feature_mask;
+    
+#if EMBREE_SYCL_GEOMETRY_CALLBACK
+    if (feature_mask & RTC_FEATURE_FLAG_FILTER_FUNCTION_IN_GEOMETRY)
+    {
+      RTCFilterFunctionSYCL gfilter = (RTCFilterFunctionSYCL) geometry->occlusionFilterN;
+      if (gfilter)
+      {
+        gfilter(args);
+        
+        if (mask == 0)
+          return false;
+      }
+    }
+#endif
+
+    if (feature_mask & RTC_FEATURE_FLAG_FILTER_FUNCTION_IN_ARGUMENTS)
+    {
+      RTCFilterFunctionSYCL cfilter = (RTCFilterFunctionSYCL) context->args->filter;
+      if (cfilter)
+      {
+        if (context->enforceArgumentFilterFunction() || geometry->hasArgumentFilterFunctions())
+          cfilter(args);
+        
+        if (mask == 0)
+          return false;
+      }
+    }
+
+    return true;
+  }
+  
+  __forceinline bool runIntersectionFilter1SYCL(Geometry* geometry, RayHit& ray, sycl::private_ptr<RayQueryContext> context, Hit& hit)
+  {
+    RTCFilterFunctionNArguments args;
+    int mask = -1;
+    args.valid = &mask;
+    args.geometryUserPtr = geometry->userPtr;
+    args.context = context->user;
+    args.ray = (RTCRayN*) &ray;
+    args.hit = (RTCHitN*) &hit;
+    args.N = 1;
+    return runIntersectionFilter1Helper(&args,mask,geometry,context);
+  }
+
+
+  __forceinline bool runIntersectionFilter1SYCL(Geometry* geometry, Ray& ray, sycl::private_ptr<RayQueryContext> context, Hit& hit)
+  {
+    RTCFilterFunctionNArguments args;
+    int mask = -1;
+    args.valid = &mask;
+    args.geometryUserPtr = geometry->userPtr;
+    args.context = context->user;
+    args.ray = (RTCRayN*) &ray;
+    args.hit = (RTCHitN*) &hit;
+    args.N = 1;
+    return runOcclusionFilter1Helper(&args,mask,geometry,context);
+  }
+}

thirdparty/embree/kernels/geometry/grid_intersector.h

@@ -0,0 +1,99 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "grid_soa.h"
+#include "grid_soa_intersector1.h"
+#include "grid_soa_intersector_packet.h"
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename T>
+      class SubdivPatch1Precalculations : public T
+    { 
+    public:
+      __forceinline SubdivPatch1Precalculations (const Ray& ray, const void* ptr)
+        : T(ray,ptr) {}
+    };
+
+    template<int K, typename T>
+      class SubdivPatch1PrecalculationsK : public T
+    { 
+    public:
+      __forceinline SubdivPatch1PrecalculationsK (const vbool<K>& valid, RayK<K>& ray)
+        : T(valid,ray) {}
+    };
+
+    class Grid1Intersector1
+    {
+    public:
+      typedef GridSOA Primitive;
+      typedef Grid1Precalculations<GridSOAIntersector1::Precalculations> Precalculations;
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node) 
+      {
+        GridSOAIntersector1::intersect(pre,ray,context,prim,lazy_node);
+      }
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, size_t& lazy_node) {
+        intersect(pre,ray,context,prim,ty,lazy_node);
+      }
+      
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline bool occluded(Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node)
+      {
+        GridSOAIntersector1::occluded(pre,ray,context,prim,lazy_node);
+      }
+      static __forceinline bool occluded(Precalculations& pre, Ray& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, size_t& lazy_node) {
+        return occluded(pre,ray,context,prim,ty,lazy_node);
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node) {
+        assert(false && "not implemented");
+        return false;
+      }
+
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, size_t& lazy_node) {
+        assert(false && "not implemented");
+        return false;
+      }
+    };
+
+    template <int K>
+      struct GridIntersectorK
+    {
+      typedef GridSOA Primitive;
+      typedef SubdivPatch1PrecalculationsK<K,typename GridSOAIntersectorK<K>::Precalculations> Precalculations;
+      
+      
+      static __forceinline void intersect(const vbool<K>& valid, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node)
+      {
+        GridSOAIntersectorK<K>::intersect(valid,pre,ray,context,prim,lazy_node);
+      }
+      
+      static __forceinline vbool<K> occluded(const vbool<K>& valid, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node)
+      {
+        GridSOAIntersectorK<K>::occluded(valid,pre,ray,context,prim,lazy_node);
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node)
+      {
+        GridSOAIntersectorK<K>::intersect(pre,ray,k,context,prim,lazy_node);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, size_t& lazy_node)
+      {
+        GridSOAIntersectorK<K>::occluded(pre,ray,k,context,prim,lazy_node);
+      }
+    };
+
+    typedef Grid1IntersectorK<4>  SubdivPatch1Intersector4;
+    typedef Grid1IntersectorK<8>  SubdivPatch1Intersector8;
+    typedef Grid1IntersectorK<16> SubdivPatch1Intersector16;
+
+  }
+}

thirdparty/embree/kernels/geometry/grid_soa.h

@@ -0,0 +1,275 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/scene_subdiv_mesh.h"
+#include "../bvh/bvh.h"
+#include "../subdiv/tessellation.h"
+#include "../subdiv/tessellation_cache.h"
+#include "subdivpatch1.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    class GridSOA
+    {
+    public:
+
+      /*! GridSOA constructor */
+      GridSOA(const SubdivPatch1Base* patches, const unsigned time_steps,
+              const unsigned x0, const unsigned x1, const unsigned y0, const unsigned y1, const unsigned swidth, const unsigned sheight,
+              const SubdivMesh* const geom, const size_t totalBvhBytes, const size_t gridBytes, BBox3fa* bounds_o = nullptr);
+
+      /*! Subgrid creation */
+      template<typename Allocator>
+        static GridSOA* create(const SubdivPatch1Base* patches, const unsigned time_steps,
+                               unsigned x0, unsigned x1, unsigned y0, unsigned y1, 
+                               const Scene* scene, Allocator& alloc, BBox3fa* bounds_o = nullptr)
+      {
+        const unsigned width = x1-x0+1;  
+        const unsigned height = y1-y0+1; 
+        const GridRange range(0,width-1,0,height-1);
+        size_t bvhBytes = 0;
+        if (time_steps == 1) 
+          bvhBytes = getBVHBytes(range,sizeof(BVH4::AABBNode),0);
+        else {
+          bvhBytes = (time_steps-1)*getBVHBytes(range,sizeof(BVH4::AABBNodeMB),0);
+          bvhBytes += getTemporalBVHBytes(make_range(0,int(time_steps-1)),sizeof(BVH4::AABBNodeMB4D));
+        }
+        const size_t gridBytes = 4*size_t(width)*size_t(height)*sizeof(float);  
+        size_t rootBytes = time_steps*sizeof(BVH4::NodeRef);
+#if !defined(__64BIT__)
+        rootBytes += 4; // We read 2 elements behind the grid. As we store at least 8 root bytes after the grid we are fine in 64 bit mode. But in 32 bit mode we have to do additional padding.
+#endif
+        void* data = alloc(offsetof(GridSOA,data)+bvhBytes+time_steps*gridBytes+rootBytes);
+        assert(data);
+        return new (data) GridSOA(patches,time_steps,x0,x1,y0,y1,patches->grid_u_res,patches->grid_v_res,scene->get<SubdivMesh>(patches->geomID()),bvhBytes,gridBytes,bounds_o);
+      }
+
+      /*! Grid creation */
+      template<typename Allocator>
+        static GridSOA* create(const SubdivPatch1Base* const patches, const unsigned time_steps,
+                               const Scene* scene, const Allocator& alloc, BBox3fa* bounds_o = nullptr) 
+      {
+        return create(patches,time_steps,0,patches->grid_u_res-1,0,patches->grid_v_res-1,scene,alloc,bounds_o);
+      }
+
+       /*! returns reference to root */
+      __forceinline       BVH4::NodeRef& root(size_t t = 0)       { return (BVH4::NodeRef&)data[rootOffset + t*sizeof(BVH4::NodeRef)]; }
+      __forceinline const BVH4::NodeRef& root(size_t t = 0) const { return (BVH4::NodeRef&)data[rootOffset + t*sizeof(BVH4::NodeRef)]; }
+
+      /*! returns pointer to BVH array */
+      __forceinline       char* bvhData()       { return &data[0]; }
+      __forceinline const char* bvhData() const { return &data[0]; }
+
+      /*! returns pointer to Grid array */
+      __forceinline       float* gridData(size_t t = 0)       { return (float*) &data[gridOffset + t*gridBytes]; }
+      __forceinline const float* gridData(size_t t = 0) const { return (float*) &data[gridOffset + t*gridBytes]; }
+      
+      __forceinline void* encodeLeaf(size_t u, size_t v) {
+        return (void*) (16*(v * width + u + 1)); // +1 to not create empty leaf
+      }
+      __forceinline float* decodeLeaf(size_t t, const void* ptr) {
+        return gridData(t) + (((size_t) (ptr) >> 4) - 1);
+      }
+
+      /*! returns the size of the BVH over the grid in bytes */
+      static size_t getBVHBytes(const GridRange& range, const size_t nodeBytes, const size_t leafBytes);
+
+      /*! returns the size of the temporal BVH over the time range BVHs */
+      static size_t getTemporalBVHBytes(const range<int> time_range, const size_t nodeBytes);
+
+      /*! calculates bounding box of grid range */
+      __forceinline BBox3fa calculateBounds(size_t time, const GridRange& range) const
+      {
+        const float* const grid_array = gridData(time);
+        const float* const grid_x_array = grid_array + 0 * dim_offset;
+        const float* const grid_y_array = grid_array + 1 * dim_offset;
+        const float* const grid_z_array = grid_array + 2 * dim_offset;
+        
+        /* compute the bounds just for the range! */
+        BBox3fa bounds( empty );
+        for (unsigned v = range.v_start; v<=range.v_end; v++) 
+        {
+          for (unsigned u = range.u_start; u<=range.u_end; u++)
+          {
+            const float x = grid_x_array[ v * width + u];
+            const float y = grid_y_array[ v * width + u];
+            const float z = grid_z_array[ v * width + u];
+            bounds.extend( Vec3fa(x,y,z) );
+          }
+        }
+        assert(is_finite(bounds));
+        return bounds;
+      }
+
+      /*! Evaluates grid over patch and builds BVH4 tree over the grid. */
+      std::pair<BVH4::NodeRef,BBox3fa> buildBVH(BBox3fa* bounds_o);
+      
+      /*! Create BVH4 tree over grid. */
+      std::pair<BVH4::NodeRef,BBox3fa> buildBVH(const GridRange& range, size_t& allocator);
+
+      /*! Evaluates grid over patch and builds MSMBlur BVH4 tree over the grid. */
+      std::pair<BVH4::NodeRef,LBBox3fa> buildMSMBlurBVH(const range<int> time_range, BBox3fa* bounds_o);
+      
+      /*! Create MBlur BVH4 tree over grid. */
+      std::pair<BVH4::NodeRef,LBBox3fa> buildMBlurBVH(size_t time, const GridRange& range, size_t& allocator);
+
+      /*! Create MSMBlur BVH4 tree over grid. */
+      std::pair<BVH4::NodeRef,LBBox3fa> buildMSMBlurBVH(const range<int> time_range, size_t& allocator, BBox3fa* bounds_o);
+
+      template<typename Loader>
+        struct MapUV
+      {
+        typedef typename Loader::vfloat vfloat;
+        const float* const grid_uv;
+        size_t line_offset;
+        size_t lines;
+
+        __forceinline MapUV(const float* const grid_uv, size_t line_offset, const size_t lines)
+          : grid_uv(grid_uv), line_offset(line_offset), lines(lines) {}
+
+        __forceinline void operator() (vfloat& u, vfloat& v, Vec3<vfloat>& Ng) const {
+          const Vec3<vfloat> tri_v012_uv = Loader::gather(grid_uv,line_offset,lines);	
+          const Vec2<vfloat> uv0 = GridSOA::decodeUV(tri_v012_uv[0]);
+          const Vec2<vfloat> uv1 = GridSOA::decodeUV(tri_v012_uv[1]);
+          const Vec2<vfloat> uv2 = GridSOA::decodeUV(tri_v012_uv[2]);        
+          const Vec2<vfloat> uv = u * uv1 + v * uv2 + (1.0f-u-v) * uv0;        
+          u = uv[0];v = uv[1]; 
+        }
+      };
+
+      struct Gather2x3
+      {
+        enum { M = 4 };
+        typedef vbool4 vbool;
+        typedef vint4 vint;
+        typedef vfloat4 vfloat;
+        
+        static __forceinline const Vec3vf4 gather(const float* const grid, const size_t line_offset, const size_t lines)
+        {
+          vfloat4 r0 = vfloat4::loadu(grid + 0*line_offset);
+          vfloat4 r1 = vfloat4::loadu(grid + 1*line_offset); // this accesses 2 elements too much in case of 2x2 grid, but this is ok as we ensure enough padding after the grid
+          if (unlikely(line_offset == 2))
+          {
+            r0 = shuffle<0,1,1,1>(r0);
+            r1 = shuffle<0,1,1,1>(r1);
+          }
+          return Vec3vf4(unpacklo(r0,r1),       // r00, r10, r01, r11
+                         shuffle<1,1,2,2>(r0),  // r01, r01, r02, r02
+                         shuffle<0,1,1,2>(r1)); // r10, r11, r11, r12
+        }
+
+        static __forceinline void gather(const float* const grid_x, 
+                                         const float* const grid_y, 
+                                         const float* const grid_z, 
+                                         const size_t line_offset,
+                                         const size_t lines,
+                                         Vec3vf4& v0_o,
+                                         Vec3vf4& v1_o,
+                                         Vec3vf4& v2_o)
+        {
+          const Vec3vf4 tri_v012_x = gather(grid_x,line_offset,lines);
+          const Vec3vf4 tri_v012_y = gather(grid_y,line_offset,lines);
+          const Vec3vf4 tri_v012_z = gather(grid_z,line_offset,lines);
+          v0_o = Vec3vf4(tri_v012_x[0],tri_v012_y[0],tri_v012_z[0]);
+          v1_o = Vec3vf4(tri_v012_x[1],tri_v012_y[1],tri_v012_z[1]);
+          v2_o = Vec3vf4(tri_v012_x[2],tri_v012_y[2],tri_v012_z[2]);
+        }
+      };
+      
+#if defined (__AVX__)
+      struct Gather3x3
+      {
+        enum { M = 8 };
+        typedef vbool8 vbool;
+        typedef vint8 vint;
+        typedef vfloat8 vfloat;
+        
+        static __forceinline const Vec3vf8 gather(const float* const grid, const size_t line_offset, const size_t lines)
+        {
+          vfloat4 ra = vfloat4::loadu(grid + 0*line_offset);
+          vfloat4 rb = vfloat4::loadu(grid + 1*line_offset); // this accesses 2 elements too much in case of 2x2 grid, but this is ok as we ensure enough padding after the grid
+          vfloat4 rc;
+          if (likely(lines > 2)) 
+            rc = vfloat4::loadu(grid + 2*line_offset);
+          else                   
+            rc = rb;
+
+          if (unlikely(line_offset == 2))
+          {
+            ra = shuffle<0,1,1,1>(ra);
+            rb = shuffle<0,1,1,1>(rb);
+            rc = shuffle<0,1,1,1>(rc);
+          }
+          
+          const vfloat8 r0 = vfloat8(ra,rb);
+          const vfloat8 r1 = vfloat8(rb,rc);
+          return Vec3vf8(unpacklo(r0,r1),         // r00, r10, r01, r11, r10, r20, r11, r21
+                         shuffle<1,1,2,2>(r0),    // r01, r01, r02, r02, r11, r11, r12, r12
+                         shuffle<0,1,1,2>(r1));   // r10, r11, r11, r12, r20, r21, r21, r22
+        }
+
+        static __forceinline void gather(const float* const grid_x, 
+                                         const float* const grid_y, 
+                                         const float* const grid_z, 
+                                         const size_t line_offset,
+                                         const size_t lines,
+                                         Vec3vf8& v0_o,
+                                         Vec3vf8& v1_o,
+                                         Vec3vf8& v2_o)
+        {
+          const Vec3vf8 tri_v012_x = gather(grid_x,line_offset,lines);
+          const Vec3vf8 tri_v012_y = gather(grid_y,line_offset,lines);
+          const Vec3vf8 tri_v012_z = gather(grid_z,line_offset,lines);
+          v0_o = Vec3vf8(tri_v012_x[0],tri_v012_y[0],tri_v012_z[0]);
+          v1_o = Vec3vf8(tri_v012_x[1],tri_v012_y[1],tri_v012_z[1]);
+          v2_o = Vec3vf8(tri_v012_x[2],tri_v012_y[2],tri_v012_z[2]);
+        }
+      };
+#endif
+
+      template<typename vfloat>
+      static __forceinline Vec2<vfloat> decodeUV(const vfloat& uv)
+      {
+        typedef typename vfloat::Int vint;
+        const vint iu  = asInt(uv) & 0xffff;
+        const vint iv  = srl(asInt(uv),16);
+	const vfloat u = (vfloat)iu * vfloat(8.0f/0x10000);
+	const vfloat v = (vfloat)iv * vfloat(8.0f/0x10000);
+	return Vec2<vfloat>(u,v);
+      }
+      
+      __forceinline unsigned int geomID() const  {
+        return _geomID;
+      } 
+      
+      __forceinline unsigned int primID() const  {
+        return _primID;
+      } 
+
+    public:
+      BVH4::NodeRef troot;
+#if !defined(__64BIT__)
+      unsigned align1;
+#endif
+      unsigned time_steps;
+      unsigned width;
+
+      unsigned height;
+      unsigned dim_offset;
+      unsigned _geomID;
+      unsigned _primID;
+
+      unsigned align2;
+      unsigned gridOffset;
+      unsigned gridBytes;
+      unsigned rootOffset;
+
+      char data[1];        //!< after the struct we first store the BVH, then the grid, and finally the roots
+    };
+  }
+}

thirdparty/embree/kernels/geometry/grid_soa_intersector1.h

@@ -0,0 +1,207 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "grid_soa.h"
+#include "../common/ray.h"
+#include "triangle_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    class GridSOAIntersector1
+    {
+    public:
+      typedef void Primitive;
+      
+      class Precalculations
+      { 
+      public:
+        __forceinline Precalculations (const Ray& ray, const void* ptr)
+          : grid(nullptr) {}
+        
+      public:
+        GridSOA* grid;
+        int itime;
+        float ftime;
+      };
+      
+      template<typename Loader>
+        static __forceinline void intersect(RayHit& ray,
+                                            RayQueryContext* context, 
+                                            const float* const grid_x,
+                                            const size_t line_offset,
+                                            const size_t lines,
+                                            Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+        Vec3<vfloat> v0, v1, v2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,v0,v1,v2);       
+        GridSOA::MapUV<Loader> mapUV(grid_uv,line_offset,lines);
+        PlueckerIntersector1<Loader::M> intersector(ray,nullptr);
+        intersector.intersect(ray,v0,v1,v2,mapUV,Intersect1EpilogMU<Loader::M,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+      };
+      
+      template<typename Loader>
+        static __forceinline bool occluded(Ray& ray,
+                                           RayQueryContext* context, 
+                                           const float* const grid_x,
+                                           const size_t line_offset,
+                                           const size_t lines,
+                                           Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        Vec3<vfloat> v0, v1, v2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,v0,v1,v2);
+        
+        GridSOA::MapUV<Loader> mapUV(grid_uv,line_offset,lines);
+        PlueckerIntersector1<Loader::M> intersector(ray,nullptr);
+        return intersector.intersect(ray,v0,v1,v2,mapUV,Occluded1EpilogMU<Loader::M,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+      }
+      
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node) 
+      {
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+        
+#if defined(__AVX__)
+        intersect<GridSOA::Gather3x3>( ray, context, grid_x, line_offset, lines, pre);
+#else
+        intersect<GridSOA::Gather2x3>(ray, context, grid_x            , line_offset, lines, pre);
+        if (likely(lines > 2))
+          intersect<GridSOA::Gather2x3>(ray, context, grid_x+line_offset, line_offset, lines, pre);
+#endif
+      }
+      
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline bool occluded(Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+        
+#if defined(__AVX__)
+        return occluded<GridSOA::Gather3x3>( ray, context, grid_x, line_offset, lines, pre);
+#else
+        if (occluded<GridSOA::Gather2x3>(ray, context, grid_x            , line_offset, lines, pre)) return true;
+        if (likely(lines > 2))
+          if (occluded<GridSOA::Gather2x3>(ray, context, grid_x+line_offset, line_offset, lines, pre)) return true;
+#endif
+        return false;
+      }      
+    };
+
+    class GridSOAMBIntersector1
+    {
+    public:
+      typedef void Primitive;
+      typedef GridSOAIntersector1::Precalculations Precalculations;
+      
+      template<typename Loader>
+        static __forceinline void intersect(RayHit& ray, const float ftime,
+                                            RayQueryContext* context, 
+                                            const float* const grid_x,
+                                            const size_t line_offset,
+                                            const size_t lines,
+                                            Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        Vec3<vfloat> a0, a1, a2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,a0,a1,a2);
+
+        Vec3<vfloat> b0, b1, b2;
+        Loader::gather(grid_x+grid_offset,grid_y+grid_offset,grid_z+grid_offset,line_offset,lines,b0,b1,b2);
+
+        Vec3<vfloat> v0 = lerp(a0,b0,vfloat(ftime));
+        Vec3<vfloat> v1 = lerp(a1,b1,vfloat(ftime));
+        Vec3<vfloat> v2 = lerp(a2,b2,vfloat(ftime));
+
+        GridSOA::MapUV<Loader> mapUV(grid_uv,line_offset,lines);
+        PlueckerIntersector1<Loader::M> intersector(ray,nullptr);
+        intersector.intersect(ray,v0,v1,v2,mapUV,Intersect1EpilogMU<Loader::M,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+      };
+      
+      template<typename Loader>
+        static __forceinline bool occluded(Ray& ray, const float ftime,
+                                           RayQueryContext* context, 
+                                           const float* const grid_x,
+                                           const size_t line_offset,
+                                           const size_t lines,
+                                           Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        Vec3<vfloat> a0, a1, a2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,a0,a1,a2);
+
+        Vec3<vfloat> b0, b1, b2;
+        Loader::gather(grid_x+grid_offset,grid_y+grid_offset,grid_z+grid_offset,line_offset,lines,b0,b1,b2);
+       
+        Vec3<vfloat> v0 = lerp(a0,b0,vfloat(ftime));
+        Vec3<vfloat> v1 = lerp(a1,b1,vfloat(ftime));
+        Vec3<vfloat> v2 = lerp(a2,b2,vfloat(ftime));
+        
+        GridSOA::MapUV<Loader> mapUV(grid_uv,line_offset,lines);
+        PlueckerIntersector1<Loader::M> intersector(ray,nullptr);
+        return intersector.intersect(ray,v0,v1,v2,mapUV,Occluded1EpilogMU<Loader::M,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+      }
+      
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node) 
+      { 
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(pre.itime,prim);
+        
+#if defined(__AVX__)
+        intersect<GridSOA::Gather3x3>( ray, pre.ftime, context, grid_x, line_offset, lines, pre);
+#else
+        intersect<GridSOA::Gather2x3>(ray, pre.ftime, context, grid_x, line_offset, lines, pre);
+        if (likely(lines > 2))
+          intersect<GridSOA::Gather2x3>(ray, pre.ftime, context, grid_x+line_offset, line_offset, lines, pre);
+#endif
+      }
+      
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline bool occluded(Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(pre.itime,prim);
+        
+#if defined(__AVX__)
+        return occluded<GridSOA::Gather3x3>( ray, pre.ftime, context, grid_x, line_offset, lines, pre);
+#else
+        if (occluded<GridSOA::Gather2x3>(ray, pre.ftime, context, grid_x            , line_offset, lines, pre)) return true;
+        if (likely(lines > 2))
+          if (occluded<GridSOA::Gather2x3>(ray, pre.ftime, context, grid_x+line_offset, line_offset, lines, pre)) return true;
+#endif
+        return false;
+      }      
+    };
+  }
+}

thirdparty/embree/kernels/geometry/grid_soa_intersector_packet.h

@@ -0,0 +1,445 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "grid_soa.h"
+#include "../common/ray.h"
+#include "triangle_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int K>
+      struct MapUV0
+    {
+      const float* const grid_uv;
+      size_t ofs00, ofs01, ofs10, ofs11;
+      
+      __forceinline MapUV0(const float* const grid_uv, size_t ofs00, size_t ofs01, size_t ofs10, size_t ofs11)
+        : grid_uv(grid_uv), ofs00(ofs00), ofs01(ofs01), ofs10(ofs10), ofs11(ofs11) {}
+      
+      __forceinline void operator() (vfloat<K>& u, vfloat<K>& v, Vec3vf<K>& Ng) const {
+        const vfloat<K> uv00(grid_uv[ofs00]);
+        const vfloat<K> uv01(grid_uv[ofs01]);
+        const vfloat<K> uv10(grid_uv[ofs10]);
+        const vfloat<K> uv11(grid_uv[ofs11]);
+        const Vec2vf<K> uv0 = GridSOA::decodeUV(uv00);
+        const Vec2vf<K> uv1 = GridSOA::decodeUV(uv01);
+        const Vec2vf<K> uv2 = GridSOA::decodeUV(uv10);
+        const Vec2vf<K> uv = madd(u,uv1,madd(v,uv2,(1.0f-u-v)*uv0));
+        u = uv[0]; v = uv[1];
+      }
+    };
+    
+    template<int K>
+      struct MapUV1
+    {
+      const float* const grid_uv;
+      size_t ofs00, ofs01, ofs10, ofs11;
+      
+      __forceinline MapUV1(const float* const grid_uv, size_t ofs00, size_t ofs01, size_t ofs10, size_t ofs11)
+        : grid_uv(grid_uv), ofs00(ofs00), ofs01(ofs01), ofs10(ofs10), ofs11(ofs11) {}
+      
+      __forceinline void operator() (vfloat<K>& u, vfloat<K>& v, Vec3vf<K>& Ng) const {
+        const vfloat<K> uv00(grid_uv[ofs00]);
+        const vfloat<K> uv01(grid_uv[ofs01]);
+        const vfloat<K> uv10(grid_uv[ofs10]);
+        const vfloat<K> uv11(grid_uv[ofs11]);
+        const Vec2vf<K> uv0 = GridSOA::decodeUV(uv10);
+        const Vec2vf<K> uv1 = GridSOA::decodeUV(uv01);
+        const Vec2vf<K> uv2 = GridSOA::decodeUV(uv11);
+        const Vec2vf<K> uv = madd(u,uv1,madd(v,uv2,(1.0f-u-v)*uv0));
+        u = uv[0]; v = uv[1];
+      }
+    };
+    
+    template<int K>
+      class GridSOAIntersectorK
+    {
+    public:
+      typedef void Primitive;
+
+      class Precalculations
+      {
+#if defined(__AVX__)
+        static const int M = 8;
+#else
+        static const int M = 4;
+#endif
+
+      public:
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray)
+          : grid(nullptr), intersector(valid,ray) {}
+
+      public:
+        GridSOA* grid;
+        PlueckerIntersectorK<M,K> intersector; // FIXME: use quad intersector
+      };
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t line_offset   = pre.grid->width;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        const size_t max_x = pre.grid->width  == 2 ? 1 : 2;
+        const size_t max_y = pre.grid->height == 2 ? 1 : 2;
+        for (size_t y=0; y<max_y; y++)        
+        {
+          for (size_t x=0; x<max_x; x++)
+          {
+            const size_t ofs00 = (y+0)*line_offset+(x+0);
+            const size_t ofs01 = (y+0)*line_offset+(x+1);
+            const size_t ofs10 = (y+1)*line_offset+(x+0);
+            const size_t ofs11 = (y+1)*line_offset+(x+1);
+            const Vec3vf<K> p00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> p01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> p10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> p11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+
+            pre.intersector.intersectK(valid_i,ray,p00,p01,p10,MapUV0<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),IntersectKEpilogMU<1,K,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+            pre.intersector.intersectK(valid_i,ray,p10,p01,p11,MapUV1<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),IntersectKEpilogMU<1,K,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+          }
+        }
+      }
+
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t line_offset   = pre.grid->width;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        vbool<K> valid = valid_i;
+        const size_t max_x = pre.grid->width  == 2 ? 1 : 2;
+        const size_t max_y = pre.grid->height == 2 ? 1 : 2;
+        for (size_t y=0; y<max_y; y++)        
+        {
+          for (size_t x=0; x<max_x; x++)
+          {
+            const size_t ofs00 = (y+0)*line_offset+(x+0);
+            const size_t ofs01 = (y+0)*line_offset+(x+1);
+            const size_t ofs10 = (y+1)*line_offset+(x+0);
+            const size_t ofs11 = (y+1)*line_offset+(x+1);
+            const Vec3vf<K> p00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> p01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> p10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> p11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+
+            pre.intersector.intersectK(valid,ray,p00,p01,p10,MapUV0<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),OccludedKEpilogMU<1,K,true>(valid,ray,context,pre.grid->geomID(),pre.grid->primID()));
+            if (none(valid)) break;
+            pre.intersector.intersectK(valid,ray,p10,p01,p11,MapUV1<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),OccludedKEpilogMU<1,K,true>(valid,ray,context,pre.grid->geomID(),pre.grid->primID()));
+            if (none(valid)) break;
+          }
+        }
+        return !valid;
+      }
+
+      template<typename Loader>
+        static __forceinline void intersect(RayHitK<K>& ray, size_t k,
+                                            RayQueryContext* context,
+                                            const float* const grid_x,
+                                            const size_t line_offset,
+                                            const size_t lines,
+                                            Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+        Vec3<vfloat> v0, v1, v2; Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,v0,v1,v2);
+        pre.intersector.intersect(ray,k,v0,v1,v2,GridSOA::MapUV<Loader>(grid_uv,line_offset,lines),Intersect1KEpilogMU<Loader::M,K,true>(ray,k,context,pre.grid->geomID(),pre.grid->primID()));
+      };
+
+      template<typename Loader>
+        static __forceinline bool occluded(RayK<K>& ray, size_t k,
+                                           RayQueryContext* context,
+                                           const float* const grid_x,
+                                           const size_t line_offset,
+                                           const size_t lines,
+                                           Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+        Vec3<vfloat> v0, v1, v2; Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,v0,v1,v2);
+        return pre.intersector.intersect(ray,k,v0,v1,v2,GridSOA::MapUV<Loader>(grid_uv,line_offset,lines),Occluded1KEpilogMU<Loader::M,K,true>(ray,k,context,pre.grid->geomID(),pre.grid->primID()));
+      }
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+#if defined(__AVX__)
+        intersect<GridSOA::Gather3x3>( ray, k, context, grid_x, line_offset, lines, pre);
+#else
+        intersect<GridSOA::Gather2x3>(ray, k, context, grid_x            , line_offset, lines, pre);
+        if (likely(lines > 2))
+          intersect<GridSOA::Gather2x3>(ray, k, context, grid_x+line_offset, line_offset, lines, pre);
+#endif
+      }
+
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(0,prim);
+
+#if defined(__AVX__)
+        return occluded<GridSOA::Gather3x3>( ray, k, context, grid_x, line_offset, lines, pre);
+#else
+        if (occluded<GridSOA::Gather2x3>(ray, k, context, grid_x            , line_offset, lines, pre)) return true;
+        if (likely(lines > 2))
+          if (occluded<GridSOA::Gather2x3>(ray, k, context, grid_x+line_offset, line_offset, lines, pre)) return true;
+#endif
+        return false;
+      }
+    };
+
+    template<int K>
+    class GridSOAMBIntersectorK
+    {
+    public:
+      typedef void Primitive;
+      typedef typename GridSOAIntersectorK<K>::Precalculations Precalculations;
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        vfloat<K> vftime;
+        vint<K> vitime = getTimeSegment<K>(ray.time(), vfloat<K>((float)(pre.grid->time_steps-1)), vftime);
+
+        vbool<K> valid1 = valid_i;
+        while (any(valid1)) {
+          const size_t j = bsf(movemask(valid1));
+          const int itime = vitime[j];
+          const vbool<K> valid2 = valid1 & (itime == vitime);
+          valid1 = valid1 & !valid2;
+          intersect(valid2,pre,ray,vftime,itime,context,prim,lazy_node);
+        }
+      }
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, const vfloat<K>& ftime, int itime, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t line_offset   = pre.grid->width;
+        const float* const grid_x  = pre.grid->decodeLeaf(itime,prim);
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        const size_t max_x = pre.grid->width  == 2 ? 1 : 2;
+        const size_t max_y = pre.grid->height == 2 ? 1 : 2;
+        for (size_t y=0; y<max_y; y++)        
+        {
+          for (size_t x=0; x<max_x; x++)
+          {
+            size_t ofs00 = (y+0)*line_offset+(x+0);
+            size_t ofs01 = (y+0)*line_offset+(x+1);
+            size_t ofs10 = (y+1)*line_offset+(x+0);
+            size_t ofs11 = (y+1)*line_offset+(x+1);
+            const Vec3vf<K> a00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> a01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> a10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> a11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+            ofs00 += grid_offset;
+            ofs01 += grid_offset;
+            ofs10 += grid_offset;
+            ofs11 += grid_offset;
+            const Vec3vf<K> b00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> b01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> b10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> b11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+            const Vec3vf<K> p00 = lerp(a00,b00,ftime);
+            const Vec3vf<K> p01 = lerp(a01,b01,ftime);
+            const Vec3vf<K> p10 = lerp(a10,b10,ftime);
+            const Vec3vf<K> p11 = lerp(a11,b11,ftime);
+
+            pre.intersector.intersectK(valid_i,ray,p00,p01,p10,MapUV0<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),IntersectKEpilogMU<1,K,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+            pre.intersector.intersectK(valid_i,ray,p10,p01,p11,MapUV1<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),IntersectKEpilogMU<1,K,true>(ray,context,pre.grid->geomID(),pre.grid->primID()));
+          }
+        }
+      }
+
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        vfloat<K> vftime;
+        vint<K> vitime = getTimeSegment<K>(ray.time(), vfloat<K>((float)(pre.grid->time_steps-1)), vftime);
+
+        vbool<K> valid_o = valid_i;
+        vbool<K> valid1 = valid_i;
+        while (any(valid1)) {
+          const int j = int(bsf(movemask(valid1)));
+          const int itime = vitime[j];
+          const vbool<K> valid2 = valid1 & (itime == vitime);
+          valid1 = valid1 & !valid2;
+          valid_o &= !valid2 | occluded(valid2,pre,ray,vftime,itime,context,prim,lazy_node);
+        }
+        return !valid_o;
+      }
+
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, const vfloat<K>& ftime, int itime, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const size_t line_offset   = pre.grid->width;
+        const float* const grid_x  = pre.grid->decodeLeaf(itime,prim);
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        vbool<K> valid = valid_i;
+        const size_t max_x = pre.grid->width  == 2 ? 1 : 2;
+        const size_t max_y = pre.grid->height == 2 ? 1 : 2;
+        for (size_t y=0; y<max_y; y++)        
+        {
+          for (size_t x=0; x<max_x; x++)
+          {
+            size_t ofs00 = (y+0)*line_offset+(x+0);
+            size_t ofs01 = (y+0)*line_offset+(x+1);
+            size_t ofs10 = (y+1)*line_offset+(x+0);
+            size_t ofs11 = (y+1)*line_offset+(x+1);
+            const Vec3vf<K> a00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> a01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> a10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> a11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+            ofs00 += grid_offset;
+            ofs01 += grid_offset;
+            ofs10 += grid_offset;
+            ofs11 += grid_offset;
+            const Vec3vf<K> b00(grid_x[ofs00],grid_y[ofs00],grid_z[ofs00]);
+            const Vec3vf<K> b01(grid_x[ofs01],grid_y[ofs01],grid_z[ofs01]);
+            const Vec3vf<K> b10(grid_x[ofs10],grid_y[ofs10],grid_z[ofs10]);
+            const Vec3vf<K> b11(grid_x[ofs11],grid_y[ofs11],grid_z[ofs11]);
+            const Vec3vf<K> p00 = lerp(a00,b00,ftime);
+            const Vec3vf<K> p01 = lerp(a01,b01,ftime);
+            const Vec3vf<K> p10 = lerp(a10,b10,ftime);
+            const Vec3vf<K> p11 = lerp(a11,b11,ftime);
+
+            pre.intersector.intersectK(valid,ray,p00,p01,p10,MapUV0<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),OccludedKEpilogMU<1,K,true>(valid,ray,context,pre.grid->geomID(),pre.grid->primID()));
+            if (none(valid)) break;
+            pre.intersector.intersectK(valid,ray,p10,p01,p11,MapUV1<K>(grid_uv,ofs00,ofs01,ofs10,ofs11),OccludedKEpilogMU<1,K,true>(valid,ray,context,pre.grid->geomID(),pre.grid->primID()));
+            if (none(valid)) break;
+          }
+        }
+        return valid;
+      }
+
+      template<typename Loader>
+        static __forceinline void intersect(RayHitK<K>& ray, size_t k,
+                                            const float ftime,
+                                            RayQueryContext* context,
+                                            const float* const grid_x,
+                                            const size_t line_offset,
+                                            const size_t lines,
+                                            Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        Vec3<vfloat> a0, a1, a2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,a0,a1,a2);
+
+        Vec3<vfloat> b0, b1, b2;
+        Loader::gather(grid_x+grid_offset,grid_y+grid_offset,grid_z+grid_offset,line_offset,lines,b0,b1,b2);
+
+        Vec3<vfloat> v0 = lerp(a0,b0,vfloat(ftime));
+        Vec3<vfloat> v1 = lerp(a1,b1,vfloat(ftime));
+        Vec3<vfloat> v2 = lerp(a2,b2,vfloat(ftime));
+
+        pre.intersector.intersect(ray,k,v0,v1,v2,GridSOA::MapUV<Loader>(grid_uv,line_offset,lines),Intersect1KEpilogMU<Loader::M,K,true>(ray,k,context,pre.grid->geomID(),pre.grid->primID()));
+      };
+
+      template<typename Loader>
+        static __forceinline bool occluded(RayK<K>& ray, size_t k,
+                                           const float ftime,
+                                           RayQueryContext* context,
+                                           const float* const grid_x,
+                                           const size_t line_offset,
+                                           const size_t lines,
+                                           Precalculations& pre)
+      {
+        typedef typename Loader::vfloat vfloat;
+        const size_t grid_offset   = pre.grid->gridBytes >> 2;
+        const size_t dim_offset    = pre.grid->dim_offset;
+        const float* const grid_y  = grid_x + 1 * dim_offset;
+        const float* const grid_z  = grid_x + 2 * dim_offset;
+        const float* const grid_uv = grid_x + 3 * dim_offset;
+
+        Vec3<vfloat> a0, a1, a2;
+        Loader::gather(grid_x,grid_y,grid_z,line_offset,lines,a0,a1,a2);
+
+        Vec3<vfloat> b0, b1, b2;
+        Loader::gather(grid_x+grid_offset,grid_y+grid_offset,grid_z+grid_offset,line_offset,lines,b0,b1,b2);
+
+        Vec3<vfloat> v0 = lerp(a0,b0,vfloat(ftime));
+        Vec3<vfloat> v1 = lerp(a1,b1,vfloat(ftime));
+        Vec3<vfloat> v2 = lerp(a2,b2,vfloat(ftime));
+
+        return pre.intersector.intersect(ray,k,v0,v1,v2,GridSOA::MapUV<Loader>(grid_uv,line_offset,lines),Occluded1KEpilogMU<Loader::M,K,true>(ray,k,context,pre.grid->geomID(),pre.grid->primID()));
+      }
+
+      /*! Intersect a ray with the primitive. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      { 
+        float ftime;
+        int itime = getTimeSegment(ray.time()[k], float(pre.grid->time_steps-1), ftime);
+
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(itime,prim);
+
+#if defined(__AVX__)
+        intersect<GridSOA::Gather3x3>( ray, k, ftime, context, grid_x, line_offset, lines, pre);
+#else
+        intersect<GridSOA::Gather2x3>(ray, k, ftime, context, grid_x, line_offset, lines, pre);
+        if (likely(lines > 2))
+          intersect<GridSOA::Gather2x3>(ray, k, ftime, context, grid_x+line_offset, line_offset, lines, pre);
+#endif
+      }
+
+      /*! Test if the ray is occluded by the primitive */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        float ftime;
+        int itime = getTimeSegment(ray.time()[k], float(pre.grid->time_steps-1), ftime);
+
+        const size_t line_offset   = pre.grid->width;
+        const size_t lines         = pre.grid->height;
+        const float* const grid_x  = pre.grid->decodeLeaf(itime,prim);
+
+#if defined(__AVX__)
+        return occluded<GridSOA::Gather3x3>( ray, k, ftime, context, grid_x, line_offset, lines, pre);
+#else
+        if (occluded<GridSOA::Gather2x3>(ray, k, ftime, context, grid_x, line_offset, lines, pre)) return true;
+        if (likely(lines > 2))
+          if (occluded<GridSOA::Gather2x3>(ray, k, ftime, context, grid_x+line_offset, line_offset, lines, pre)) return true;
+#endif
+        return false;
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/instance.h

@@ -0,0 +1,78 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../common/scene_instance.h"
+
+namespace embree
+{
+  struct InstancePrimitive
+  {
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return 1; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return N; }
+
+  public:
+
+    InstancePrimitive (const Instance* instance, unsigned int instID) 
+    : instance(instance) 
+    , instID_(instID)
+    {}
+
+    __forceinline void fill(const PrimRef* prims, size_t& i, size_t end, Scene* scene)
+    {
+      assert(end-i == 1);
+      const PrimRef& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const Instance* instance = scene->get<Instance>(geomID);
+      new (this) InstancePrimitive(instance, geomID);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& i, size_t end, Scene* scene, size_t itime)
+    {
+      assert(end-i == 1);
+      const PrimRef& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const Instance* instance = scene->get<Instance>(geomID);
+      new (this) InstancePrimitive(instance,geomID);
+      return instance->linearBounds(0,itime);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& i, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      assert(end-i == 1);
+      const PrimRefMB& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const Instance* instance = scene->get<Instance>(geomID);
+      new (this) InstancePrimitive(instance,geomID);
+      return instance->linearBounds(0,time_range);
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(Instance* instance) {
+      return instance->bounds(0);
+    }
+
+  public:
+    const Instance* instance;
+    const unsigned int instID_ = std::numeric_limits<unsigned int>::max ();
+  };
+}

thirdparty/embree/kernels/geometry/instance_array.h

@@ -0,0 +1,85 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../common/scene_instance_array.h"
+
+namespace embree
+{
+  //template<int M>
+  struct InstanceArrayPrimitive
+  {
+    struct Type : public PrimitiveType
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return 1; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return N; }
+
+  public:
+
+    InstanceArrayPrimitive (const uint32_t geomID, const uint32_t primID)
+    : primID_(primID)
+    , instID_(geomID)
+    {}
+
+    __forceinline bool valid() const { 
+      return primID_ != -1;
+    }
+
+    void fill(const PrimRef* prims, size_t& i, size_t end, Scene* scene)
+    {
+      assert(end-i == 1);
+      const PrimRef& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const unsigned int primID = prim.primID();
+      new (this) InstanceArrayPrimitive(geomID, primID);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& i, size_t end, Scene* scene, size_t itime)
+    {
+      assert(end-i == 1);
+      const PrimRef& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const size_t primID = prim.primID();
+      new (this) InstanceArrayPrimitive(geomID, primID);
+      const InstanceArray* instanceArray = scene->get<InstanceArray>(geomID);
+      return instanceArray->linearBounds(primID,itime);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& i, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      assert(end-i == 1);
+      const PrimRefMB& prim = prims[i]; i++;
+      const unsigned int geomID = prim.geomID();
+      const size_t primID = prim.primID();
+      new (this) InstanceArrayPrimitive(geomID, primID);
+      const InstanceArray* instanceArray = scene->get<InstanceArray>(geomID);
+      return instanceArray->linearBounds(primID,time_range);
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(InstanceArray* instanceArray) {
+      return instanceArray->bounds(0);
+    }
+
+  public:
+    unsigned int primID_;
+    unsigned int instID_;
+  };
+}

thirdparty/embree/kernels/geometry/instance_array_intersector.h

@@ -0,0 +1,85 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "instance_array.h"
+#include "../common/ray.h"
+#include "../common/point_query.h"
+#include "../common/scene.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct InstanceArrayIntersector1
+    {
+      typedef InstanceArrayPrimitive Primitive;
+
+      struct Precalculations {
+        __forceinline Precalculations (const Ray& ray, const void *ptr) {}
+      };
+
+      static void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim);
+      static bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim);
+      static bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim);
+    };
+
+    struct InstanceArrayIntersector1MB
+    {
+      typedef InstanceArrayPrimitive Primitive;
+
+      struct Precalculations {
+        __forceinline Precalculations (const Ray& ray, const void *ptr) {}
+      };
+      
+      static void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim);
+      static bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim);
+      static bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim);
+    };
+
+    template<int K>
+      struct InstanceArrayIntersectorK
+    {
+      typedef InstanceArrayPrimitive Primitive;
+      
+      struct Precalculations {
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray) {}
+      };
+      
+      static void intersect(const vbool<K>& valid_i, const Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& prim);
+      static vbool<K> occluded(const vbool<K>& valid_i, const Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& prim);
+
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        intersect(vbool<K>(1<<int(k)),pre,ray,context,prim);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        occluded(vbool<K>(1<<int(k)),pre,ray,context,prim);
+        return ray.tfar[k] < 0.0f; 
+      }
+    };
+
+    template<int K>
+      struct InstanceArrayIntersectorKMB
+    {
+      typedef InstanceArrayPrimitive Primitive;
+      
+      struct Precalculations {
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray) {}
+      };
+      
+      static void intersect(const vbool<K>& valid_i, const Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& prim);
+      static vbool<K> occluded(const vbool<K>& valid_i, const Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& prim);
+
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        intersect(vbool<K>(1<<int(k)),pre,ray,context,prim);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        occluded(vbool<K>(1<<int(k)),pre,ray,context,prim);
+        return ray.tfar[k] < 0.0f; 
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/instance_intersector.h

@@ -0,0 +1,84 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "instance.h"
+#include "../common/ray.h"
+#include "../common/point_query.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct InstanceIntersector1
+    {
+      typedef InstancePrimitive Primitive;
+
+      struct Precalculations {
+        __forceinline Precalculations (const Ray& ray, const void *ptr) {}
+      };
+      
+      static void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim);
+      static bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim);
+      static bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim);
+    };
+
+    struct InstanceIntersector1MB
+    {
+      typedef InstancePrimitive Primitive;
+
+      struct Precalculations {
+        __forceinline Precalculations (const Ray& ray, const void *ptr) {}
+      };
+      
+      static void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim);
+      static bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim);
+      static bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim);
+    };
+
+    template<int K>
+      struct InstanceIntersectorK
+    {
+      typedef InstancePrimitive Primitive;
+      
+      struct Precalculations {
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray) {}
+      };
+      
+      static void intersect(const vbool<K>& valid_i, const Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& prim);
+      static vbool<K> occluded(const vbool<K>& valid_i, const Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& prim);
+
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        intersect(vbool<K>(1<<int(k)),pre,ray,context,prim);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        occluded(vbool<K>(1<<int(k)),pre,ray,context,prim);
+        return ray.tfar[k] < 0.0f; 
+      }
+    };
+
+    template<int K>
+      struct InstanceIntersectorKMB
+    {
+      typedef InstancePrimitive Primitive;
+      
+      struct Precalculations {
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray) {}
+      };
+      
+      static void intersect(const vbool<K>& valid_i, const Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& prim);
+      static vbool<K> occluded(const vbool<K>& valid_i, const Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& prim);
+
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        intersect(vbool<K>(1<<int(k)),pre,ray,context,prim);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        occluded(vbool<K>(1<<int(k)),pre,ray,context,prim);
+        return ray.tfar[k] < 0.0f; 
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/intersector_epilog_sycl.h

@@ -0,0 +1,207 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/context.h"
+#include "filter_sycl.h"
+
+namespace embree
+{
+  template<typename Ray>
+  struct Intersect1Epilog1_HWIF;
+
+  template<>
+  struct Intersect1Epilog1_HWIF<RayHit>
+  {
+    RayHit& ray;
+    sycl::private_ptr<RayQueryContext> context;
+    const unsigned int geomID;
+    const unsigned int primID;
+    const bool filter;
+    
+    __forceinline Intersect1Epilog1_HWIF(RayHit& ray,
+                                         sycl::private_ptr<RayQueryContext> context,
+                                         const unsigned int geomID,
+                                         const unsigned int primID,
+                                         const bool filter)
+      : ray(ray), context(context), geomID(geomID), primID(primID), filter(filter) {}
+    
+    template<typename Hit_i>
+    __forceinline bool operator() (Hit_i& hit_i) const
+    {
+      hit_i.finalize();
+
+      Scene* scene MAYBE_UNUSED = context->scene;
+      Geometry* geometry MAYBE_UNUSED = scene->get(geomID);
+
+      /* ray mask test */
+#if defined(EMBREE_RAY_MASK)
+      if ((geometry->mask & ray.mask) == 0)
+        return false;
+#endif
+
+      /* call intersection filter function */
+#if defined(EMBREE_FILTER_FUNCTION) 
+      if (filter && (unlikely(context->hasContextFilter() || geometry->hasIntersectionFilter())))
+      {
+        Hit h(context->user,geomID,primID,Vec2f(hit_i.u,hit_i.v),hit_i.Ng);
+        float old_t = ray.tfar;
+        ray.tfar = hit_i.t;
+        bool found = runIntersectionFilter1SYCL(geometry,ray,context,h);
+        if (!found) {
+          ray.tfar = old_t;
+          return false;
+        }
+      }
+#endif
+      
+      ray.tfar = hit_i.t;
+      ray.u = hit_i.u;
+      ray.v = hit_i.v;
+      ray.Ng.x = hit_i.Ng.x;
+      ray.Ng.y = hit_i.Ng.y;
+      ray.Ng.z = hit_i.Ng.z;
+      ray.geomID = geomID;
+      ray.primID = primID;
+      instance_id_stack::copy_UU(context->user, context->user->instID, ray.instID);
+#if defined(EMBREE_GEOMETRY_INSTANCE_ARRAY)
+      instance_id_stack::copy_UU(context->user, context->user->instPrimID, ray.instPrimID);
+#endif
+      return true;
+    }
+    
+    template<typename Hit_i>
+    __forceinline bool operator() (bool, Hit_i& hit_i) const
+    {
+      hit_i.finalize();
+
+      Scene* scene MAYBE_UNUSED = context->scene;
+      Geometry* geometry MAYBE_UNUSED = scene->get(geomID);
+
+      /* ray mask test */
+#if defined(EMBREE_RAY_MASK)
+      if ((geometry->mask & ray.mask) == 0)
+        return false;
+#endif
+
+      const Vec3fa Ng = hit_i.Ng();
+      const Vec2f uv = hit_i.uv();
+      
+      /* call intersection filter function */
+#if defined(EMBREE_FILTER_FUNCTION) 
+      if (filter && (unlikely(context->hasContextFilter() || geometry->hasIntersectionFilter())))
+      {
+        Hit h(context->user,geomID,primID,uv,Ng);
+        float old_t = ray.tfar;
+        ray.tfar = hit_i.t();
+        bool found = runIntersectionFilter1SYCL(geometry,ray,context,h);
+        if (!found) {
+          ray.tfar = old_t;
+          return false;
+        }
+      }
+#endif
+      
+      ray.tfar = hit_i.t();
+      ray.u = uv.x;
+      ray.v = uv.y;
+      ray.Ng.x = Ng.x;
+      ray.Ng.y = Ng.y;
+      ray.Ng.z = Ng.z;
+      ray.geomID = geomID;
+      ray.primID = primID;
+      instance_id_stack::copy_UU(context->user, context->user->instID, ray.instID);
+#if defined(EMBREE_GEOMETRY_INSTANCE_ARRAY)
+      instance_id_stack::copy_UU(context->user, context->user->instPrimID, ray.instPrimID);
+#endif
+      return true;
+    }
+  };
+
+  template<>
+  struct Intersect1Epilog1_HWIF<Ray>
+  {
+    Ray& ray;
+    sycl::private_ptr<RayQueryContext> context;
+    const unsigned int geomID;
+    const unsigned int primID;
+    const bool filter;
+    
+    __forceinline Intersect1Epilog1_HWIF(Ray& ray,
+                                         sycl::private_ptr<RayQueryContext> context,
+                                         const unsigned int geomID,
+                                         const unsigned int primID,
+                                         const bool filter)
+      : ray(ray), context(context), geomID(geomID), primID(primID), filter(filter) {}
+    
+    template<typename Hit_i>
+    __forceinline bool operator() (Hit_i& hit_i) const
+    {
+      hit_i.finalize();
+
+      Scene* scene MAYBE_UNUSED = context->scene;
+      Geometry* geometry MAYBE_UNUSED = scene->get(geomID);
+
+      /* ray mask test */
+#if defined(EMBREE_RAY_MASK)
+      if ((geometry->mask & ray.mask) == 0)
+        return false;
+#endif
+
+      /* call intersection filter function */
+#if defined(EMBREE_FILTER_FUNCTION) 
+      if (filter && (unlikely(context->hasContextFilter() || geometry->hasOcclusionFilter())))
+      {
+        Hit h(context->user,geomID,primID,Vec2f(hit_i.u,hit_i.v),hit_i.Ng);
+        float old_t = ray.tfar;
+        ray.tfar = hit_i.t;
+        bool found = runIntersectionFilter1SYCL(geometry,ray,context,h);
+        if (!found) {
+          ray.tfar = old_t;
+          return false;
+        }
+      }
+#endif
+      
+      ray.tfar = neg_inf;
+      return true;
+    }
+    
+    template<typename Hit_i>
+    __forceinline bool operator() (bool, Hit_i& hit_i) const
+    {
+      hit_i.finalize();
+
+      Scene* scene MAYBE_UNUSED = context->scene;
+      Geometry* geometry MAYBE_UNUSED = scene->get(geomID);
+
+      /* ray mask test */
+#if defined(EMBREE_RAY_MASK)
+      if ((geometry->mask & ray.mask) == 0)
+        return false;
+#endif
+
+      /* call intersection filter function */
+#if defined(EMBREE_FILTER_FUNCTION) 
+      if (filter && (unlikely(context->hasContextFilter() || geometry->hasOcclusionFilter())))
+      {
+        const Vec3fa Ng = hit_i.Ng();
+        const Vec2f uv = hit_i.uv();
+        Hit h(context->user,geomID,primID,uv,Ng);
+        float old_t = ray.tfar;
+        ray.tfar = hit_i.t();
+        bool found = runIntersectionFilter1SYCL(geometry,ray,context,h);
+        if (!found) {
+          ray.tfar = old_t;
+          return false;
+        }
+      }
+#endif
+      
+      ray.tfar = neg_inf;
+      return true;
+    }
+  };
+}

thirdparty/embree/kernels/geometry/intersector_iterators.h

@@ -0,0 +1,173 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/scene.h"
+#include "../common/ray.h"
+#include "../common/point_query.h"
+#include "../bvh/node_intersector1.h"
+#include "../bvh/node_intersector_packet.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename Intersector>
+    struct ArrayIntersector1
+    {
+      typedef typename Intersector::Primitive Primitive;
+      typedef typename Intersector::Precalculations Precalculations;
+
+      template<int N, bool robust>
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        for (size_t i=0; i<num; i++) {
+          Intersector::intersect(pre,ray,context,prim[i]);
+        }
+      }
+
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        for (size_t i=0; i<num; i++) {
+          if (Intersector::occluded(pre,ray,context,prim[i]))
+            return true;
+        }
+        return false;
+      }
+
+      template<int N>
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t num, const TravPointQuery<N> &tquery, size_t& lazy_node)
+      {
+        bool changed = false;
+        for (size_t i=0; i<num; i++)
+          changed |= Intersector::pointQuery(query, context, prim[i]);
+        return changed;
+      }
+
+      template<int K>
+      static __forceinline void intersectK(const vbool<K>& valid, /* PrecalculationsK& pre, */ RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+      {
+      }
+
+      template<int K>
+      static __forceinline vbool<K> occludedK(const vbool<K>& valid, /* PrecalculationsK& pre, */ RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+      {
+        return valid;
+      }
+    };
+
+    template<int K, typename Intersector>
+    struct ArrayIntersectorK_1
+    {
+      typedef typename Intersector::Primitive Primitive;
+      typedef typename Intersector::Precalculations Precalculations;
+
+      template<bool robust>
+      static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        for (size_t i=0; i<num; i++) {
+          Intersector::intersect(valid,pre,ray,context,prim[i]);
+        }
+      }
+
+      template<bool robust>
+      static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        vbool<K> valid0 = valid;
+        for (size_t i=0; i<num; i++) {
+          valid0 &= !Intersector::occluded(valid0,pre,ray,context,prim[i]);
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+
+      template<int N, bool robust>
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        for (size_t i=0; i<num; i++) {
+          Intersector::intersect(pre,ray,k,context,prim[i]);
+        }
+      }
+
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        for (size_t i=0; i<num; i++) {
+          if (Intersector::occluded(pre,ray,k,context,prim[i]))
+            return true;
+        }
+        return false;
+      }
+    };
+
+    // =============================================================================================
+
+    template<int K, typename IntersectorK>
+    struct ArrayIntersectorKStream
+    {
+      typedef typename IntersectorK::Primitive PrimitiveK;
+      typedef typename IntersectorK::Precalculations PrecalculationsK;
+
+      static __forceinline void intersectK(const vbool<K>& valid, const Accel::Intersectors* This, /* PrecalculationsK& pre, */ RayHitK<K>& ray, RayQueryContext* context, const PrimitiveK* prim, size_t num, size_t& lazy_node)
+      {
+        PrecalculationsK pre(valid,ray); // FIXME: might cause trouble
+
+        for (size_t i=0; i<num; i++) {
+          IntersectorK::intersect(valid,pre,ray,context,prim[i]);
+        }
+      }
+
+      static __forceinline vbool<K> occludedK(const vbool<K>& valid, const Accel::Intersectors* This, /* PrecalculationsK& pre, */ RayK<K>& ray, RayQueryContext* context, const PrimitiveK* prim, size_t num, size_t& lazy_node)
+      {
+        PrecalculationsK pre(valid,ray); // FIXME: might cause trouble
+        vbool<K> valid0 = valid;
+        for (size_t i=0; i<num; i++) {
+          valid0 &= !IntersectorK::occluded(valid0,pre,ray,context,prim[i]);
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+
+      static __forceinline void intersect(const Accel::Intersectors* This, RayHitK<K>& ray, size_t k, RayQueryContext* context, const PrimitiveK* prim, size_t num, size_t& lazy_node)
+      {
+        PrecalculationsK pre(ray.tnear() <= ray.tfar,ray); // FIXME: might cause trouble
+        for (size_t i=0; i<num; i++) {
+          IntersectorK::intersect(pre,ray,k,context,prim[i]);
+        }
+      }
+
+      static __forceinline bool occluded(const Accel::Intersectors* This, RayK<K>& ray, size_t k, RayQueryContext* context, const PrimitiveK* prim, size_t num, size_t& lazy_node)
+      {
+        PrecalculationsK pre(ray.tnear() <= ray.tfar,ray); // FIXME: might cause trouble
+        for (size_t i=0; i<num; i++) {
+          if (IntersectorK::occluded(pre,ray,k,context,prim[i]))
+            return true;
+        }
+        return false;
+      }
+
+      static __forceinline size_t occluded(const Accel::Intersectors* This, size_t cur_mask, RayK<K>** __restrict__ inputPackets, RayQueryContext* context, const PrimitiveK* prim, size_t num, size_t& lazy_node)
+      {
+        size_t m_occluded = 0;
+        for (size_t i=0; i<num; i++) {
+          size_t bits = cur_mask & (~m_occluded);
+          for (; bits!=0; )
+          {
+            const size_t rayID = bscf(bits);
+            RayHitK<K> &ray = *inputPackets[rayID / K];
+            const size_t k = rayID % K;
+            PrecalculationsK pre(ray.tnear() <= ray.tfar,ray); // FIXME: might cause trouble
+            if (IntersectorK::occluded(pre,ray,k,context,prim[i]))
+            {
+              m_occluded |= (size_t)1 << rayID;
+              ray.tfar[k] = neg_inf;
+            }
+          }
+        }
+        return m_occluded;
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/line_intersector.h

@@ -0,0 +1,145 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+      struct LineIntersectorHitM
+      {
+        __forceinline LineIntersectorHitM() {}
+
+        __forceinline LineIntersectorHitM(const vfloat<M>& u, const vfloat<M>& v, const vfloat<M>& t, const Vec3vf<M>& Ng)
+          : vu(u), vv(v), vt(t), vNg(Ng) {}
+        
+        __forceinline void finalize() {}
+        
+        __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+        __forceinline float t  (const size_t i) const { return vt[i]; }
+        __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+
+        __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); }
+        __forceinline vfloat<M> t () const { return vt; }
+        __forceinline Vec3vf<M> Ng() const { return vNg; }
+        
+      public:
+        vfloat<M> vu;
+        vfloat<M> vv;
+        vfloat<M> vt;
+        Vec3vf<M> vNg;
+      };
+    
+    template<int M>
+      struct FlatLinearCurveIntersector1
+      {
+        typedef CurvePrecalculations1 Precalculations;
+        
+        template<typename Ray, typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            Ray& ray,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const Epilog& epilog)
+        {
+          /* transform end points into ray space */
+          vbool<M> valid = valid_i;
+          vfloat<M> depth_scale = pre.depth_scale;
+          LinearSpace3<Vec3vf<M>> ray_space = pre.ray_space;
+
+          const Vec3vf<M> ray_org ((Vec3fa)ray.org);
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          
+          Vec4vf<M> p0(xfmVector(ray_space,v0.xyz()-ray_org), v0.w);
+          Vec4vf<M> p1(xfmVector(ray_space,v1.xyz()-ray_org), v1.w);
+          
+          /* approximative intersection with cone */
+          const Vec4vf<M> v = p1-p0;
+          const Vec4vf<M> w = -p0;
+          const vfloat<M> d0 = madd(w.x,v.x,w.y*v.y);
+          const vfloat<M> d1 = madd(v.x,v.x,v.y*v.y);
+          const vfloat<M> u = clamp(d0*rcp(d1),vfloat<M>(zero),vfloat<M>(one));
+          const Vec4vf<M> p = madd(u,v,p0);
+          const vfloat<M> t = p.z;
+          const vfloat<M> d2 = madd(p.x,p.x,p.y*p.y);
+          const vfloat<M> r = p.w;
+          const vfloat<M> r2 = r*r;
+          valid &= (d2 <= r2) & (vfloat<M>(ray.tnear()) <= t) & (t <= vfloat<M>(ray.tfar));
+          if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) 
+            valid &= t > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale; // ignore self intersections
+          if (unlikely(none(valid))) return false;
+          
+          /* ignore denormalized segments */
+          const Vec3vf<M> T = v1.xyz()-v0.xyz();
+          valid &= (T.x != vfloat<M>(zero)) | (T.y != vfloat<M>(zero)) | (T.z != vfloat<M>(zero));
+          if (unlikely(none(valid))) return false;
+          
+          /* update hit information */
+          LineIntersectorHitM<M> hit(u,zero,t,T);
+          return epilog(valid,hit);
+        }
+      };
+    
+    template<int M, int K>
+      struct FlatLinearCurveIntersectorK
+      {
+        typedef CurvePrecalculationsK<K> Precalculations;
+        
+        template<typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            RayK<K>& ray, size_t k,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const Epilog& epilog)
+        {
+          /* transform end points into ray space */
+          vbool<M> valid = valid_i;
+          vfloat<M> depth_scale = pre.depth_scale[k];
+          LinearSpace3<Vec3vf<M>> ray_space = pre.ray_space[k];
+          const Vec3vf<M> ray_org(ray.org.x[k],ray.org.y[k],ray.org.z[k]);
+          const Vec3vf<M> ray_dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]);
+
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          
+          Vec4vf<M> p0(xfmVector(ray_space,v0.xyz()-ray_org), v0.w);
+          Vec4vf<M> p1(xfmVector(ray_space,v1.xyz()-ray_org), v1.w);
+          
+          /* approximative intersection with cone */
+          const Vec4vf<M> v = p1-p0;
+          const Vec4vf<M> w = -p0;
+          const vfloat<M> d0 = madd(w.x,v.x,w.y*v.y);
+          const vfloat<M> d1 = madd(v.x,v.x,v.y*v.y);
+          const vfloat<M> u = clamp(d0*rcp(d1),vfloat<M>(zero),vfloat<M>(one));
+          const Vec4vf<M> p = madd(u,v,p0);
+          const vfloat<M> t = p.z;
+          const vfloat<M> d2 = madd(p.x,p.x,p.y*p.y);
+          const vfloat<M> r = p.w;
+          const vfloat<M> r2 = r*r;
+          valid &= (d2 <= r2) & (vfloat<M>(ray.tnear()[k]) <= t) & (t <= vfloat<M>(ray.tfar[k]));
+          if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) 
+            valid &= t > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale; // ignore self intersections
+          if (unlikely(none(valid))) return false;
+          
+          /* ignore denormalized segments */
+          const Vec3vf<M> T = v1.xyz()-v0.xyz();
+          valid &= (T.x != vfloat<M>(zero)) | (T.y != vfloat<M>(zero)) | (T.z != vfloat<M>(zero));
+          if (unlikely(none(valid))) return false;
+          
+          /* update hit information */
+          LineIntersectorHitM<M> hit(u,zero,t,T);
+          return epilog(valid,hit);
+        }
+      };
+  }
+}

thirdparty/embree/kernels/geometry/linei.h

@@ -0,0 +1,711 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  template<int M>
+  struct LineMi
+  {
+    /* Virtual interface to query information about the line segment type */
+    struct Type : public PrimitiveType
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;      
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored line segments */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N line segments */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+
+    /* Returns required number of bytes for N line segments */
+    static __forceinline size_t bytes(size_t N) { return blocks(N)*sizeof(LineMi); }
+
+  public:
+
+    /* Default constructor */
+    __forceinline LineMi() {  }
+
+    /* Construction from vertices and IDs */
+    __forceinline LineMi(const vuint<M>& v0, unsigned short leftExists, unsigned short rightExists, const vuint<M>& geomIDs, const vuint<M>& primIDs, Geometry::GType gtype)
+      : gtype((unsigned char)gtype), m((unsigned char)popcnt(vuint<M>(primIDs) != vuint<M>(-1))), sharedGeomID(geomIDs[0]), leftExists (leftExists), rightExists(rightExists), v0(v0), primIDs(primIDs)
+    {
+      assert(all(vuint<M>(geomID()) == geomIDs));
+    }
+
+    /* Returns a mask that tells which line segments are valid */
+    __forceinline vbool<M> valid() const { return primIDs != vuint<M>(-1); }
+
+    /* Returns if the specified line segment is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return primIDs[i] != -1; }
+
+    /* Returns the number of stored line segments */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    //template<class T>
+    //static __forceinline T unmask(T &index) { return index & 0x3fffffff; }
+
+    __forceinline     unsigned int geomID(unsigned int i = 0) const { return sharedGeomID; }
+    //__forceinline       vuint<M> geomID()       { return unmask(geomIDs); }
+    //__forceinline const vuint<M> geomID() const { return unmask(geomIDs); }
+    //__forceinline unsigned int geomID(const size_t i) const { assert(i<M); return unmask(geomIDs[i]); }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M>& primID()       { return primIDs; }
+    __forceinline const vuint<M>& primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* gather the line segments */
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              const LineSegments* geom) const;
+
+    __forceinline void gatheri(Vec4vf<M>& p0,
+                               Vec4vf<M>& p1,
+                               const LineSegments* geom,
+                               const int itime) const;
+
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              const LineSegments* geom,
+                              float time) const;
+
+    /* gather the line segments with lateral info */
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              Vec4vf<M>& pL,
+                              Vec4vf<M>& pR,
+                              const LineSegments* geom) const;
+
+    __forceinline void gatheri(Vec4vf<M>& p0,
+                               Vec4vf<M>& p1,
+                               Vec4vf<M>& pL,
+                               Vec4vf<M>& pR,
+                               const LineSegments* geom,
+                               const int itime) const;
+
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              Vec4vf<M>& pL,
+                              Vec4vf<M>& pR,
+                              const LineSegments* geom,
+                              float time) const;
+
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              vbool<M>& cL,
+                              vbool<M>& cR,
+                              const LineSegments* geom) const;
+
+    __forceinline void gatheri(Vec4vf<M>& p0,
+                               Vec4vf<M>& p1,
+                               vbool<M>& cL,
+                               vbool<M>& cR,
+                               const LineSegments* geom,
+                               const int itime) const;
+
+    __forceinline void gather(Vec4vf<M>& p0,
+                              Vec4vf<M>& p1,
+                              vbool<M>& cL,
+                              vbool<M>& cR,
+                              const LineSegments* geom,
+                              float time) const;
+
+    /* Calculate the bounds of the line segments */
+    __forceinline const BBox3fa bounds(const Scene* scene, size_t itime = 0) const
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const LineSegments* geom = scene->get<LineSegments>(geomID(i));
+        const Vec3ff& p0 = geom->vertex(v0[i]+0,itime);
+        const Vec3ff& p1 = geom->vertex(v0[i]+1,itime);
+        BBox3fa b = merge(BBox3fa(p0),BBox3fa(p1));
+        b = enlarge(b,Vec3fa(max(p0.w,p1.w)));
+        bounds.extend(b);
+      }
+      return bounds;
+    }
+
+    /* Calculate the linear bounds of the primitive */
+    __forceinline LBBox3fa linearBounds(const Scene* scene, size_t itime) {
+      return LBBox3fa(bounds(scene,itime+0), bounds(scene,itime+1));
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, size_t itime, size_t numTimeSteps) {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const LineSegments* geom = scene->get<LineSegments>(geomID(i));
+        allBounds.extend(geom->linearBounds(primID(i), itime, numTimeSteps));
+      }
+      return allBounds;
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, const BBox1f time_range) 
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const LineSegments* geom = scene->get<LineSegments>(geomID((unsigned int)i));
+        allBounds.extend(geom->linearBounds(primID(i), time_range));
+      }
+      return allBounds;
+    }
+
+    /* Fill line segment from line segment list */
+    template<typename PrimRefT>
+    __forceinline void fill(const PrimRefT* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      Geometry::GType gty = scene->get(prims[begin].geomID())->getType();
+      vuint<M> geomID, primID;
+      vuint<M> v0;
+      unsigned short leftExists = 0;
+      unsigned short rightExists = 0;
+      const PrimRefT* prim = &prims[begin];
+
+      for (size_t i=0; i<M; i++)
+      {
+        const LineSegments* geom = scene->get<LineSegments>(prim->geomID());
+        if (begin<end) {
+          geomID[i] = prim->geomID();
+          primID[i] = prim->primID();
+          v0[i] = geom->segment(prim->primID());
+          leftExists |= geom->segmentLeftExists(primID[i]) << i;
+          rightExists |= geom->segmentRightExists(primID[i]) << i;         
+          begin++;
+        } else {
+          assert(i);
+          if (i>0) {
+            geomID[i] = geomID[i-1];
+            primID[i] = -1;
+            v0[i] = v0[i-1];
+          }
+        }
+        if (begin<end) prim = &prims[begin]; // FIXME: remove this line
+      }
+      new (this) LineMi(v0,leftExists,rightExists,geomID,primID,gty); // FIXME: use non temporal store
+    }
+
+     template<typename BVH, typename Allocator>
+      __forceinline static typename BVH::NodeRef createLeaf (BVH* bvh, const PrimRef* prims, const range<size_t>& set, const Allocator& alloc)
+    {
+      size_t start = set.begin();
+      size_t items = LineMi::blocks(set.size());
+      size_t numbytes = LineMi::bytes(set.size());
+      LineMi* accel = (LineMi*) alloc.malloc1(numbytes,M*sizeof(float));
+      for (size_t i=0; i<items; i++) {
+        accel[i].fill(prims,start,set.end(),bvh->scene);
+      }
+      return bvh->encodeLeaf((char*)accel,items);
+    };
+    
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& begin, size_t end, Scene* scene, size_t itime)
+    {
+      fill(prims,begin,end,scene);
+      return linearBounds(scene,itime);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& begin, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      fill(prims,begin,end,scene);
+      return linearBounds(scene,time_range);
+    }
+
+      template<typename BVH, typename SetMB, typename Allocator>
+    __forceinline static typename BVH::NodeRecordMB4D createLeafMB(BVH* bvh, const SetMB& prims, const Allocator& alloc)
+    {
+      size_t start = prims.begin();
+      size_t end   = prims.end();
+      size_t items = LineMi::blocks(prims.size());
+      size_t numbytes = LineMi::bytes(prims.size());
+      LineMi* accel = (LineMi*) alloc.malloc1(numbytes,M*sizeof(float));
+      const typename BVH::NodeRef node = bvh->encodeLeaf((char*)accel,items);
+      
+      LBBox3fa bounds = empty;
+      for (size_t i=0; i<items; i++)
+        bounds.extend(accel[i].fillMB(prims.prims->data(),start,end,bvh->scene,prims.time_range));
+      
+      return typename BVH::NodeRecordMB4D(node,bounds,prims.time_range);
+    };
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(LineSegments* geom)
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const Vec3ff& p0 = geom->vertex(v0[i]+0);
+        const Vec3ff& p1 = geom->vertex(v0[i]+1);
+        BBox3fa b = merge(BBox3fa(p0),BBox3fa(p1));
+        b = enlarge(b,Vec3fa(max(p0.w,p1.w)));
+        bounds.extend(b);
+      }
+      return bounds;
+    }
+
+    /*! output operator */
+    friend __forceinline embree_ostream operator<<(embree_ostream cout, const LineMi& line) {
+      return cout << "Line" << M << "i {" << line.v0 << ", " << line.geomID() << ", " << line.primID() << "}";
+    }
+    
+  public:
+    unsigned char gtype;
+    unsigned char m;
+    unsigned int sharedGeomID;
+    unsigned short leftExists, rightExists;
+    vuint<M> v0;      // index of start vertex
+  private:
+    vuint<M> primIDs; // primitive ID
+  };
+
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                         Vec4vf4& p1,
+                                         const LineSegments* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0]));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1]));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2]));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3]));
+    transpose(a0,a1,a2,a3,p0.x,p0.y,p0.z,p0.w);
+
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1));
+    transpose(b0,b1,b2,b3,p1.x,p1.y,p1.z,p1.w);
+  }
+
+  template<>
+  __forceinline void LineMi<4>::gatheri(Vec4vf4& p0,
+                                       Vec4vf4& p1,
+                                       const LineSegments* geom,
+                                       const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0],itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1],itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2],itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3],itime));
+    transpose(a0,a1,a2,a3,p0.x,p0.y,p0.z,p0.w);
+
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1,itime));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1,itime));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1,itime));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1,itime));
+    transpose(b0,b1,b2,b3,p1.x,p1.y,p1.z,p1.w);
+  }
+
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                         Vec4vf4& p1,
+                                         const LineSegments* geom,
+                                         float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf4 a0,a1;
+    gatheri(a0,a1,geom,itime);
+    Vec4vf4 b0,b1;
+    gatheri(b0,b1,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat4(ftime));
+    p1 = lerp(a1,b1,vfloat4(ftime));
+  }
+
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                         Vec4vf4& p1,
+                                         vbool4&  cL,
+                                         vbool4&  cR,
+                                         const LineSegments* geom) const
+  {
+    gather(p0,p1,geom);
+    cL = !vbool4(leftExists);
+    cR = !vbool4(rightExists);
+  }
+
+  template<>
+    __forceinline void LineMi<4>::gatheri(Vec4vf4& p0,
+                                          Vec4vf4& p1,
+                                          vbool4&  cL,
+                                          vbool4&  cR,
+                                          const LineSegments* geom,
+                                          const int itime) const
+  {
+    gatheri(p0,p1,geom,itime);
+    cL = !vbool4(leftExists);
+    cR = !vbool4(rightExists);
+  }
+
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                         Vec4vf4& p1,
+                                         vbool4&  cL,
+                                         vbool4&  cR,
+                                         const LineSegments* geom,
+                                         float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+    
+    Vec4vf4 a0,a1;
+    gatheri(a0,a1,geom,itime);
+    Vec4vf4 b0,b1;
+    gatheri(b0,b1,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat4(ftime));
+    p1 = lerp(a1,b1,vfloat4(ftime));
+    cL = !vbool4(leftExists);
+    cR = !vbool4(rightExists);
+  }
+
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                              Vec4vf4& p1,
+                                              Vec4vf4& pL,
+                                              Vec4vf4& pR,
+                                              const LineSegments* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0]));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1]));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2]));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3]));
+    transpose(a0,a1,a2,a3,p0.x,p0.y,p0.z,p0.w);
+
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1));
+    transpose(b0,b1,b2,b3,p1.x,p1.y,p1.z,p1.w);
+    
+    const vfloat4 l0 = (leftExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]-1)) : vfloat4(inf);
+    const vfloat4 l1 = (leftExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]-1)) : vfloat4(inf);
+    const vfloat4 l2 = (leftExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]-1)) : vfloat4(inf);
+    const vfloat4 l3 = (leftExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]-1)) : vfloat4(inf);
+    transpose(l0,l1,l2,l3,pL.x,pL.y,pL.z,pL.w);
+    
+    const vfloat4 r0 = (rightExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]+2)) : vfloat4(inf);
+    const vfloat4 r1 = (rightExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]+2)) : vfloat4(inf);
+    const vfloat4 r2 = (rightExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]+2)) : vfloat4(inf);
+    const vfloat4 r3 = (rightExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]+2)) : vfloat4(inf);
+    transpose(r0,r1,r2,r3,pR.x,pR.y,pR.z,pR.w);
+  }
+  
+  template<>
+    __forceinline void LineMi<4>::gatheri(Vec4vf4& p0,
+                                              Vec4vf4& p1,
+                                              Vec4vf4& pL,
+                                              Vec4vf4& pR,
+                                              const LineSegments* geom,
+                                              const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0],itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1],itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2],itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3],itime));
+    transpose(a0,a1,a2,a3,p0.x,p0.y,p0.z,p0.w);
+    
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1,itime));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1,itime));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1,itime));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1,itime));
+    transpose(b0,b1,b2,b3,p1.x,p1.y,p1.z,p1.w);
+    
+    const vfloat4 l0 = (leftExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]-1,itime)) : vfloat4(inf);
+    const vfloat4 l1 = (leftExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]-1,itime)) : vfloat4(inf);
+    const vfloat4 l2 = (leftExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]-1,itime)) : vfloat4(inf);
+    const vfloat4 l3 = (leftExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]-1,itime)) : vfloat4(inf);
+    transpose(l0,l1,l2,l3,pL.x,pL.y,pL.z,pL.w);
+    
+    const vfloat4 r0 = (rightExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]+2,itime)) : vfloat4(inf);
+    const vfloat4 r1 = (rightExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]+2,itime)) : vfloat4(inf);
+    const vfloat4 r2 = (rightExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]+2,itime)) : vfloat4(inf);
+    const vfloat4 r3 = (rightExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]+2,itime)) : vfloat4(inf);
+    transpose(r0,r1,r2,r3,pR.x,pR.y,pR.z,pR.w);
+  }
+  
+  template<>
+    __forceinline void LineMi<4>::gather(Vec4vf4& p0,
+                                              Vec4vf4& p1,
+                                              Vec4vf4& pL,
+                                              Vec4vf4& pR,
+                                              const LineSegments* geom,
+                                              float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+    
+    Vec4vf4 a0,a1,aL,aR;
+    gatheri(a0,a1,aL,aR,geom,itime);
+    Vec4vf4 b0,b1,bL,bR;
+    gatheri(b0,b1,bL,bR,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat4(ftime));
+    p1 = lerp(a1,b1,vfloat4(ftime));
+    pL = lerp(aL,bL,vfloat4(ftime));
+    pR = lerp(aR,bR,vfloat4(ftime));
+
+    pL = select(vboolf4(leftExists), pL, Vec4vf4(inf));
+    pR = select(vboolf4(rightExists), pR, Vec4vf4(inf));
+  }
+
+#if defined(__AVX__)
+
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                         Vec4vf8& p1,
+                                         const LineSegments* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0]));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1]));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2]));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3]));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(v0[4]));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(v0[5]));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(v0[6]));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(v0[7]));
+    transpose(a0,a1,a2,a3,a4,a5,a6,a7,p0.x,p0.y,p0.z,p0.w);
+
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1));
+    const vfloat4 b4 = vfloat4::loadu(geom->vertexPtr(v0[4]+1));
+    const vfloat4 b5 = vfloat4::loadu(geom->vertexPtr(v0[5]+1));
+    const vfloat4 b6 = vfloat4::loadu(geom->vertexPtr(v0[6]+1));
+    const vfloat4 b7 = vfloat4::loadu(geom->vertexPtr(v0[7]+1));
+    transpose(b0,b1,b2,b3,b4,b5,b6,b7,p1.x,p1.y,p1.z,p1.w);
+  }
+
+  template<>
+  __forceinline void LineMi<8>::gatheri(Vec4vf8& p0,
+                                       Vec4vf8& p1,
+                                       const LineSegments* geom,
+                                       const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0],itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1],itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2],itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3],itime));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(v0[4],itime));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(v0[5],itime));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(v0[6],itime));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(v0[7],itime));
+    transpose(a0,a1,a2,a3,a4,a5,a6,a7,p0.x,p0.y,p0.z,p0.w);
+
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1,itime));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1,itime));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1,itime));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1,itime));
+    const vfloat4 b4 = vfloat4::loadu(geom->vertexPtr(v0[4]+1,itime));
+    const vfloat4 b5 = vfloat4::loadu(geom->vertexPtr(v0[5]+1,itime));
+    const vfloat4 b6 = vfloat4::loadu(geom->vertexPtr(v0[6]+1,itime));
+    const vfloat4 b7 = vfloat4::loadu(geom->vertexPtr(v0[7]+1,itime));
+    transpose(b0,b1,b2,b3,b4,b5,b6,b7,p1.x,p1.y,p1.z,p1.w);
+  }
+
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                         Vec4vf8& p1,
+                                         const LineSegments* geom,
+                                         float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf8 a0,a1;
+    gatheri(a0,a1,geom,itime);
+    Vec4vf8 b0,b1;
+    gatheri(b0,b1,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat8(ftime));
+    p1 = lerp(a1,b1,vfloat8(ftime));
+  }
+  
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                              Vec4vf8& p1,
+                                              Vec4vf8& pL,
+                                              Vec4vf8& pR,
+                                              const LineSegments* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0]));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1]));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2]));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3]));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(v0[4]));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(v0[5]));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(v0[6]));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(v0[7]));
+    transpose(a0,a1,a2,a3,a4,a5,a6,a7,p0.x,p0.y,p0.z,p0.w);
+    
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1));
+    const vfloat4 b4 = vfloat4::loadu(geom->vertexPtr(v0[4]+1));
+    const vfloat4 b5 = vfloat4::loadu(geom->vertexPtr(v0[5]+1));
+    const vfloat4 b6 = vfloat4::loadu(geom->vertexPtr(v0[6]+1));
+    const vfloat4 b7 = vfloat4::loadu(geom->vertexPtr(v0[7]+1));
+    transpose(b0,b1,b2,b3,b4,b5,b6,b7,p1.x,p1.y,p1.z,p1.w);
+    
+    const vfloat4 l0 = (leftExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]-1)) : vfloat4(inf);
+    const vfloat4 l1 = (leftExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]-1)) : vfloat4(inf);
+    const vfloat4 l2 = (leftExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]-1)) : vfloat4(inf);
+    const vfloat4 l3 = (leftExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]-1)) : vfloat4(inf);
+    const vfloat4 l4 = (leftExists & (1<<4)) ? vfloat4::loadu(geom->vertexPtr(v0[4]-1)) : vfloat4(inf);
+    const vfloat4 l5 = (leftExists & (1<<5)) ? vfloat4::loadu(geom->vertexPtr(v0[5]-1)) : vfloat4(inf);
+    const vfloat4 l6 = (leftExists & (1<<6)) ? vfloat4::loadu(geom->vertexPtr(v0[6]-1)) : vfloat4(inf);
+    const vfloat4 l7 = (leftExists & (1<<7)) ? vfloat4::loadu(geom->vertexPtr(v0[7]-1)) : vfloat4(inf);
+    transpose(l0,l1,l2,l3,l4,l5,l6,l7,pL.x,pL.y,pL.z,pL.w);
+    
+    const vfloat4 r0 = (rightExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]+2)) : vfloat4(inf);
+    const vfloat4 r1 = (rightExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]+2)) : vfloat4(inf);
+    const vfloat4 r2 = (rightExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]+2)) : vfloat4(inf);
+    const vfloat4 r3 = (rightExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]+2)) : vfloat4(inf);
+    const vfloat4 r4 = (rightExists & (1<<4)) ? vfloat4::loadu(geom->vertexPtr(v0[4]+2)) : vfloat4(inf);
+    const vfloat4 r5 = (rightExists & (1<<5)) ? vfloat4::loadu(geom->vertexPtr(v0[5]+2)) : vfloat4(inf);
+    const vfloat4 r6 = (rightExists & (1<<6)) ? vfloat4::loadu(geom->vertexPtr(v0[6]+2)) : vfloat4(inf);
+    const vfloat4 r7 = (rightExists & (1<<7)) ? vfloat4::loadu(geom->vertexPtr(v0[7]+2)) : vfloat4(inf);
+    transpose(r0,r1,r2,r3,r4,r5,r6,r7,pR.x,pR.y,pR.z,pR.w);
+  }
+  
+  template<>
+    __forceinline void LineMi<8>::gatheri(Vec4vf8& p0,
+                                              Vec4vf8& p1,
+                                              Vec4vf8& pL,
+                                              Vec4vf8& pR,
+                                              const LineSegments* geom,
+                                              const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(v0[0],itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(v0[1],itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(v0[2],itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(v0[3],itime));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(v0[4],itime));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(v0[5],itime));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(v0[6],itime));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(v0[7],itime));
+    transpose(a0,a1,a2,a3,a4,a5,a6,a7,p0.x,p0.y,p0.z,p0.w);
+    
+    const vfloat4 b0 = vfloat4::loadu(geom->vertexPtr(v0[0]+1,itime));
+    const vfloat4 b1 = vfloat4::loadu(geom->vertexPtr(v0[1]+1,itime));
+    const vfloat4 b2 = vfloat4::loadu(geom->vertexPtr(v0[2]+1,itime));
+    const vfloat4 b3 = vfloat4::loadu(geom->vertexPtr(v0[3]+1,itime));
+    const vfloat4 b4 = vfloat4::loadu(geom->vertexPtr(v0[4]+1,itime));
+    const vfloat4 b5 = vfloat4::loadu(geom->vertexPtr(v0[5]+1,itime));
+    const vfloat4 b6 = vfloat4::loadu(geom->vertexPtr(v0[6]+1,itime));
+    const vfloat4 b7 = vfloat4::loadu(geom->vertexPtr(v0[7]+1,itime));
+    transpose(b0,b1,b2,b3,b4,b5,b6,b7,p1.x,p1.y,p1.z,p1.w);
+    
+    const vfloat4 l0 = (leftExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]-1,itime)) : vfloat4(inf);
+    const vfloat4 l1 = (leftExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]-1,itime)) : vfloat4(inf);
+    const vfloat4 l2 = (leftExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]-1,itime)) : vfloat4(inf);
+    const vfloat4 l3 = (leftExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]-1,itime)) : vfloat4(inf);
+    const vfloat4 l4 = (leftExists & (1<<4)) ? vfloat4::loadu(geom->vertexPtr(v0[4]-1,itime)) : vfloat4(inf);
+    const vfloat4 l5 = (leftExists & (1<<5)) ? vfloat4::loadu(geom->vertexPtr(v0[5]-1,itime)) : vfloat4(inf);
+    const vfloat4 l6 = (leftExists & (1<<6)) ? vfloat4::loadu(geom->vertexPtr(v0[6]-1,itime)) : vfloat4(inf);
+    const vfloat4 l7 = (leftExists & (1<<7)) ? vfloat4::loadu(geom->vertexPtr(v0[7]-1,itime)) : vfloat4(inf);
+    transpose(l0,l1,l2,l3,l4,l5,l6,l7,pL.x,pL.y,pL.z,pL.w);
+    
+    const vfloat4 r0 = (rightExists & (1<<0)) ? vfloat4::loadu(geom->vertexPtr(v0[0]+2,itime)) : vfloat4(inf);
+    const vfloat4 r1 = (rightExists & (1<<1)) ? vfloat4::loadu(geom->vertexPtr(v0[1]+2,itime)) : vfloat4(inf);
+    const vfloat4 r2 = (rightExists & (1<<2)) ? vfloat4::loadu(geom->vertexPtr(v0[2]+2,itime)) : vfloat4(inf);
+    const vfloat4 r3 = (rightExists & (1<<3)) ? vfloat4::loadu(geom->vertexPtr(v0[3]+2,itime)) : vfloat4(inf);
+    const vfloat4 r4 = (rightExists & (1<<4)) ? vfloat4::loadu(geom->vertexPtr(v0[4]+2,itime)) : vfloat4(inf);
+    const vfloat4 r5 = (rightExists & (1<<5)) ? vfloat4::loadu(geom->vertexPtr(v0[5]+2,itime)) : vfloat4(inf);
+    const vfloat4 r6 = (rightExists & (1<<6)) ? vfloat4::loadu(geom->vertexPtr(v0[6]+2,itime)) : vfloat4(inf);
+    const vfloat4 r7 = (rightExists & (1<<7)) ? vfloat4::loadu(geom->vertexPtr(v0[7]+2,itime)) : vfloat4(inf);
+    transpose(r0,r1,r2,r3,r4,r5,r6,r7,pR.x,pR.y,pR.z,pR.w);
+  }
+  
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                              Vec4vf8& p1,
+                                              Vec4vf8& pL,
+                                              Vec4vf8& pR,
+                                              const LineSegments* geom,
+                                              float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+    
+    Vec4vf8 a0,a1,aL,aR;
+    gatheri(a0,a1,aL,aR,geom,itime);
+    Vec4vf8 b0,b1,bL,bR;
+    gatheri(b0,b1,bL,bR,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat8(ftime));
+    p1 = lerp(a1,b1,vfloat8(ftime));
+    pL = lerp(aL,bL,vfloat8(ftime));
+    pR = lerp(aR,bR,vfloat8(ftime));
+    
+    pL = select(vboolf4(leftExists), pL, Vec4vf8(inf));
+    pR = select(vboolf4(rightExists), pR, Vec4vf8(inf));
+  }
+
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                         Vec4vf8& p1,
+                                         vbool8& cL,
+                                         vbool8& cR,
+                                         const LineSegments* geom) const
+  {
+    gather(p0,p1,geom);
+    cL = !vbool8(leftExists);
+    cR = !vbool8(rightExists);
+  }
+  
+  template<>
+    __forceinline void LineMi<8>::gatheri(Vec4vf8& p0,
+                                              Vec4vf8& p1,
+                                              vbool8& cL,
+                                              vbool8& cR,
+                                              const LineSegments* geom,
+                                              const int itime) const
+  {
+    gatheri(p0,p1,geom,itime);
+    cL = !vbool8(leftExists);
+    cR = !vbool8(rightExists);
+  }
+  
+  template<>
+    __forceinline void LineMi<8>::gather(Vec4vf8& p0,
+                                              Vec4vf8& p1,
+                                              vbool8& cL,
+                                              vbool8& cR,
+                                              const LineSegments* geom,
+                                              float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+    
+    Vec4vf8 a0,a1;
+    gatheri(a0,a1,geom,itime);
+    Vec4vf8 b0,b1;
+    gatheri(b0,b1,geom,itime+1);
+    p0 = lerp(a0,b0,vfloat8(ftime));
+    p1 = lerp(a1,b1,vfloat8(ftime));
+    cL = !vbool8(leftExists);
+    cR = !vbool8(rightExists);
+  }
+  
+#endif
+  
+  template<int M>
+  typename LineMi<M>::Type LineMi<M>::type;
+
+  typedef LineMi<4> Line4i;
+  typedef LineMi<8> Line8i;
+}

thirdparty/embree/kernels/geometry/linei_intersector.h

@@ -0,0 +1,124 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "linei.h"
+#include "line_intersector.h"
+#include "intersector_epilog.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, bool filter>
+    struct FlatLinearCurveMiIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom);
+        const vbool<M> valid = line.valid();
+        FlatLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom);
+        const vbool<M> valid = line.valid();
+        return FlatLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, bool filter>
+    struct FlatLinearCurveMiMBIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        FlatLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        return FlatLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct FlatLinearCurveMiIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom);
+        const vbool<M> valid = line.valid();
+        FlatLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom);
+        const vbool<M> valid = line.valid();
+        return FlatLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct FlatLinearCurveMiMBIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context,  const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        FlatLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1; line.gather(v0,v1,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        return FlatLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/object.h

@@ -0,0 +1,95 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  struct Object
+  {
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored primitives */
+    static __forceinline size_t max_size() { return 1; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return N; }
+
+  public:
+
+    /*! constructs a virtual object */
+    Object (unsigned geomID, unsigned primID) 
+    : _geomID(geomID), _primID(primID) {}
+
+    __forceinline unsigned geomID() const {
+      return _geomID;
+    }
+
+    __forceinline unsigned primID() const {
+      return _primID;
+    }
+
+    /*! fill triangle from triangle list */
+    __forceinline void fill(const PrimRef* prims, size_t& i, size_t end, Scene* scene)
+    {
+      const PrimRef& prim = prims[i]; i++;
+      new (this) Object(prim.geomID(), prim.primID());
+    }
+
+    /*! fill triangle from triangle list */
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& i, size_t end, Scene* scene, size_t itime)
+    {
+      const PrimRef& prim = prims[i]; i++;
+      const unsigned geomID = prim.geomID();
+      const unsigned primID = prim.primID();
+      new (this) Object(geomID, primID);
+      AccelSet* accel = (AccelSet*) scene->get(geomID);
+      return accel->linearBounds(primID,itime);
+    }
+
+    /*! fill triangle from triangle list */
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t i, Scene* scene, const BBox1f time_range)
+    {
+      const PrimRefMB& prim = prims[i]; i++;
+      const unsigned geomID = prim.geomID();
+      const unsigned primID = prim.primID();
+      new (this) Object(geomID, primID);
+      AccelSet* accel = (AccelSet*) scene->get(geomID);
+      return accel->linearBounds(primID,time_range);
+    }
+
+     /*! fill triangle from triangle list */
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& i, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      const PrimRefMB& prim = prims[i]; i++;
+      const unsigned geomID = prim.geomID();
+      const unsigned primID = prim.primID();
+      new (this) Object(geomID, primID);
+      AccelSet* accel = (AccelSet*) scene->get(geomID);
+      return accel->linearBounds(primID,time_range);
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(AccelSet* mesh) {
+      return mesh->bounds(primID());
+    }
+
+  private:
+    unsigned int _geomID;  //!< geometry ID
+    unsigned int _primID;  //!< primitive ID
+  };
+}

thirdparty/embree/kernels/geometry/object_intersector.h

@@ -0,0 +1,161 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "object.h"
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<bool mblur>
+    struct ObjectIntersector1
+    {
+      typedef Object Primitive;
+     
+      static const bool validIntersectorK = false;
+
+      struct Precalculations {
+        __forceinline Precalculations() {}
+        __forceinline Precalculations (const Ray& ray, const void *ptr) {}
+      };
+      
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim) 
+      {
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        if ((ray.mask & accel->mask) == 0) 
+          return;
+#endif
+
+        accel->intersect(ray,prim.geomID(),prim.primID(),context);
+      }
+      
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        if ((ray.mask & accel->mask) == 0) 
+          return false;
+#endif
+
+        accel->occluded(ray,prim.geomID(),prim.primID(),context);
+        return ray.tfar < 0.0f;
+      }
+
+      static __forceinline bool intersect(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim) {
+        return occluded(pre,ray,context,prim);
+      }
+
+      static __forceinline void intersect(unsigned int k, const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& prim) 
+      {
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        if ((ray.mask & accel->mask) == 0) 
+          return;
+#endif
+
+        accel->intersect(k,ray,prim.geomID(),prim.primID(),context);
+      }
+      
+      static __forceinline bool occluded(unsigned int k, const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        if ((ray.mask & accel->mask) == 0) 
+          return false;
+#endif
+
+        accel->occluded(k, ray,prim.geomID(),prim.primID(),context);
+        return ray.tfar < 0.0f;
+      }
+
+      static __forceinline bool intersect(unsigned int k, const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& prim) {
+        return occluded(k,pre,ray,context,prim);
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim)
+      {
+        AccelSet* accel = (AccelSet*)context->scene->get(prim.geomID());
+        context->geomID = prim.geomID();
+        context->primID = prim.primID();
+        return accel->pointQuery(query, context);
+      }
+      
+      template<int K>
+      static __forceinline void intersectK(const vbool<K>& valid, /* PrecalculationsK& pre, */ RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+      {
+        assert(false);
+      }
+
+      template<int K>
+      static __forceinline vbool<K> occludedK(const vbool<K>& valid, /* PrecalculationsK& pre, */ RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, size_t& lazy_node)
+      {
+        assert(false);
+        return valid;
+      }
+    };
+
+    template<int K, bool mblur>
+    struct ObjectIntersectorK
+    {
+      typedef Object Primitive;
+      
+      struct Precalculations {
+        __forceinline Precalculations (const vbool<K>& valid, const RayK<K>& ray) {}
+      };
+      
+      static __forceinline void intersect(const vbool<K>& valid_i, const Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vbool<K> valid = valid_i;
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+        
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        valid &= (ray.mask & accel->mask) != 0;
+        if (none(valid)) return;
+#endif
+        accel->intersect(valid,ray,prim.geomID(),prim.primID(),context);
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, const Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& prim)
+      {
+        vbool<K> valid = valid_i;
+        AccelSet* accel = (AccelSet*) context->scene->get(prim.geomID());
+        
+        /* perform ray mask test */
+#if defined(EMBREE_RAY_MASK)
+        valid &= (ray.mask & accel->mask) != 0;
+        if (none(valid)) return false;
+#endif
+        accel->occluded(valid,ray,prim.geomID(),prim.primID(),context);
+        return ray.tfar < 0.0f;
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        intersect(vbool<K>(1<<int(k)),pre,ray,context,prim);
+      }
+      
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& prim) {
+        occluded(vbool<K>(1<<int(k)),pre,ray,context,prim);
+        return ray.tfar[k] < 0.0f; 
+      }
+    };
+
+    typedef ObjectIntersectorK<4,false>  ObjectIntersector4;
+    typedef ObjectIntersectorK<8,false>  ObjectIntersector8;
+    typedef ObjectIntersectorK<16,false> ObjectIntersector16;
+
+    typedef ObjectIntersectorK<4,true>  ObjectIntersector4MB;
+    typedef ObjectIntersectorK<8,true>  ObjectIntersector8MB;
+    typedef ObjectIntersectorK<16,true> ObjectIntersector16MB;
+  }
+}

thirdparty/embree/kernels/geometry/plane.h

@@ -0,0 +1,57 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    struct HalfPlane
+    {
+      const Vec3fa P;  //!< plane origin
+      const Vec3fa N;  //!< plane normal
+
+      __forceinline HalfPlane(const Vec3fa& P, const Vec3fa& N) 
+        : P(P), N(N) {}
+      
+      __forceinline BBox1f intersect(const Vec3fa& ray_org, const Vec3fa& ray_dir) const
+      {
+        Vec3fa O = Vec3fa(ray_org) - P;
+        Vec3fa D = Vec3fa(ray_dir);
+        float ON = dot(O,N);
+        float DN = dot(D,N);
+        bool eps = abs(DN) < min_rcp_input;
+        float t = -ON*rcp(DN);
+        float lower = select(eps || DN < 0.0f, float(neg_inf), t);
+        float upper = select(eps || DN > 0.0f, float(pos_inf), t);
+        return BBox1f(lower,upper);
+      }
+    };
+
+    template<int M>
+      struct HalfPlaneN
+      {
+        const Vec3vf<M> P;  //!< plane origin
+        const Vec3vf<M> N;  //!< plane normal
+
+        __forceinline HalfPlaneN(const Vec3vf<M>& P, const Vec3vf<M>& N)
+          : P(P), N(N) {}
+
+        __forceinline BBox<vfloat<M>> intersect(const Vec3fa& ray_org, const Vec3fa& ray_dir) const
+        {
+          Vec3vf<M> O = Vec3vf<M>((Vec3fa)ray_org) - P;
+          Vec3vf<M> D = Vec3vf<M>((Vec3fa)ray_dir);
+          vfloat<M> ON = dot(O,N);
+          vfloat<M> DN = dot(D,N);
+          vbool<M> eps = abs(DN) < min_rcp_input;
+          vfloat<M> t = -ON*rcp(DN);
+          vfloat<M> lower = select(eps | DN < 0.0f, vfloat<M>(neg_inf), t);
+          vfloat<M> upper = select(eps | DN > 0.0f, vfloat<M>(pos_inf), t);
+          return BBox<vfloat<M>>(lower,upper);
+        }
+      };
+  }
+}

thirdparty/embree/kernels/geometry/pointi.h

@@ -0,0 +1,412 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  template<int M>
+  struct PointMi
+  {
+    /* Virtual interface to query information about the line segment type */
+    struct Type : public PrimitiveType
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+   public:
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored line segments */
+    static __forceinline size_t max_size()
+    {
+      return M;
+    }
+
+    /* Returns required number of primitive blocks for N line segments */
+    static __forceinline size_t blocks(size_t N)
+    {
+      return (N + max_size() - 1) / max_size();
+    }
+
+    /* Returns required number of bytes for N line segments */
+    static __forceinline size_t bytes(size_t N)
+    {
+      return blocks(N) * sizeof(PointMi);
+    }
+
+   public:
+    /* Default constructor */
+    __forceinline PointMi() {}
+
+    /* Construction from vertices and IDs */
+    __forceinline PointMi(const vuint<M>& geomIDs, const vuint<M>& primIDs, Geometry::GType gtype, uint32_t numPrimitives)
+        : gtype((unsigned char)gtype),
+          numPrimitives(numPrimitives),
+          sharedGeomID(geomIDs[0]),
+          primIDs(primIDs)
+    {
+      assert(all(vuint<M>(geomID()) == geomIDs));
+    }
+
+    /* Returns a mask that tells which line segments are valid */
+    __forceinline vbool<M> valid() const {
+      return vint<M>(step) < vint<M>(numPrimitives);
+    }
+
+    /* Returns if the specified line segment is valid */
+    __forceinline bool valid(const size_t i) const
+    {
+      assert(i < M);
+      return i < numPrimitives;
+    }
+
+    /* Returns the number of stored line segments */
+    __forceinline size_t size() const {
+      return numPrimitives;
+    }
+
+    __forceinline unsigned int geomID(unsigned int i = 0) const {
+      return sharedGeomID;
+    }
+
+    __forceinline vuint<M>& primID() {
+      return primIDs;
+    }
+    __forceinline const vuint<M>& primID() const {
+      return primIDs;
+    }
+    __forceinline unsigned int primID(const size_t i) const {
+      assert(i < M);
+      return primIDs[i];
+    }
+
+    /* gather the line segments */
+    __forceinline void gather(Vec4vf<M>& p0, const Points* geom) const;
+    __forceinline void gather(Vec4vf<M>& p0, Vec3vf<M>& n0, const Points* geom) const;
+
+    __forceinline void gatheri(Vec4vf<M>& p0, const Points* geom, const int itime) const;
+    __forceinline void gatheri(Vec4vf<M>& p0, Vec3vf<M>& n0, const Points* geom, const int itime) const;
+
+    __forceinline void gather(Vec4vf<M>& p0, const Points* geom, float time) const;
+    __forceinline void gather(Vec4vf<M>& p0, Vec3vf<M>& n0, const Points* geom, float time) const;
+
+    /* Calculate the bounds of the line segments */
+    __forceinline const BBox3fa bounds(const Scene* scene, size_t itime = 0) const
+    {
+      BBox3fa bounds = empty;
+      for (size_t i = 0; i < M && valid(i); i++) {
+        const Points* geom = scene->get<Points>(geomID(i));
+        bounds.extend(geom->bounds(primID(i),itime));
+      }
+      return bounds;
+    }
+
+    /* Calculate the linear bounds of the primitive */
+    __forceinline LBBox3fa linearBounds(const Scene* scene, size_t itime) {
+      return LBBox3fa(bounds(scene, itime + 0), bounds(scene, itime + 1));
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene* const scene, size_t itime, size_t numTimeSteps)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i = 0; i < M && valid(i); i++) {
+        const Points* geom = scene->get<Points>(geomID(i));
+        allBounds.extend(geom->linearBounds(primID(i), itime, numTimeSteps));
+      }
+      return allBounds;
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene* const scene, const BBox1f time_range)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i = 0; i < M && valid(i); i++) {
+        const Points* geom = scene->get<Points>(geomID((unsigned int)i));
+        allBounds.extend(geom->linearBounds(primID(i), time_range));
+      }
+      return allBounds;
+    }
+
+    /* Fill line segment from line segment list */
+    template<typename PrimRefT>
+    __forceinline void fill(const PrimRefT* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      Geometry::GType gty = scene->get(prims[begin].geomID())->getType();
+      vuint<M> geomID, primID;
+      vuint<M> v0;
+      const PrimRefT* prim = &prims[begin];
+
+      int numPrimitives = 0;
+      for (size_t i = 0; i < M; i++) {
+        if (begin < end) {
+          geomID[i] = prim->geomID();
+          primID[i] = prim->primID();
+          begin++;
+          numPrimitives++;
+        } else {
+          assert(i);
+          if (i > 0) {
+            geomID[i] = geomID[i - 1];
+            primID[i] = primID[i - 1];
+          }
+        }
+        if (begin < end)
+          prim = &prims[begin];  // FIXME: remove this line
+      }
+      new (this) PointMi(geomID, primID, gty, numPrimitives);  // FIXME: use non temporal store
+    }
+
+    template<typename BVH, typename Allocator>
+    __forceinline static typename BVH::NodeRef createLeaf(BVH* bvh,
+                                                          const PrimRef* prims,
+                                                          const range<size_t>& set,
+                                                          const Allocator& alloc)
+    {
+      size_t start    = set.begin();
+      size_t items    = PointMi::blocks(set.size());
+      size_t numbytes = PointMi::bytes(set.size());
+      PointMi* accel  = (PointMi*)alloc.malloc1(numbytes, M * sizeof(float));
+      for (size_t i = 0; i < items; i++) {
+        accel[i].fill(prims, start, set.end(), bvh->scene);
+      }
+      return bvh->encodeLeaf((char*)accel, items);
+    };
+
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& begin, size_t end, Scene* scene, size_t itime)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, itime);
+    }
+
+    __forceinline LBBox3fa fillMB(
+        const PrimRefMB* prims, size_t& begin, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, time_range);
+    }
+
+    template<typename BVH, typename SetMB, typename Allocator>
+    __forceinline static typename BVH::NodeRecordMB4D createLeafMB(BVH* bvh, const SetMB& prims, const Allocator& alloc)
+    {
+      size_t start                     = prims.object_range.begin();
+      size_t end                       = prims.object_range.end();
+      size_t items                     = PointMi::blocks(prims.object_range.size());
+      size_t numbytes                  = PointMi::bytes(prims.object_range.size());
+      PointMi* accel                   = (PointMi*)alloc.malloc1(numbytes, M * sizeof(float));
+      const typename BVH::NodeRef node = bvh->encodeLeaf((char*)accel, items);
+
+      LBBox3fa bounds = empty;
+      for (size_t i = 0; i < items; i++)
+        bounds.extend(accel[i].fillMB(prims.prims->data(), start, end, bvh->scene, prims.time_range));
+
+      return typename BVH::NodeRecordMB4D(node, bounds, prims.time_range);
+    };
+
+    /*! output operator */
+    friend __forceinline embree_ostream operator<<(embree_ostream cout, const PointMi& point)
+    {
+      return cout << "Point" << M << "i {" << point.geomID() << ", " << point.primID() << "}";
+    }
+
+   public:
+    unsigned char gtype;
+    unsigned char numPrimitives;
+    unsigned int sharedGeomID;
+
+   private:
+    vuint<M> primIDs;  // primitive ID
+  };
+
+  template<>
+  __forceinline void PointMi<4>::gather(Vec4vf4& p0, const Points* geom) const
+  {
+    const vfloat4 a0   = vfloat4::loadu(geom->vertexPtr(primID(0)));
+    const vfloat4 a1   = vfloat4::loadu(geom->vertexPtr(primID(1)));
+    const vfloat4 a2   = vfloat4::loadu(geom->vertexPtr(primID(2)));
+    const vfloat4 a3   = vfloat4::loadu(geom->vertexPtr(primID(3)));
+    transpose(a0, a1, a2, a3, p0.x, p0.y, p0.z, p0.w);
+  }
+
+  template<>
+  __forceinline void PointMi<4>::gather(Vec4vf4& p0, Vec3vf4& n0, const Points* geom) const
+  {
+    const vfloat4 a0   = vfloat4::loadu(geom->vertexPtr(primID(0)));
+    const vfloat4 a1   = vfloat4::loadu(geom->vertexPtr(primID(1)));
+    const vfloat4 a2   = vfloat4::loadu(geom->vertexPtr(primID(2)));
+    const vfloat4 a3   = vfloat4::loadu(geom->vertexPtr(primID(3)));
+    transpose(a0, a1, a2, a3, p0.x, p0.y, p0.z, p0.w);
+    const vfloat4 b0 = vfloat4(geom->normal(primID(0)));
+    const vfloat4 b1 = vfloat4(geom->normal(primID(1)));
+    const vfloat4 b2 = vfloat4(geom->normal(primID(2)));
+    const vfloat4 b3 = vfloat4(geom->normal(primID(3)));
+    transpose(b0, b1, b2, b3, n0.x, n0.y, n0.z);
+  }
+
+  template<>
+  __forceinline void PointMi<4>::gatheri(Vec4vf4& p0, const Points* geom, const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0), itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1), itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2), itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3), itime));
+    transpose(a0, a1, a2, a3, p0.x, p0.y, p0.z, p0.w);
+  }
+
+  template<>
+  __forceinline void PointMi<4>::gatheri(Vec4vf4& p0, Vec3vf4& n0, const Points* geom, const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0), itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1), itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2), itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3), itime));
+    transpose(a0, a1, a2, a3, p0.x, p0.y, p0.z, p0.w);
+    const vfloat4 b0 = vfloat4(geom->normal((size_t)primID(0), (size_t)itime));
+    const vfloat4 b1 = vfloat4(geom->normal((size_t)primID(1), (size_t)itime));
+    const vfloat4 b2 = vfloat4(geom->normal((size_t)primID(2), (size_t)itime));
+    const vfloat4 b3 = vfloat4(geom->normal((size_t)primID(3), (size_t)itime));
+    transpose(b0, b1, b2, b3, n0.x, n0.y, n0.z);
+  }
+
+  template<>
+  __forceinline void PointMi<4>::gather(Vec4vf4& p0, const Points* geom, float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf4 a0; gatheri(a0, geom, itime);
+    Vec4vf4 b0; gatheri(b0, geom, itime + 1);
+    p0 = lerp(a0, b0, vfloat4(ftime));
+  }
+
+  template<>
+  __forceinline void PointMi<4>::gather(Vec4vf4& p0, Vec3vf4& n0, const Points* geom, float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf4 a0, b0;
+    Vec3vf4 norm0, norm1;
+    gatheri(a0, norm0, geom, itime);
+    gatheri(b0, norm1, geom, itime + 1);
+    p0 = lerp(a0, b0, vfloat4(ftime));
+    n0 = lerp(norm0, norm1, vfloat4(ftime));
+  }
+
+#if defined(__AVX__)
+
+  template<>
+  __forceinline void PointMi<8>::gather(Vec4vf8& p0, const Points* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0)));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1)));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2)));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3)));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(primID(4)));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(primID(5)));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(primID(6)));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(primID(7)));
+    transpose(a0, a1, a2, a3, a4, a5, a6, a7, p0.x, p0.y, p0.z, p0.w);
+  }
+
+  template<>
+  __forceinline void PointMi<8>::gather(Vec4vf8& p0, Vec3vf8& n0, const Points* geom) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0)));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1)));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2)));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3)));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(primID(4)));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(primID(5)));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(primID(6)));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(primID(7)));
+    transpose(a0, a1, a2, a3, a4, a5, a6, a7, p0.x, p0.y, p0.z, p0.w);
+    const vfloat4 b0 = vfloat4(geom->normal(primID(0)));
+    const vfloat4 b1 = vfloat4(geom->normal(primID(1)));
+    const vfloat4 b2 = vfloat4(geom->normal(primID(2)));
+    const vfloat4 b3 = vfloat4(geom->normal(primID(3)));
+    const vfloat4 b4 = vfloat4(geom->normal(primID(4)));
+    const vfloat4 b5 = vfloat4(geom->normal(primID(5)));
+    const vfloat4 b6 = vfloat4(geom->normal(primID(6)));
+    const vfloat4 b7 = vfloat4(geom->normal(primID(7)));
+    transpose(b0, b1, b2, b3, b4, b5, b6, b7, n0.x, n0.y, n0.z);
+  }
+
+  template<>
+  __forceinline void PointMi<8>::gatheri(Vec4vf8& p0, const Points* geom, const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0), itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1), itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2), itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3), itime));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(primID(4), itime));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(primID(5), itime));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(primID(6), itime));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(primID(7), itime));
+    transpose(a0, a1, a2, a3, a4, a5, a6, a7, p0.x, p0.y, p0.z, p0.w);
+  }
+
+  template<>
+  __forceinline void PointMi<8>::gatheri(Vec4vf8& p0, Vec3vf8& n0, const Points* geom, const int itime) const
+  {
+    const vfloat4 a0 = vfloat4::loadu(geom->vertexPtr(primID(0), itime));
+    const vfloat4 a1 = vfloat4::loadu(geom->vertexPtr(primID(1), itime));
+    const vfloat4 a2 = vfloat4::loadu(geom->vertexPtr(primID(2), itime));
+    const vfloat4 a3 = vfloat4::loadu(geom->vertexPtr(primID(3), itime));
+    const vfloat4 a4 = vfloat4::loadu(geom->vertexPtr(primID(4), itime));
+    const vfloat4 a5 = vfloat4::loadu(geom->vertexPtr(primID(5), itime));
+    const vfloat4 a6 = vfloat4::loadu(geom->vertexPtr(primID(6), itime));
+    const vfloat4 a7 = vfloat4::loadu(geom->vertexPtr(primID(7), itime));
+    transpose(a0, a1, a2, a3, a4, a5, a6, a7, p0.x, p0.y, p0.z, p0.w);
+    const vfloat4 b0 = vfloat4(geom->normal((size_t)primID(0), (size_t)itime));
+    const vfloat4 b1 = vfloat4(geom->normal((size_t)primID(1), (size_t)itime));
+    const vfloat4 b2 = vfloat4(geom->normal((size_t)primID(2), (size_t)itime));
+    const vfloat4 b3 = vfloat4(geom->normal((size_t)primID(3), (size_t)itime));
+    const vfloat4 b4 = vfloat4(geom->normal((size_t)primID(4), (size_t)itime));
+    const vfloat4 b5 = vfloat4(geom->normal((size_t)primID(5), (size_t)itime));
+    const vfloat4 b6 = vfloat4(geom->normal((size_t)primID(6), (size_t)itime));
+    const vfloat4 b7 = vfloat4(geom->normal((size_t)primID(7), (size_t)itime));
+    transpose(b0, b1, b2, b3, b4, b5, b6, b7, n0.x, n0.y, n0.z);
+  }
+
+  template<>
+  __forceinline void PointMi<8>::gather(Vec4vf8& p0, const Points* geom, float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf8 a0;
+    gatheri(a0, geom, itime);
+    Vec4vf8 b0;
+    gatheri(b0, geom, itime + 1);
+    p0 = lerp(a0, b0, vfloat8(ftime));
+  }
+
+  template<>
+  __forceinline void PointMi<8>::gather(Vec4vf8& p0, Vec3vf8& n0, const Points* geom, float time) const
+  {
+    float ftime;
+    const int itime = geom->timeSegment(time, ftime);
+
+    Vec4vf8 a0, b0;
+    Vec3vf8 norm0, norm1;
+    gatheri(a0, norm0, geom, itime);
+    gatheri(b0, norm1, geom, itime + 1);
+    p0 = lerp(a0, b0, vfloat8(ftime));
+    n0 = lerp(norm0, norm1, vfloat8(ftime));
+  }
+#endif
+
+  template<int M>
+  typename PointMi<M>::Type PointMi<M>::type;
+
+  typedef PointMi<4> Point4i;
+  typedef PointMi<8> Point8i;
+  
+}  // namespace embree

thirdparty/embree/kernels/geometry/primitive.h

@@ -0,0 +1,49 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/default.h"
+#include "../common/scene.h"
+#include "../../common/simd/simd.h"
+#include "../builders/primref.h"
+#include "../builders/primref_mb.h"
+
+namespace embree
+{
+  struct PrimitiveType
+  {
+    /*! returns name of this primitive type */
+    virtual const char* name() const = 0;
+    
+    /*! Returns the number of stored active primitives in a block. */
+    virtual size_t sizeActive(const char* This) const = 0;
+
+    /*! Returns the number of stored active and inactive primitives in a block. */
+    virtual size_t sizeTotal(const char* This) const = 0;
+
+    /*! Returns the number of bytes of block. */
+    virtual size_t getBytes(const char* This) const = 0;
+  };
+  
+  template<typename Primitive>
+  struct PrimitivePointQuery1
+  {
+    static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& prim)
+    {
+      bool changed = false;
+      for (size_t i = 0; i < Primitive::max_size(); i++)
+      {
+        if (!prim.valid(i)) break;
+        STAT3(point_query.trav_prims,1,1,1);
+        AccelSet* accel = (AccelSet*)context->scene->get(prim.geomID(i));
+        context->geomID = prim.geomID(i);
+        context->primID = prim.primID(i);
+        changed |= accel->pointQuery(query, context);
+      }
+      return changed;
+    }
+    
+    static __forceinline void pointQueryNoop(PointQuery* query, PointQueryContext* context, const Primitive& prim) { }
+  };
+}

thirdparty/embree/kernels/geometry/primitive4.cpp

@@ -0,0 +1,400 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "primitive.h"
+#include "curveNv.h"
+#include "curveNi.h"
+#include "curveNi_mb.h"
+#include "linei.h"
+#include "triangle.h"
+#include "trianglev.h"
+#include "trianglev_mb.h"
+#include "trianglei.h"
+#include "quadv.h"
+#include "quadi.h"
+#include "subdivpatch1.h"
+#include "object.h"
+#include "instance.h"
+#include "instance_array.h"
+#include "subgrid.h"
+
+namespace embree
+{
+  /********************** Curve4v **************************/
+
+  template<>
+  const char* Curve4v::Type::name () const {
+    return "curve4v";
+  }
+
+  template<>
+  size_t Curve4v::Type::sizeActive(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return ((Line4i*)This)->size();
+    else
+      return ((Curve4v*)This)->N;
+  }
+
+  template<>
+  size_t Curve4v::Type::sizeTotal(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return 4;
+    else
+      return ((Curve4v*)This)->N;
+  }
+
+  template<>
+  size_t Curve4v::Type::getBytes(const char* This) const
+  {
+     if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return Line4i::bytes(sizeActive(This));
+     else
+        return Curve4v::bytes(sizeActive(This));
+  }
+
+  /********************** Curve4i **************************/
+
+  template<>
+  const char* Curve4i::Type::name () const {
+    return "curve4i";
+  }
+
+  template<>
+  size_t Curve4i::Type::sizeActive(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return ((Line4i*)This)->size();
+    else
+      return ((Curve4i*)This)->N;
+  }
+
+  template<>
+  size_t Curve4i::Type::sizeTotal(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return 4;
+    else
+      return ((Curve4i*)This)->N;
+  }
+
+  template<>
+  size_t Curve4i::Type::getBytes(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return Line4i::bytes(sizeActive(This));
+    else
+      return Curve4i::bytes(sizeActive(This));
+  }
+
+  /********************** Curve4iMB **************************/
+
+  template<>
+  const char* Curve4iMB::Type::name () const {
+    return "curve4imb";
+  }
+
+  template<>
+  size_t Curve4iMB::Type::sizeActive(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return ((Line4i*)This)->size();
+    else
+      return ((Curve4iMB*)This)->N;
+  }
+
+  template<>
+  size_t Curve4iMB::Type::sizeTotal(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return 4;
+    else
+      return ((Curve4iMB*)This)->N;
+  }
+
+  template<>
+  size_t Curve4iMB::Type::getBytes(const char* This) const
+  {
+    if ((*This & Geometry::GType::GTY_BASIS_MASK) == Geometry::GType::GTY_BASIS_LINEAR)
+      return Line4i::bytes(sizeActive(This));
+    else
+      return Curve4iMB::bytes(sizeActive(This));
+  }
+
+  /********************** Line4i **************************/
+
+  template<>
+  const char* Line4i::Type::name () const {
+    return "line4i";
+  }
+
+  template<>
+  size_t Line4i::Type::sizeActive(const char* This) const {
+    return ((Line4i*)This)->size();
+  }
+
+  template<>
+  size_t Line4i::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Line4i::Type::getBytes(const char* This) const {
+    return sizeof(Line4i);
+  }
+
+  /********************** Triangle4 **************************/
+
+  template<>
+  const char* Triangle4::Type::name () const {
+    return "triangle4";
+  }
+
+  template<>
+  size_t Triangle4::Type::sizeActive(const char* This) const {
+    return ((Triangle4*)This)->size();
+  }
+
+  template<>
+  size_t Triangle4::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Triangle4::Type::getBytes(const char* This) const {
+    return sizeof(Triangle4);
+  }
+
+  /********************** Triangle4v **************************/
+
+  template<>
+  const char* Triangle4v::Type::name () const {
+    return "triangle4v";
+  }
+
+  template<>
+  size_t Triangle4v::Type::sizeActive(const char* This) const {
+    return ((Triangle4v*)This)->size();
+  }
+
+  template<>
+  size_t Triangle4v::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Triangle4v::Type::getBytes(const char* This) const {
+    return sizeof(Triangle4v);
+  }
+
+  /********************** Triangle4i **************************/
+
+  template<>
+  const char* Triangle4i::Type::name () const {
+    return "triangle4i";
+  }
+
+  template<>
+  size_t Triangle4i::Type::sizeActive(const char* This) const {
+    return ((Triangle4i*)This)->size();
+  }
+
+  template<>
+  size_t Triangle4i::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Triangle4i::Type::getBytes(const char* This) const {
+    return sizeof(Triangle4i);
+  }
+
+  /********************** Triangle4vMB **************************/
+
+  template<>
+  const char* Triangle4vMB::Type::name () const {
+    return  "triangle4vmb";
+  }
+
+  template<>
+  size_t Triangle4vMB::Type::sizeActive(const char* This) const {
+    return ((Triangle4vMB*)This)->size();
+  }
+
+  template<>
+  size_t Triangle4vMB::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Triangle4vMB::Type::getBytes(const char* This) const {
+    return sizeof(Triangle4vMB);
+  }
+
+  /********************** Quad4v **************************/
+
+  template<>
+  const char* Quad4v::Type::name () const {
+    return "quad4v";
+  }
+
+  template<>
+  size_t Quad4v::Type::sizeActive(const char* This) const {
+    return ((Quad4v*)This)->size();
+  }
+
+  template<>
+  size_t Quad4v::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Quad4v::Type::getBytes(const char* This) const {
+    return sizeof(Quad4v);
+  }
+
+  /********************** Quad4i **************************/
+
+  template<>
+  const char* Quad4i::Type::name () const {
+    return "quad4i";
+  }
+
+  template<>
+  size_t Quad4i::Type::sizeActive(const char* This) const {
+    return ((Quad4i*)This)->size();
+  }
+
+  template<>
+  size_t Quad4i::Type::sizeTotal(const char* This) const {
+    return 4;
+  }
+
+  template<>
+  size_t Quad4i::Type::getBytes(const char* This) const {
+    return sizeof(Quad4i);
+  }
+
+  /********************** SubdivPatch1 **************************/
+
+  const char* SubdivPatch1::Type::name () const {
+    return "subdivpatch1";
+  }
+
+  size_t SubdivPatch1::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  size_t SubdivPatch1::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  size_t SubdivPatch1::Type::getBytes(const char* This) const {
+    return sizeof(SubdivPatch1);
+  }
+
+  SubdivPatch1::Type SubdivPatch1::type;
+
+  /********************** Virtual Object **************************/
+
+  const char* Object::Type::name () const {
+    return "object";
+  }
+
+  size_t Object::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  size_t Object::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  size_t Object::Type::getBytes(const char* This) const {
+    return sizeof(Object);
+  }
+
+  Object::Type Object::type;
+
+  /********************** Instance **************************/
+
+  const char* InstancePrimitive::Type::name () const {
+    return "instance";
+  }
+
+  size_t InstancePrimitive::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  size_t InstancePrimitive::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  size_t InstancePrimitive::Type::getBytes(const char* This) const {
+    return sizeof(InstancePrimitive);
+  }
+
+  InstancePrimitive::Type InstancePrimitive::type;
+
+  /********************** InstanceArray4 **************************/
+
+  const char* InstanceArrayPrimitive::Type::name () const {
+    return "instance_array";
+  }
+
+  size_t InstanceArrayPrimitive::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  size_t InstanceArrayPrimitive::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  size_t InstanceArrayPrimitive::Type::getBytes(const char* This) const {
+    return sizeof(InstanceArrayPrimitive);
+  }
+
+  InstanceArrayPrimitive::Type InstanceArrayPrimitive::type;
+
+  /********************** SubGrid **************************/
+
+  const char* SubGrid::Type::name () const {
+    return "subgrid";
+  }
+
+  size_t SubGrid::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  size_t SubGrid::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  size_t SubGrid::Type::getBytes(const char* This) const {
+    return sizeof(SubGrid);
+  }
+
+  SubGrid::Type SubGrid::type;
+  
+  /********************** SubGridQBVH4 **************************/
+
+  template<>
+  const char* SubGridQBVH4::Type::name () const {
+    return "SubGridQBVH4";
+  }
+
+  template<>
+  size_t SubGridQBVH4::Type::sizeActive(const char* This) const {
+    return 1;
+  }
+
+  template<>
+  size_t SubGridQBVH4::Type::sizeTotal(const char* This) const {
+    return 1;
+  }
+
+  template<>
+  size_t SubGridQBVH4::Type::getBytes(const char* This) const {
+    return sizeof(SubGridQBVH4);
+  }
+}

thirdparty/embree/kernels/geometry/quad_intersector.h

@@ -0,0 +1,76 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects a ray with a quad with backface culling
+     *  enabled. The quad v0,v1,v2,v3 is split into two triangles
+     *  v0,v1,v3 and v2,v3,v1. The edge v1,v2 decides which of the two
+     *  triangles gets intersected. */
+    template<int N>
+    __forceinline vbool<N> intersect_quad_backface_culling(const vbool<N>& valid0,
+                                                           const Vec3fa& ray_org,
+                                                           const Vec3fa& ray_dir,
+                                                           const float ray_tnear,
+                                                           const float ray_tfar,
+                                                           const Vec3vf<N>& quad_v0,
+                                                           const Vec3vf<N>& quad_v1,
+                                                           const Vec3vf<N>& quad_v2,
+                                                           const Vec3vf<N>& quad_v3,
+                                                           vfloat<N>& u_o,
+                                                           vfloat<N>& v_o,
+                                                           vfloat<N>& t_o)
+    {
+      /* calculate vertices relative to ray origin */
+      vbool<N> valid = valid0;
+      const Vec3vf<N> O = Vec3vf<N>(ray_org);
+      const Vec3vf<N> D = Vec3vf<N>(ray_dir);
+      const Vec3vf<N> va = quad_v0-O;
+      const Vec3vf<N> vb = quad_v1-O;
+      const Vec3vf<N> vc = quad_v2-O;
+      const Vec3vf<N> vd = quad_v3-O;
+
+      const Vec3vf<N> edb = vb-vd;
+      const vfloat<N> WW = dot(cross(vd,edb),D);
+      const Vec3vf<N> v0 = select(WW <= 0.0f,va,vc);
+      const Vec3vf<N> v1 = select(WW <= 0.0f,vb,vd);
+      const Vec3vf<N> v2 = select(WW <= 0.0f,vd,vb);
+
+      /* calculate edges */
+      const Vec3vf<N> e0 = v2-v0;
+      const Vec3vf<N> e1 = v0-v1;
+
+      /* perform edge tests */
+      const vfloat<N> U = dot(cross(v0,e0),D);
+      const vfloat<N> V = dot(cross(v1,e1),D);
+      valid &= max(U,V) <= 0.0f;
+      if (unlikely(none(valid))) return false;
+
+      /* calculate geometry normal and denominator */
+      const Vec3vf<N> Ng = cross(e1,e0);
+      const vfloat<N> den = dot(Ng,D);
+      const vfloat<N> rcpDen = rcp(den);
+
+      /* perform depth test */
+      const vfloat<N> t = rcpDen*dot(v0,Ng);
+      valid &= vfloat<N>(ray_tnear) <= t & t <= vfloat<N>(ray_tfar);
+      if (unlikely(none(valid))) return false;
+
+      /* avoid division by 0 */
+      valid &= den != vfloat<N>(zero);
+      if (unlikely(none(valid))) return false;
+
+      /* update hit information */
+      t_o = t;
+      u_o = U * rcpDen;
+      v_o = V * rcpDen;
+      u_o = select(WW <= 0.0f,u_o,1.0f-u_o);
+      v_o = select(WW <= 0.0f,v_o,1.0f-v_o);
+      return valid;
+    }
+  }
+}

thirdparty/embree/kernels/geometry/quad_intersector_moeller.h

@@ -0,0 +1,460 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "quadv.h"
+#include "triangle_intersector_moeller.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+    struct QuadHitM
+    {
+      __forceinline QuadHitM() {}
+
+      __forceinline QuadHitM(const vbool<M>& valid,
+                             const vfloat<M>& U,
+                             const vfloat<M>& V,
+                             const vfloat<M>& T,
+                             const vfloat<M>& absDen,
+                             const Vec3vf<M>& Ng,
+                             const vbool<M>& flags)
+        : U(U), V(V), T(T), absDen(absDen), tri_Ng(Ng), valid(valid), flags(flags) {}
+
+      __forceinline void finalize()
+      {
+        const vfloat<M> rcpAbsDen = rcp(absDen);
+        vt = T * rcpAbsDen;
+        const vfloat<M> u = min(U * rcpAbsDen,1.0f);
+        const vfloat<M> v = min(V * rcpAbsDen,1.0f);
+        const vfloat<M> u1 = vfloat<M>(1.0f) - u;
+        const vfloat<M> v1 = vfloat<M>(1.0f) - v;
+#if !defined(__AVX__) || defined(EMBREE_BACKFACE_CULLING)
+        vu = select(flags,u1,u);
+        vv = select(flags,v1,v);
+        vNg = Vec3vf<M>(tri_Ng.x,tri_Ng.y,tri_Ng.z);
+#else
+        const vfloat<M> flip = select(flags,vfloat<M>(-1.0f),vfloat<M>(1.0f));
+        vv = select(flags,u1,v);
+        vu = select(flags,v1,u);
+        vNg = Vec3vf<M>(flip*tri_Ng.x,flip*tri_Ng.y,flip*tri_Ng.z);
+#endif
+      }
+
+      __forceinline Vec2f uv(const size_t i)
+      {
+        const float u = vu[i];
+        const float v = vv[i];
+        return Vec2f(u,v);
+      }
+
+      __forceinline float   t(const size_t i) { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+
+    private:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> T;
+      vfloat<M> absDen;
+      Vec3vf<M> tri_Ng;
+
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+
+    public:
+      const vbool<M> flags;
+    };
+
+    template<int K>
+    struct QuadHitK
+    {
+      __forceinline QuadHitK(const vfloat<K>& U,
+                             const vfloat<K>& V,
+                             const vfloat<K>& T,
+                             const vfloat<K>& absDen,
+                             const Vec3vf<K>& Ng,
+                             const vbool<K>& flags)
+        : U(U), V(V), T(T), absDen(absDen), flags(flags), tri_Ng(Ng) {}
+
+      __forceinline std::tuple<vfloat<K>,vfloat<K>,vfloat<K>,Vec3vf<K>> operator() () const
+      {
+        const vfloat<K> rcpAbsDen = rcp(absDen);
+        const vfloat<K> t = T * rcpAbsDen;
+        const vfloat<K> u0 = min(U * rcpAbsDen,1.0f);
+        const vfloat<K> v0 = min(V * rcpAbsDen,1.0f);
+        const vfloat<K> u1 = vfloat<K>(1.0f) - u0;
+        const vfloat<K> v1 = vfloat<K>(1.0f) - v0;
+        const vfloat<K> u = select(flags,u1,u0);
+        const vfloat<K> v = select(flags,v1,v0);
+        const Vec3vf<K> Ng(tri_Ng.x,tri_Ng.y,tri_Ng.z);
+        return std::make_tuple(u,v,t,Ng);
+      }
+
+    private:
+      const vfloat<K> U;
+      const vfloat<K> V;
+      const vfloat<K> T;
+      const vfloat<K> absDen;
+      const vbool<K> flags;
+      const Vec3vf<K> tri_Ng;
+    };
+
+    /* ----------------------------- */
+    /* -- single ray intersectors -- */
+    /* ----------------------------- */
+
+
+    template<int M, bool filter>
+    struct QuadMIntersector1MoellerTrumbore;
+
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMIntersector1MoellerTrumbore
+    {
+      __forceinline QuadMIntersector1MoellerTrumbore() {}
+
+      __forceinline QuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {}
+
+      __forceinline void intersect(RayHit& ray, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                   const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        UVIdentity<M> mapUV;
+        MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+        MoellerTrumboreIntersector1<M> intersector(ray,nullptr);
+        Intersect1EpilogM<M,filter> epilog(ray,context,geomID,primID);
+
+        /* intersect first triangle */
+        if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+          epilog(hit.valid,hit);
+
+        /* intersect second triangle */
+        if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+        {
+          hit.U = hit.absDen - hit.U;
+          hit.V = hit.absDen - hit.V;
+          epilog(hit.valid,hit);
+        }
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                  const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                  const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        UVIdentity<M> mapUV;
+        MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+        MoellerTrumboreIntersector1<M> intersector(ray,nullptr);
+        Occluded1EpilogM<M,filter> epilog(ray,context,geomID,primID);
+
+        /* intersect first triangle */
+        if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+        {
+          if (epilog(hit.valid,hit))
+            return true;
+        }
+
+        /* intersect second triangle */
+        if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+        {
+          hit.U = hit.absDen - hit.U;
+          hit.V = hit.absDen - hit.V;
+          if (epilog(hit.valid,hit))
+            return true;
+        }
+        return false;
+      }
+    };
+
+#if defined(__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<bool filter>
+    struct QuadMIntersector1MoellerTrumbore<4,filter>
+    {
+      __forceinline QuadMIntersector1MoellerTrumbore() {}
+
+      __forceinline QuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {}
+      
+      template<typename Epilog>
+      __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
+        UVIdentity<8> mapUV;
+        MoellerTrumboreHitM<8,UVIdentity<8>> hit(mapUV);
+        MoellerTrumboreIntersector1<8> intersector(ray,nullptr);
+        const vbool8 flags(0,0,0,0,1,1,1,1);
+        if (unlikely(intersector.intersect(ray,vtx0,vtx1,vtx2,mapUV,hit)))
+        {
+          vfloat8 U = hit.U, V = hit.V, absDen = hit.absDen;
+
+#if !defined(EMBREE_BACKFACE_CULLING)
+          hit.U = select(flags,absDen-V,U);
+          hit.V = select(flags,absDen-U,V);
+          hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); // FIXME: use XOR
+#else
+          hit.U = select(flags,absDen-U,U);
+          hit.V = select(flags,absDen-V,V);
+#endif
+          if (unlikely(epilog(hit.valid,hit)))
+            return true;
+        }
+        return false;
+      }
+      
+      __forceinline bool intersect(RayHit& ray, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect(ray,v0,v1,v2,v3,Intersect1EpilogM<8,filter>(ray,context,vuint8(geomID),vuint8(primID)));
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                  const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                  const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect(ray,v0,v1,v2,v3,Occluded1EpilogM<8,filter>(ray,context,vuint8(geomID),vuint8(primID)));
+      }
+    };
+
+#endif
+
+    /* ----------------------------- */
+    /* -- ray packet intersectors -- */
+    /* ----------------------------- */
+
+
+    struct MoellerTrumboreIntersector1KTriangleM
+    {
+      /*! Intersect k'th ray from ray packet of size K with M triangles. */
+      template<int M, int K, typename Epilog>
+      static  __forceinline bool intersect(RayK<K>& ray,
+                                           size_t k,
+                                           const Vec3vf<M>& tri_v0,
+                                           const Vec3vf<M>& tri_e1,
+                                           const Vec3vf<M>& tri_e2,
+                                           const Vec3vf<M>& tri_Ng,
+                                           const vbool<M>& flags,
+                                           const Epilog& epilog)
+      {
+        /* calculate denominator */
+        const Vec3vf<M> O = broadcast<vfloat<M>>(ray.org,k);
+        const Vec3vf<M> D = broadcast<vfloat<M>>(ray.dir,k);
+        const Vec3vf<M> C = Vec3vf<M>(tri_v0) - O;
+        const Vec3vf<M> R = cross(C,D);
+        const vfloat<M> den = dot(Vec3vf<M>(tri_Ng),D);
+        const vfloat<M> absDen = abs(den);
+        const vfloat<M> sgnDen = signmsk(den);
+        
+        /* perform edge tests */
+        const vfloat<M> U = dot(R,Vec3vf<M>(tri_e2)) ^ sgnDen;
+        const vfloat<M> V = dot(R,Vec3vf<M>(tri_e1)) ^ sgnDen;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        vbool<M> valid = (den < vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#else
+        vbool<M> valid = (den != vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#endif
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<M> T = dot(Vec3vf<M>(tri_Ng),C) ^ sgnDen;
+        valid &= (absDen*vfloat<M>(ray.tnear()[k]) < T) & (T <= absDen*vfloat<M>(ray.tfar[k]));
+        if (likely(none(valid))) return false;
+        
+        /* calculate hit information */
+        QuadHitM<M> hit(valid,U,V,T,absDen,tri_Ng,flags);
+        return epilog(valid,hit);
+      }
+      
+      template<int M, int K, typename Epilog>
+      static __forceinline bool intersect1(RayK<K>& ray,
+                                           size_t k,
+                                           const Vec3vf<M>& v0,
+                                           const Vec3vf<M>& v1,
+                                           const Vec3vf<M>& v2,
+                                           const vbool<M>& flags,
+                                           const Epilog& epilog)
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        const Vec3vf<M> Ng = cross(e2,e1);
+        return intersect<M,K>(ray,k,v0,e1,e2,Ng,flags,epilog);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct QuadMIntersectorKMoellerTrumboreBase
+    {
+      __forceinline QuadMIntersectorKMoellerTrumboreBase(const vbool<K>& valid, const RayK<K>& ray) {}
+            
+      /*! Intersects K rays with one of M triangles. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_e1,
+                                        const Vec3vf<K>& tri_e2,
+                                        const Vec3vf<K>& tri_Ng,
+                                        const vbool<K>& flags,
+                                        const Epilog& epilog) const
+      { 
+        /* calculate denominator */
+        vbool<K> valid = valid0;
+        const Vec3vf<K> C = tri_v0 - ray.org;
+        const Vec3vf<K> R = cross(C,ray.dir);
+        const vfloat<K> den = dot(tri_Ng,ray.dir);
+        const vfloat<K> absDen = abs(den);
+        const vfloat<K> sgnDen = signmsk(den);
+        
+        /* test against edge p2 p0 */
+        const vfloat<K> U = dot(R,tri_e2) ^ sgnDen;
+        valid &= U >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p0 p1 */
+        const vfloat<K> V = dot(R,tri_e1) ^ sgnDen;
+        valid &= V >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p1 p2 */
+        const vfloat<K> W = absDen-U-V;
+        valid &= W >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<K> T = dot(tri_Ng,C) ^ sgnDen;
+        valid &= (absDen*ray.tnear() < T) & (T <= absDen*ray.tfar);
+        if (unlikely(none(valid))) return false;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= den < vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#else
+        valid &= den != vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#endif
+        
+        /* calculate hit information */
+        QuadHitK<K> hit(U,V,T,absDen,tri_Ng,flags);
+        return epilog(valid,hit);
+      }
+      
+      /*! Intersects K rays with one of M quads. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0, 
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const vbool<K>& flags,
+                                        const Epilog& epilog) const
+      {
+        const Vec3vf<K> e1 = tri_v0-tri_v1;
+        const Vec3vf<K> e2 = tri_v2-tri_v0;
+        const Vec3vf<K> Ng = cross(e2,e1);
+        return intersectK(valid0,ray,tri_v0,e1,e2,Ng,flags,epilog);
+      }
+
+      /*! Intersects K rays with one of M quads. */
+      template<typename Epilog>
+      __forceinline bool intersectK(const vbool<K>& valid0, 
+                                    RayK<K>& ray,
+                                    const Vec3vf<K>& v0,
+                                    const Vec3vf<K>& v1,
+                                    const Vec3vf<K>& v2,
+                                    const Vec3vf<K>& v3,
+                                    const Epilog& epilog) const
+      {
+        intersectK(valid0,ray,v0,v1,v3,vbool<K>(false),epilog);
+        if (none(valid0)) return true;
+        intersectK(valid0,ray,v2,v3,v1,vbool<K>(true ),epilog);
+        return none(valid0);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct QuadMIntersectorKMoellerTrumbore : public QuadMIntersectorKMoellerTrumboreBase<M,K,filter>
+    {
+      __forceinline QuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray)
+        : QuadMIntersectorKMoellerTrumboreBase<M,K,filter>(valid,ray) {}
+
+      __forceinline void intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                    const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Intersect1KEpilogM<M,K,filter> epilog(ray,k,context,geomID,primID);
+        MoellerTrumboreIntersector1KTriangleM::intersect1<M,K>(ray,k,v0,v1,v3,vbool<M>(false),epilog);
+        MoellerTrumboreIntersector1KTriangleM::intersect1<M,K>(ray,k,v2,v3,v1,vbool<M>(true ),epilog);
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                   const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Occluded1KEpilogM<M,K,filter> epilog(ray,k,context,geomID,primID);
+        if (MoellerTrumboreIntersector1KTriangleM::intersect1<M,K>(ray,k,v0,v1,v3,vbool<M>(false),epilog)) return true;
+        if (MoellerTrumboreIntersector1KTriangleM::intersect1<M,K>(ray,k,v2,v3,v1,vbool<M>(true ),epilog)) return true;
+        return false;
+      }
+    };
+
+
+#if defined(__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<int K, bool filter>
+    struct QuadMIntersectorKMoellerTrumbore<4,K,filter> : public QuadMIntersectorKMoellerTrumboreBase<4,K,filter>
+    {
+      __forceinline QuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray)
+        : QuadMIntersectorKMoellerTrumboreBase<4,K,filter>(valid,ray) {}
+      
+      template<typename Epilog>
+      __forceinline bool intersect1(RayK<K>& 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));
+#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 MoellerTrumboreIntersector1KTriangleM::intersect1<8,K>(ray,k,vtx0,vtx1,vtx2,flags,epilog); 
+      }
+      
+      __forceinline bool intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                    const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,Intersect1KEpilogM<8,K,filter>(ray,k,context,vuint8(geomID),vuint8(primID)));
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,Occluded1KEpilogM<8,K,filter>(ray,k,context,vuint8(geomID),vuint8(primID)));
+      }
+    };
+
+#endif
+  }
+}

thirdparty/embree/kernels/geometry/quad_intersector_pluecker.h

@@ -0,0 +1,438 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#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
+  {
+    template<int M>
+    struct QuadHitPlueckerM
+    {
+      __forceinline QuadHitPlueckerM() {}
+
+      __forceinline QuadHitPlueckerM(const vbool<M>& valid,
+                                     const vfloat<M>& U,
+                                     const vfloat<M>& V,
+                                     const vfloat<M>& UVW,
+                                     const vfloat<M>& t,
+                                     const Vec3vf<M>& Ng,
+                                     const vbool<M>& flags)
+        : U(U), V(V), UVW(UVW), tri_Ng(Ng), valid(valid), vt(t), flags(flags) {}
+
+      __forceinline void finalize()
+      {
+        const vbool<M> invalid = abs(UVW) < min_rcp_input;
+        const vfloat<M> rcpUVW = select(invalid,vfloat<M>(0.0f),rcp(UVW));
+        const vfloat<M> u = min(U * rcpUVW,1.0f);
+        const vfloat<M> v = min(V * rcpUVW,1.0f);
+        const vfloat<M> u1 = vfloat<M>(1.0f) - u;
+        const vfloat<M> v1 = vfloat<M>(1.0f) - v;
+#if !defined(__AVX__) || defined(EMBREE_BACKFACE_CULLING)
+        vu = select(flags,u1,u);
+        vv = select(flags,v1,v);
+        vNg = Vec3vf<M>(tri_Ng.x,tri_Ng.y,tri_Ng.z);
+#else
+        const vfloat<M> flip = select(flags,vfloat<M>(-1.0f),vfloat<M>(1.0f));
+        vv = select(flags,u1,v);
+        vu = select(flags,v1,u);
+        vNg = Vec3vf<M>(flip*tri_Ng.x,flip*tri_Ng.y,flip*tri_Ng.z);
+#endif
+      }
+
+      __forceinline Vec2f uv(const size_t i)
+      {
+        const float u = vu[i];
+        const float v = vv[i];
+        return Vec2f(u,v);
+      }
+
+      __forceinline float   t(const size_t i) { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+
+    private:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> UVW;
+      Vec3vf<M> tri_Ng;
+
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+
+    public:
+      const vbool<M> flags;
+    };
+
+    template<int K>
+    struct QuadHitPlueckerK
+    {
+      __forceinline QuadHitPlueckerK(const vfloat<K>& U,
+                                     const vfloat<K>& V,
+                                     const vfloat<K>& UVW,
+                                     const vfloat<K>& t,
+                                     const Vec3vf<K>& Ng,
+                                     const vbool<K>& flags)
+        : U(U), V(V), UVW(UVW), t(t), flags(flags), tri_Ng(Ng) {}
+
+      __forceinline std::tuple<vfloat<K>,vfloat<K>,vfloat<K>,Vec3vf<K>> operator() () const
+      {
+        const vbool<K> invalid = abs(UVW) < min_rcp_input;
+        const vfloat<K> rcpUVW = select(invalid,vfloat<K>(0.0f),rcp(UVW));
+        const vfloat<K> u0 = min(U * rcpUVW,1.0f);
+        const vfloat<K> v0 = min(V * rcpUVW,1.0f);
+        const vfloat<K> u1 = vfloat<K>(1.0f) - u0;
+        const vfloat<K> v1 = vfloat<K>(1.0f) - v0;
+        const vfloat<K> u = select(flags,u1,u0);
+        const vfloat<K> v = select(flags,v1,v0);
+        const Vec3vf<K> Ng(tri_Ng.x,tri_Ng.y,tri_Ng.z);
+        return std::make_tuple(u,v,t,Ng);
+      }
+
+    private:
+      const vfloat<K> U;
+      const vfloat<K> V;
+      const vfloat<K> UVW;
+      const vfloat<K> t;
+      const vbool<K> flags;
+      const Vec3vf<K> tri_Ng;
+    };
+
+    struct PlueckerIntersectorTriangle1
+    {
+      template<int M, typename Epilog>
+      static __forceinline bool intersect(Ray& ray,
+                                          const Vec3vf<M>& tri_v0,
+                                          const Vec3vf<M>& tri_v1,
+                                          const Vec3vf<M>& tri_v2,
+                                          const vbool<M>& flags,
+                                          const Epilog& epilog)
+      {
+        /* calculate vertices relative to ray origin */
+        const Vec3vf<M> O = Vec3vf<M>((Vec3fa)ray.org);
+        const Vec3vf<M> D = Vec3vf<M>((Vec3fa)ray.dir);
+        const Vec3vf<M> v0 = tri_v0-O;
+        const Vec3vf<M> v1 = tri_v1-O;
+        const Vec3vf<M> v2 = tri_v2-O;
+
+        /* calculate triangle edges */
+        const Vec3vf<M> e0 = v2-v0;
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v1-v2;
+
+        /* perform edge tests */
+        const vfloat<M> U = dot(cross(e0,v2+v0),D);
+        const vfloat<M> V = dot(cross(e1,v0+v1),D);
+        const vfloat<M> W = dot(cross(e2,v1+v2),D);
+        const vfloat<M> UVW = U+V+W;
+        const vfloat<M> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+        vbool<M> valid = max(U,V,W) <= eps;
+#else
+        vbool<M> valid =  (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif
+        if (unlikely(none(valid))) return false;
+
+        /* calculate geometry normal and denominator */
+        const Vec3vf<M> Ng = stable_triangle_normal(e0,e1,e2);
+        const vfloat<M> den = twice(dot(Ng,D));
+
+         /* perform depth test */
+        const vfloat<M> T = twice(dot(v0,Ng));
+        const vfloat<M> t = rcp(den)*T;
+        valid &= vfloat<M>(ray.tnear()) <= t & t <= vfloat<M>(ray.tfar);
+        valid &= den != vfloat<M>(zero);
+        if (unlikely(none(valid))) return false;
+
+        /* update hit information */
+        QuadHitPlueckerM<M> hit(valid,U,V,UVW,t,Ng,flags);
+        return epilog(valid,hit);
+      }
+    };
+
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMIntersector1Pluecker
+    {
+      __forceinline QuadMIntersector1Pluecker() {}
+
+      __forceinline QuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {}
+
+      __forceinline void intersect(RayHit& ray, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                   const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Intersect1EpilogM<M,filter> epilog(ray,context,geomID,primID);
+        PlueckerIntersectorTriangle1::intersect<M>(ray,v0,v1,v3,vbool<M>(false),epilog);
+        PlueckerIntersectorTriangle1::intersect<M>(ray,v2,v3,v1,vbool<M>(true),epilog);
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                  const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                  const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Occluded1EpilogM<M,filter> epilog(ray,context,geomID,primID);
+        if (PlueckerIntersectorTriangle1::intersect<M>(ray,v0,v1,v3,vbool<M>(false),epilog)) return true;
+        if (PlueckerIntersectorTriangle1::intersect<M>(ray,v2,v3,v1,vbool<M>(true ),epilog)) return true;
+        return false;
+      }
+    };
+
+#if defined(__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<bool filter>
+    struct QuadMIntersector1Pluecker<4,filter>
+    {
+      __forceinline QuadMIntersector1Pluecker() {}
+
+      __forceinline QuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {}
+      
+      template<typename Epilog>
+      __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<8>(ray,vtx0,vtx1,vtx2,flags,epilog); 
+      }
+      
+      __forceinline bool intersect(RayHit& ray, RayQueryContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect(ray,v0,v1,v2,v3,Intersect1EpilogM<8,filter>(ray,context,vuint8(geomID),vuint8(primID)));
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context, const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3,
+                                  const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect(ray,v0,v1,v2,v3,Occluded1EpilogM<8,filter>(ray,context,vuint8(geomID),vuint8(primID)));
+      }
+    };
+
+#endif
+
+
+    /* ----------------------------- */
+    /* -- ray packet intersectors -- */
+    /* ----------------------------- */
+
+    struct PlueckerIntersector1KTriangleM
+    {
+      /*! Intersect k'th ray from ray packet of size K with M triangles. */
+      template<int M, int K, typename Epilog>
+      static  __forceinline bool intersect1(RayK<K>& ray,
+                                            size_t k,
+                                            const Vec3vf<M>& tri_v0,
+                                            const Vec3vf<M>& tri_v1,
+                                            const Vec3vf<M>& tri_v2,
+                                            const vbool<M>& flags,
+                                            const Epilog& epilog)
+      {
+        /* calculate vertices relative to ray origin */
+          const Vec3vf<M> O = broadcast<vfloat<M>>(ray.org,k);
+          const Vec3vf<M> D = broadcast<vfloat<M>>(ray.dir,k);
+          const Vec3vf<M> v0 = tri_v0-O;
+          const Vec3vf<M> v1 = tri_v1-O;
+          const Vec3vf<M> v2 = tri_v2-O;
+          
+          /* calculate triangle edges */
+          const Vec3vf<M> e0 = v2-v0;
+          const Vec3vf<M> e1 = v0-v1;
+          const Vec3vf<M> e2 = v1-v2;
+	  
+          /* perform edge tests */
+          const vfloat<M> U = dot(cross(e0,v2+v0),D);
+          const vfloat<M> V = dot(cross(e1,v0+v1),D);
+          const vfloat<M> W = dot(cross(e2,v1+v2),D);
+	  
+          const vfloat<M> UVW = U+V+W;
+          const vfloat<M> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+          vbool<M> valid = max(U,V,W) <= eps;
+#else
+          vbool<M> valid = (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif	
+          if (unlikely(none(valid))) return false;
+          
+          /* calculate geometry normal and denominator */
+          const Vec3vf<M> Ng = stable_triangle_normal(e0,e1,e2);
+          const vfloat<M> den = twice(dot(Ng,D));
+
+          /* perform depth test */
+          const vfloat<M> T = twice(dot(v0,Ng));
+          const vfloat<M> t = rcp(den)*T;
+          valid &= vfloat<M>(ray.tnear()[k]) <= t & t <= vfloat<M>(ray.tfar[k]);
+          if (unlikely(none(valid))) return false;
+          
+          /* avoid division by 0 */
+          valid &= den != vfloat<M>(zero);
+          if (unlikely(none(valid))) return false;
+          
+          /* update hit information */
+          QuadHitPlueckerM<M> hit(valid,U,V,UVW,t,Ng,flags);
+          return epilog(valid,hit);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct QuadMIntersectorKPlueckerBase
+    {
+      __forceinline QuadMIntersectorKPlueckerBase(const vbool<K>& valid, const RayK<K>& ray) {}
+            
+      /*! Intersects K rays with one of M triangles. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const vbool<K>& flags,
+                                        const Epilog& epilog) const
+      {
+        /* calculate vertices relative to ray origin */
+          vbool<K> valid = valid0;
+          const Vec3vf<K> O = ray.org;
+          const Vec3vf<K> D = ray.dir;
+          const Vec3vf<K> v0 = tri_v0-O;
+          const Vec3vf<K> v1 = tri_v1-O;
+          const Vec3vf<K> v2 = tri_v2-O;
+          
+          /* calculate triangle edges */
+          const Vec3vf<K> e0 = v2-v0;
+          const Vec3vf<K> e1 = v0-v1;
+          const Vec3vf<K> e2 = v1-v2;
+           
+          /* perform edge tests */
+          const vfloat<K> U = dot(Vec3vf<K>(cross(e0,v2+v0)),D);
+          const vfloat<K> V = dot(Vec3vf<K>(cross(e1,v0+v1)),D);
+          const vfloat<K> W = dot(Vec3vf<K>(cross(e2,v1+v2)),D);
+          const vfloat<K> UVW = U+V+W;
+          const vfloat<K> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+          valid &= max(U,V,W) <= eps;
+#else
+          valid &= (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif
+          if (unlikely(none(valid))) return false;
+          
+           /* calculate geometry normal and denominator */
+          const Vec3vf<K> Ng = stable_triangle_normal(e0,e1,e2);
+          const vfloat<K> den = twice(dot(Vec3vf<K>(Ng),D));
+
+          /* perform depth test */
+          const vfloat<K> T = twice(dot(v0,Vec3vf<K>(Ng)));
+          const vfloat<K> t = rcp(den)*T;
+          valid &= ray.tnear() <= t & t <= ray.tfar;
+          valid &= den != vfloat<K>(zero);
+          if (unlikely(none(valid))) return false;
+          
+          /* calculate hit information */
+          QuadHitPlueckerK<K> hit(U,V,UVW,t,Ng,flags);
+          return epilog(valid,hit);
+      }
+      
+      /*! Intersects K rays with one of M quads. */
+      template<typename Epilog>
+      __forceinline bool intersectK(const vbool<K>& valid0, 
+                                    RayK<K>& ray,
+                                    const Vec3vf<K>& v0,
+                                    const Vec3vf<K>& v1,
+                                    const Vec3vf<K>& v2,
+                                    const Vec3vf<K>& v3,
+                                    const Epilog& epilog) const
+      {
+        intersectK(valid0,ray,v0,v1,v3,vbool<K>(false),epilog);
+        if (none(valid0)) return true;
+        intersectK(valid0,ray,v2,v3,v1,vbool<K>(true ),epilog);
+        return none(valid0);
+      }
+    };
+
+    template<int M, int K, bool filter>
+      struct QuadMIntersectorKPluecker : public QuadMIntersectorKPlueckerBase<M,K,filter>
+    {
+      __forceinline QuadMIntersectorKPluecker(const vbool<K>& valid, const RayK<K>& ray)
+        : QuadMIntersectorKPlueckerBase<M,K,filter>(valid,ray) {}
+
+      __forceinline void intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                    const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Intersect1KEpilogM<M,K,filter> epilog(ray,k,context,geomID,primID);
+        PlueckerIntersector1KTriangleM::intersect1<M,K>(ray,k,v0,v1,v3,vbool<M>(false),epilog);
+        PlueckerIntersector1KTriangleM::intersect1<M,K>(ray,k,v2,v3,v1,vbool<M>(true ),epilog);
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                   const vuint<M>& geomID, const vuint<M>& primID) const
+      {
+        Occluded1KEpilogM<M,K,filter> epilog(ray,k,context,geomID,primID);
+        if (PlueckerIntersector1KTriangleM::intersect1<M,K>(ray,k,v0,v1,v3,vbool<M>(false),epilog)) return true;
+        if (PlueckerIntersector1KTriangleM::intersect1<M,K>(ray,k,v2,v3,v1,vbool<M>(true ),epilog)) return true;
+        return false;
+      }
+    };
+
+#if defined(__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<int K, bool filter>
+    struct QuadMIntersectorKPluecker<4,K,filter> : public QuadMIntersectorKPlueckerBase<4,K,filter>
+    {
+      __forceinline QuadMIntersectorKPluecker(const vbool<K>& valid, const RayK<K>& ray)
+        : QuadMIntersectorKPlueckerBase<4,K,filter>(valid,ray) {}
+      
+      template<typename Epilog>
+      __forceinline bool intersect1(RayK<K>& 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<8,K>(ray,k,vtx0,vtx1,vtx2,flags,epilog); 
+      }
+      
+      __forceinline bool intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                    const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,Intersect1KEpilogM<8,K,filter>(ray,k,context,vuint8(geomID),vuint8(primID)));
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const vuint4& geomID, const vuint4& primID) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,Occluded1KEpilogM<8,K,filter>(ray,k,context,vuint8(geomID),vuint8(primID)));
+      }
+    };
+
+#endif
+  }
+}

thirdparty/embree/kernels/geometry/quadi.h

@@ -0,0 +1,483 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../common/scene.h"
+
+namespace embree
+{
+  /* Stores M quads from an indexed face set */
+  template <int M>
+  struct QuadMi
+  {
+    /* Virtual interface to query information about the quad type */
+    struct Type : public PrimitiveType
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored quads */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+
+  public:
+
+    /* Default constructor */
+    __forceinline QuadMi() {  }
+
+    /* Construction from vertices and IDs */
+    __forceinline QuadMi(const vuint<M>& v0,
+                         const vuint<M>& v1,
+                         const vuint<M>& v2,
+                         const vuint<M>& v3,
+                         const vuint<M>& geomIDs,
+                         const vuint<M>& primIDs)
+#if defined(EMBREE_COMPACT_POLYS)
+      : geomIDs(geomIDs), primIDs(primIDs) {}
+#else
+     : v0_(v0),v1_(v1), v2_(v2), v3_(v3), geomIDs(geomIDs), primIDs(primIDs) {}
+#endif
+
+    /* Returns a mask that tells which quads are valid */
+    __forceinline vbool<M> valid() const { return primIDs != vuint<M>(-1); }
+
+    /* Returns if the specified quad is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return primIDs[i] != -1; }
+
+    /* Returns the number of stored quads */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    __forceinline       vuint<M>& geomID()       { return geomIDs; }
+    __forceinline const vuint<M>& geomID() const { return geomIDs; }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); assert(geomIDs[i] != -1); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M>& primID()       { return primIDs; }
+    __forceinline const vuint<M>& primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the quads */
+    __forceinline const BBox3fa bounds(const Scene *const scene, const size_t itime=0) const
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M && valid(i); i++) {
+        const QuadMesh* mesh = scene->get<QuadMesh>(geomID(i));
+        bounds.extend(mesh->bounds(primID(i),itime));
+      }
+      return bounds;
+    }
+
+    /* Calculate the linear bounds of the primitive */
+    __forceinline LBBox3fa linearBounds(const Scene* const scene, const size_t itime) {
+      return LBBox3fa(bounds(scene,itime+0),bounds(scene,itime+1));
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, size_t itime, size_t numTimeSteps)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const QuadMesh* mesh = scene->get<QuadMesh>(geomID(i));
+        allBounds.extend(mesh->linearBounds(primID(i), itime, numTimeSteps));
+      }
+      return allBounds;
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, const BBox1f time_range)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const QuadMesh* mesh = scene->get<QuadMesh>(geomID(i));
+        allBounds.extend(mesh->linearBounds(primID(i), time_range));
+      }
+      return allBounds;
+    }
+
+    /* Fill quad from quad list */
+    template<typename PrimRefT>
+    __forceinline void fill(const PrimRefT* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      vuint<M> geomID = -1, primID = -1;
+      const PrimRefT* prim = &prims[begin];
+      vuint<M> v0 = zero, v1 = zero, v2 = zero, v3 = zero;
+
+      for (size_t i=0; i<M; i++)
+      {
+        if (begin<end) {
+          geomID[i] = prim->geomID();
+          primID[i] = prim->primID();
+#if !defined(EMBREE_COMPACT_POLYS)
+          const QuadMesh* mesh = scene->get<QuadMesh>(prim->geomID());
+          const QuadMesh::Quad& q = mesh->quad(prim->primID());
+          unsigned int_stride = mesh->vertices0.getStride()/4;
+          v0[i] = q.v[0] * int_stride;
+          v1[i] = q.v[1] * int_stride;
+          v2[i] = q.v[2] * int_stride;
+          v3[i] = q.v[3] * int_stride;
+#endif
+          begin++;
+        } else {
+          assert(i);
+          if (likely(i > 0)) {
+            geomID[i] = geomID[0]; // always valid geomIDs
+            primID[i] = -1;        // indicates invalid data
+            v0[i] = v0[0];
+            v1[i] = v0[0];
+            v2[i] = v0[0];
+            v3[i] = v0[0];
+          }
+        }
+        if (begin<end) prim = &prims[begin];
+      }
+      new (this) QuadMi(v0,v1,v2,v3,geomID,primID); // FIXME: use non temporal store
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& begin, size_t end, Scene* scene, size_t itime)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, itime);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& begin, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, time_range);
+    }
+
+    friend embree_ostream operator<<(embree_ostream cout, const QuadMi& quad) {
+      return cout << "QuadMi<" << M << ">( "
+#if !defined(EMBREE_COMPACT_POLYS)
+                  << "v0 = " << quad.v0_ << ", v1 = " << quad.v1_ << ", v2 = " << quad.v2_ << ", v3 = " << quad.v3_ << ", "
+#endif
+                  << "geomID = " << quad.geomIDs << ", primID = " << quad.primIDs << " )";
+    }
+
+  protected:
+#if !defined(EMBREE_COMPACT_POLYS)
+    vuint<M> v0_;         // 4 byte offset of 1st vertex
+    vuint<M> v1_;         // 4 byte offset of 2nd vertex
+    vuint<M> v2_;         // 4 byte offset of 3rd vertex
+    vuint<M> v3_;         // 4 byte offset of 4th vertex
+#endif
+    vuint<M> geomIDs;    // geometry ID of mesh
+    vuint<M> primIDs;    // primitive ID of primitive inside mesh
+  };
+
+  namespace isa
+  {
+    
+  template<int M>
+    struct QuadMi : public embree::QuadMi<M>
+  {
+#if !defined(EMBREE_COMPACT_POLYS)
+    using embree::QuadMi<M>::v0_;
+    using embree::QuadMi<M>::v1_;
+    using embree::QuadMi<M>::v2_;
+    using embree::QuadMi<M>::v3_;
+#endif
+    using embree::QuadMi<M>::geomIDs;
+    using embree::QuadMi<M>::primIDs;
+    using embree::QuadMi<M>::geomID;
+    using embree::QuadMi<M>::primID;
+    using embree::QuadMi<M>::valid;
+    
+    template<int vid>
+    __forceinline Vec3f getVertex(const size_t index, const Scene *const scene) const
+    {
+#if defined(EMBREE_COMPACT_POLYS)
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID(index));
+      const QuadMesh::Quad& quad = mesh->quad(primID(index));
+      return (Vec3f) mesh->vertices[0][quad.v[vid]];
+#else
+      const vuint<M>& v = getVertexOffset<vid>();
+      const float* vertices = scene->vertices[geomID(index)];
+      return (Vec3f&) vertices[v[index]];
+#endif
+    }
+
+    template<int vid, typename T>
+    __forceinline Vec3<T> getVertex(const size_t index, const Scene *const scene, const size_t itime, const T& ftime) const
+    {
+#if defined(EMBREE_COMPACT_POLYS)
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID(index));
+      const QuadMesh::Quad& quad = mesh->quad(primID(index));
+      const Vec3fa v0 = mesh->vertices[itime+0][quad.v[vid]];
+      const Vec3fa v1 = mesh->vertices[itime+1][quad.v[vid]];
+#else
+      const vuint<M>& v = getVertexOffset<vid>();
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID(index));
+      const float* vertices0 = (const float*) mesh->vertexPtr(0,itime+0);
+      const float* vertices1 = (const float*) mesh->vertexPtr(0,itime+1);
+      const Vec3fa v0 = Vec3fa::loadu(vertices0+v[index]);
+      const Vec3fa v1 = Vec3fa::loadu(vertices1+v[index]);
+#endif
+      const Vec3<T> p0(v0.x,v0.y,v0.z);
+      const Vec3<T> p1(v1.x,v1.y,v1.z);
+      return lerp(p0,p1,ftime);
+    }
+
+    template<int vid, int K, typename T>
+    __forceinline Vec3<T> getVertex(const vbool<K>& valid, const size_t index, const Scene *const scene, const vint<K>& itime, const T& ftime) const
+    {
+      Vec3<T> p0, p1;
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID(index));
+
+      for (size_t mask=movemask(valid), i=bsf(mask); mask; mask=btc(mask,i), i=bsf(mask))
+      {
+#if defined(EMBREE_COMPACT_POLYS)
+        const QuadMesh::Quad& quad = mesh->quad(primID(index));
+        const Vec3fa v0 = mesh->vertices[itime[i]+0][quad.v[vid]];
+        const Vec3fa v1 = mesh->vertices[itime[i]+1][quad.v[vid]];
+#else
+        const vuint<M>& v = getVertexOffset<vid>();
+        const float* vertices0 = (const float*) mesh->vertexPtr(0,itime[i]+0);
+        const float* vertices1 = (const float*) mesh->vertexPtr(0,itime[i]+1);
+        const Vec3fa v0 = Vec3fa::loadu(vertices0+v[index]);
+        const Vec3fa v1 = Vec3fa::loadu(vertices1+v[index]);
+#endif
+        p0.x[i] = v0.x; p0.y[i] = v0.y; p0.z[i] = v0.z;
+        p1.x[i] = v1.x; p1.y[i] = v1.y; p1.z[i] = v1.z;
+      }
+      return (T(one)-ftime)*p0 + ftime*p1;
+    }
+
+    struct Quad {
+      vfloat4 v0,v1,v2,v3;
+    };
+
+#if defined(EMBREE_COMPACT_POLYS)
+    
+    __forceinline Quad loadQuad(const int i, const Scene* const scene) const 
+    {
+      const unsigned int geomID = geomIDs[i];
+      const unsigned int primID = primIDs[i];
+      if (unlikely(primID == -1)) return { zero, zero, zero, zero };
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID);
+      const QuadMesh::Quad& quad = mesh->quad(primID);
+      const vfloat4 v0 = (vfloat4) mesh->vertices0[quad.v[0]];
+      const vfloat4 v1 = (vfloat4) mesh->vertices0[quad.v[1]];
+      const vfloat4 v2 = (vfloat4) mesh->vertices0[quad.v[2]];
+      const vfloat4 v3 = (vfloat4) mesh->vertices0[quad.v[3]];
+      return { v0, v1, v2, v3 };
+    }
+
+    __forceinline Quad loadQuad(const int i, const int itime, const Scene* const scene) const 
+    {
+      const unsigned int geomID = geomIDs[i];
+      const unsigned int primID = primIDs[i];
+      if (unlikely(primID == -1)) return { zero, zero, zero, zero };
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID);
+      const QuadMesh::Quad& quad = mesh->quad(primID);
+      const vfloat4 v0 = (vfloat4) mesh->vertices[itime][quad.v[0]];
+      const vfloat4 v1 = (vfloat4) mesh->vertices[itime][quad.v[1]];
+      const vfloat4 v2 = (vfloat4) mesh->vertices[itime][quad.v[2]];
+      const vfloat4 v3 = (vfloat4) mesh->vertices[itime][quad.v[3]];
+      return { v0, v1, v2, v3 };
+    }
+    
+#else
+
+    __forceinline Quad loadQuad(const int i, const Scene* const scene) const 
+    {
+      const float* vertices = scene->vertices[geomID(i)];
+      const vfloat4 v0 = vfloat4::loadu(vertices + v0_[i]);
+      const vfloat4 v1 = vfloat4::loadu(vertices + v1_[i]);
+      const vfloat4 v2 = vfloat4::loadu(vertices + v2_[i]);
+      const vfloat4 v3 = vfloat4::loadu(vertices + v3_[i]);
+      return { v0, v1, v2, v3 };
+    }
+
+    __forceinline Quad loadQuad(const int i, const int itime, const Scene* const scene) const 
+    {
+      const unsigned int geomID = geomIDs[i];
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID);
+      const float* vertices = (const float*) mesh->vertexPtr(0,itime);
+      const vfloat4 v0 = vfloat4::loadu(vertices + v0_[i]);
+      const vfloat4 v1 = vfloat4::loadu(vertices + v1_[i]);
+      const vfloat4 v2 = vfloat4::loadu(vertices + v2_[i]);
+      const vfloat4 v3 = vfloat4::loadu(vertices + v3_[i]);
+      return { v0, v1, v2, v3 };
+    }
+    
+#endif
+
+    /* Gather the quads */
+    __forceinline void gather(Vec3vf<M>& p0,
+                              Vec3vf<M>& p1,
+                              Vec3vf<M>& p2,
+                              Vec3vf<M>& p3,
+                              const Scene *const scene) const;
+
+#if defined(__AVX512F__)
+    __forceinline void gather(Vec3vf16& p0,
+                              Vec3vf16& p1,
+                              Vec3vf16& p2,
+                              Vec3vf16& p3,
+                              const Scene *const scene) const;
+#endif
+
+    template<int K>
+#if defined(__INTEL_COMPILER) && (__INTEL_COMPILER < 2000) // workaround for compiler bug in ICC 2019
+    __noinline
+#else
+    __forceinline
+#endif
+    void gather(const vbool<K>& valid,
+      Vec3vf<K>& p0,
+      Vec3vf<K>& p1,
+      Vec3vf<K>& p2,
+      Vec3vf<K>& p3,
+      const size_t index,
+      const Scene* const scene,
+      const vfloat<K>& time) const
+    {
+      const QuadMesh* mesh = scene->get<QuadMesh>(geomID(index));
+
+      vfloat<K> ftime;
+      const vint<K> itime = mesh->timeSegment<K>(time, ftime);
+
+      const size_t first = bsf(movemask(valid));
+      if (likely(all(valid,itime[first] == itime)))
+      {
+        p0 = getVertex<0>(index, scene, itime[first], ftime);
+        p1 = getVertex<1>(index, scene, itime[first], ftime);
+        p2 = getVertex<2>(index, scene, itime[first], ftime);
+        p3 = getVertex<3>(index, scene, itime[first], ftime);
+      }
+      else
+      {
+        p0 = getVertex<0,K>(valid, index, scene, itime, ftime);
+        p1 = getVertex<1,K>(valid, index, scene, itime, ftime);
+        p2 = getVertex<2,K>(valid, index, scene, itime, ftime);
+        p3 = getVertex<3,K>(valid, index, scene, itime, ftime);
+      }
+    }
+
+    __forceinline void gather(Vec3vf<M>& p0,
+                              Vec3vf<M>& p1,
+                              Vec3vf<M>& p2,
+                              Vec3vf<M>& p3,
+                              const QuadMesh* mesh,
+                              const Scene *const scene,
+                              const int itime) const;
+
+    __forceinline void gather(Vec3vf<M>& p0,
+                              Vec3vf<M>& p1,
+                              Vec3vf<M>& p2,
+                              Vec3vf<M>& p3,
+                              const Scene *const scene,
+                              const float time) const;
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(QuadMesh* mesh)
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M; i++)
+      {
+        if (!valid(i)) break;
+        const unsigned primId = primID(i);
+        const QuadMesh::Quad& q = mesh->quad(primId);
+        const Vec3fa p0 = mesh->vertex(q.v[0]);
+        const Vec3fa p1 = mesh->vertex(q.v[1]);
+        const Vec3fa p2 = mesh->vertex(q.v[2]);
+        const Vec3fa p3 = mesh->vertex(q.v[3]);
+        bounds.extend(merge(BBox3fa(p0),BBox3fa(p1),BBox3fa(p2),BBox3fa(p3)));
+      }
+      return bounds;
+    }
+
+  private:
+#if !defined(EMBREE_COMPACT_POLYS)
+    template<int N> const vuint<M>& getVertexOffset() const;
+#endif
+  };
+
+#if !defined(EMBREE_COMPACT_POLYS)
+  template<> template<> __forceinline const vuint<4>& QuadMi<4>::getVertexOffset<0>() const { return v0_; }
+  template<> template<> __forceinline const vuint<4>& QuadMi<4>::getVertexOffset<1>() const { return v1_; }
+  template<> template<> __forceinline const vuint<4>& QuadMi<4>::getVertexOffset<2>() const { return v2_; }
+  template<> template<> __forceinline const vuint<4>& QuadMi<4>::getVertexOffset<3>() const { return v3_; }
+#endif
+
+  template<>
+  __forceinline void QuadMi<4>::gather(Vec3vf4& p0,
+                                       Vec3vf4& p1,
+                                       Vec3vf4& p2,
+                                       Vec3vf4& p3,
+                                       const Scene *const scene) const
+  {
+    prefetchL1(((char*)this)+0*64);
+    prefetchL1(((char*)this)+1*64);
+    const Quad tri0 = loadQuad(0,scene);
+    const Quad tri1 = loadQuad(1,scene);
+    const Quad tri2 = loadQuad(2,scene);
+    const Quad tri3 = loadQuad(3,scene);
+    transpose(tri0.v0,tri1.v0,tri2.v0,tri3.v0,p0.x,p0.y,p0.z);
+    transpose(tri0.v1,tri1.v1,tri2.v1,tri3.v1,p1.x,p1.y,p1.z);
+    transpose(tri0.v2,tri1.v2,tri2.v2,tri3.v2,p2.x,p2.y,p2.z);
+    transpose(tri0.v3,tri1.v3,tri2.v3,tri3.v3,p3.x,p3.y,p3.z);
+  }
+
+  template<>
+  __forceinline void QuadMi<4>::gather(Vec3vf4& p0,
+                                       Vec3vf4& p1,
+                                       Vec3vf4& p2,
+                                       Vec3vf4& p3,
+                                       const QuadMesh* mesh,
+                                       const Scene *const scene,
+                                       const int itime) const
+  {
+    // FIXME: for trianglei there all geometries are identical, is this the case here too?
+    
+    const Quad tri0 = loadQuad(0,itime,scene);
+    const Quad tri1 = loadQuad(1,itime,scene);
+    const Quad tri2 = loadQuad(2,itime,scene);
+    const Quad tri3 = loadQuad(3,itime,scene);
+    transpose(tri0.v0,tri1.v0,tri2.v0,tri3.v0,p0.x,p0.y,p0.z);
+    transpose(tri0.v1,tri1.v1,tri2.v1,tri3.v1,p1.x,p1.y,p1.z);
+    transpose(tri0.v2,tri1.v2,tri2.v2,tri3.v2,p2.x,p2.y,p2.z);
+    transpose(tri0.v3,tri1.v3,tri2.v3,tri3.v3,p3.x,p3.y,p3.z);
+  }
+
+  template<>
+  __forceinline void QuadMi<4>::gather(Vec3vf4& p0,
+                                       Vec3vf4& p1,
+                                       Vec3vf4& p2,
+                                       Vec3vf4& p3,
+                                       const Scene *const scene,
+                                       const float time) const
+  {
+    const QuadMesh* mesh = scene->get<QuadMesh>(geomID(0)); // in mblur mode all geometries are identical
+
+    float ftime;
+    const int itime = mesh->timeSegment(time, ftime);
+
+    Vec3vf4 a0,a1,a2,a3; gather(a0,a1,a2,a3,mesh,scene,itime);
+    Vec3vf4 b0,b1,b2,b3; gather(b0,b1,b2,b3,mesh,scene,itime+1);
+    p0 = lerp(a0,b0,vfloat4(ftime));
+    p1 = lerp(a1,b1,vfloat4(ftime));
+    p2 = lerp(a2,b2,vfloat4(ftime));
+    p3 = lerp(a3,b3,vfloat4(ftime));
+  }
+  }
+
+  template<int M>
+  typename QuadMi<M>::Type QuadMi<M>::type;
+
+  typedef QuadMi<4> Quad4i;
+}

thirdparty/embree/kernels/geometry/quadi_intersector.h

@@ -0,0 +1,350 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "quadi.h"
+#include "quad_intersector_moeller.h"
+#include "quad_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMiIntersector1Moeller
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersector1MoellerTrumbore<M,filter> Precalculations;
+
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMiIntersectorKMoeller
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersectorKMoellerTrumbore<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        Scene* scene = context->scene;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> p0 = quad.template getVertex<0>(i,scene);
+          const Vec3vf<K> p1 = quad.template getVertex<1>(i,scene);
+          const Vec3vf<K> p2 = quad.template getVertex<2>(i,scene);
+          const Vec3vf<K> p3 = quad.template getVertex<3>(i,scene);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        Scene* scene = context->scene;
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> p0 = quad.template getVertex<0>(i,scene);
+          const Vec3vf<K> p1 = quad.template getVertex<1>(i,scene);
+          const Vec3vf<K> p2 = quad.template getVertex<2>(i,scene);
+          const Vec3vf<K> p3 = quad.template getVertex<3>(i,scene);
+          if (pre.intersectK(valid0,ray,p0,p1,p2,p3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf4 v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf4 v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+    };
+
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMiIntersector1Pluecker
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersector1Pluecker<M,filter> Precalculations;
+
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMiIntersectorKPluecker
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersectorKPluecker<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        Scene* scene = context->scene;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> p0 = quad.template getVertex<0>(i,scene);
+          const Vec3vf<K> p1 = quad.template getVertex<1>(i,scene);
+          const Vec3vf<K> p2 = quad.template getVertex<2>(i,scene);
+          const Vec3vf<K> p3 = quad.template getVertex<3>(i,scene);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        Scene* scene = context->scene;
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> p0 = quad.template getVertex<0>(i,scene);
+          const Vec3vf<K> p1 = quad.template getVertex<1>(i,scene);
+          const Vec3vf<K> p2 = quad.template getVertex<2>(i,scene);
+          const Vec3vf<K> p3 = quad.template getVertex<3>(i,scene);
+          if (pre.intersectK(valid0,ray,p0,p1,p2,p3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf4 v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf4 v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+    };
+
+    /*! Intersects M motion blur quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMiMBIntersector1Moeller
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersector1MoellerTrumbore<M,filter> Precalculations;
+
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time());
+        pre.intersect(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time());
+        return pre.occluded(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M motion blur quads with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMiMBIntersectorKMoeller
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersectorKMoellerTrumbore<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M quads. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          Vec3vf<K> v0,v1,v2,v3; quad.template gather<K>(valid_i,v0,v1,v2,v3,i,context->scene,ray.time());
+          pre.intersectK(valid_i,ray,v0,v1,v2,v3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M quads. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          Vec3vf<K> v0,v1,v2,v3; quad.template gather<K>(valid_i,v0,v1,v2,v3,i,context->scene,ray.time());
+          if (pre.intersectK(valid0,ray,v0,v1,v2,v3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M quads and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time()[k]);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M quads. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time()[k]);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+    };
+
+    /*! Intersects M motion blur quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMiMBIntersector1Pluecker
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersector1Pluecker<M,filter> Precalculations;
+
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time());
+        pre.intersect(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time());
+        return pre.occluded(ray,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M motion blur quads with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMiMBIntersectorKPluecker
+    {
+      typedef QuadMi<M> Primitive;
+      typedef QuadMIntersectorKPluecker<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M quads. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          Vec3vf<K> v0,v1,v2,v3; quad.template gather<K>(valid_i,v0,v1,v2,v3,i,context->scene,ray.time());
+          pre.intersectK(valid_i,ray,v0,v1,v2,v3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M quads. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<QuadMi<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          Vec3vf<K> v0,v1,v2,v3; quad.template gather<K>(valid_i,v0,v1,v2,v3,i,context->scene,ray.time());
+          if (pre.intersectK(valid0,ray,v0,v1,v2,v3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M quads and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time()[k]);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M quads. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMi<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2,v3; quad.gather(v0,v1,v2,v3,context->scene,ray.time()[k]);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,quad.geomID(),quad.primID());
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/quadv.h

@@ -0,0 +1,165 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  /* Stores the vertices of M quads in struct of array layout */
+  template <int M>
+  struct QuadMv
+  { 
+  public:
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* Returns maximum number of stored quads */
+    static __forceinline size_t max_size() { return M; }
+    
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+   
+  public:
+
+    /* Default constructor */
+    __forceinline QuadMv() {}
+
+    /* Construction from vertices and IDs */
+    __forceinline QuadMv(const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const vuint<M>& geomIDs, const vuint<M>& primIDs)
+      : v0(v0), v1(v1), v2(v2), v3(v3), geomIDs(geomIDs), primIDs(primIDs) {}
+    
+    /* Returns a mask that tells which quads are valid */
+    __forceinline vbool<M> valid() const { return geomIDs != vuint<M>(-1); }
+
+    /* Returns true if the specified quad is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return geomIDs[i] != -1; }
+
+    /* Returns the number of stored quads */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    __forceinline       vuint<M>& geomID()       { return geomIDs; }
+    __forceinline const vuint<M>& geomID() const { return geomIDs; }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M> primID()       { return primIDs; }
+    __forceinline const vuint<M> primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the quads */
+    __forceinline BBox3fa bounds() const 
+    {
+      Vec3vf<M> lower = min(v0,v1,v2,v3);
+      Vec3vf<M> upper = max(v0,v1,v2,v3);
+      vbool<M> mask = valid();
+      lower.x = select(mask,lower.x,vfloat<M>(pos_inf));
+      lower.y = select(mask,lower.y,vfloat<M>(pos_inf));
+      lower.z = select(mask,lower.z,vfloat<M>(pos_inf));
+      upper.x = select(mask,upper.x,vfloat<M>(neg_inf));
+      upper.y = select(mask,upper.y,vfloat<M>(neg_inf));
+      upper.z = select(mask,upper.z,vfloat<M>(neg_inf));
+      return BBox3fa(Vec3fa(reduce_min(lower.x),reduce_min(lower.y),reduce_min(lower.z)),
+                     Vec3fa(reduce_max(upper.x),reduce_max(upper.y),reduce_max(upper.z)));
+    }
+    
+    /* Non temporal store */
+    __forceinline static void store_nt(QuadMv* dst, const QuadMv& src)
+    {
+      vfloat<M>::store_nt(&dst->v0.x,src.v0.x);
+      vfloat<M>::store_nt(&dst->v0.y,src.v0.y);
+      vfloat<M>::store_nt(&dst->v0.z,src.v0.z);
+      vfloat<M>::store_nt(&dst->v1.x,src.v1.x);
+      vfloat<M>::store_nt(&dst->v1.y,src.v1.y);
+      vfloat<M>::store_nt(&dst->v1.z,src.v1.z);
+      vfloat<M>::store_nt(&dst->v2.x,src.v2.x);
+      vfloat<M>::store_nt(&dst->v2.y,src.v2.y);
+      vfloat<M>::store_nt(&dst->v2.z,src.v2.z);
+      vfloat<M>::store_nt(&dst->v3.x,src.v3.x);
+      vfloat<M>::store_nt(&dst->v3.y,src.v3.y);
+      vfloat<M>::store_nt(&dst->v3.z,src.v3.z);
+      vuint<M>::store_nt(&dst->geomIDs,src.geomIDs);
+      vuint<M>::store_nt(&dst->primIDs,src.primIDs);
+    }
+
+    /* Fill quad from quad list */
+    __forceinline void fill(const PrimRef* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero, v3 = zero;
+      
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+	const PrimRef& prim = prims[begin];
+        const unsigned geomID = prim.geomID();
+        const unsigned primID = prim.primID();
+        const QuadMesh* __restrict__ const mesh = scene->get<QuadMesh>(geomID);
+        const QuadMesh::Quad& quad = mesh->quad(primID);
+        const Vec3fa& p0 = mesh->vertex(quad.v[0]);
+        const Vec3fa& p1 = mesh->vertex(quad.v[1]);
+        const Vec3fa& p2 = mesh->vertex(quad.v[2]);
+        const Vec3fa& p3 = mesh->vertex(quad.v[3]);
+        vgeomID [i] = geomID;
+        vprimID [i] = primID;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+        v3.x[i] = p3.x; v3.y[i] = p3.y; v3.z[i] = p3.z;
+      }
+      QuadMv::store_nt(this,QuadMv(v0,v1,v2,v3,vgeomID,vprimID));
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(QuadMesh* mesh)
+    {
+      BBox3fa bounds = empty;
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero;
+	
+      for (size_t i=0; i<M; i++)
+      {
+        if (primID(i) == -1) break;
+        const unsigned geomId = geomID(i);
+        const unsigned primId = primID(i);
+        const QuadMesh::Quad& quad = mesh->quad(primId);
+        const Vec3fa p0 = mesh->vertex(quad.v[0]);
+        const Vec3fa p1 = mesh->vertex(quad.v[1]);
+        const Vec3fa p2 = mesh->vertex(quad.v[2]);
+        const Vec3fa p3 = mesh->vertex(quad.v[3]);
+        bounds.extend(merge(BBox3fa(p0),BBox3fa(p1),BBox3fa(p2),BBox3fa(p3)));
+        vgeomID [i] = geomId;
+        vprimID [i] = primId;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+        v3.x[i] = p3.x; v3.y[i] = p3.y; v3.z[i] = p3.z;
+      }
+      new (this) QuadMv(v0,v1,v2,v3,vgeomID,vprimID);
+      return bounds;
+    }
+   
+  public:
+    Vec3vf<M> v0;      // 1st vertex of the quads
+    Vec3vf<M> v1;      // 2nd vertex of the quads
+    Vec3vf<M> v2;      // 3rd vertex of the quads
+    Vec3vf<M> v3;      // 4th vertex of the quads
+  private:
+    vuint<M> geomIDs; // geometry ID
+    vuint<M> primIDs; // primitive ID
+  };
+
+  template<int M>
+  typename QuadMv<M>::Type QuadMv<M>::type;
+
+  typedef QuadMv<4> Quad4v;
+}

thirdparty/embree/kernels/geometry/quadv_intersector.h

@@ -0,0 +1,181 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "quadv.h"
+#include "quad_intersector_moeller.h"
+#include "quad_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMvIntersector1Moeller
+    {
+      typedef QuadMv<M> Primitive;
+      typedef QuadMIntersector1MoellerTrumbore<M,filter> Precalculations;
+        
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+        
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.occluded(ray,context, quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMvIntersectorKMoeller
+    {
+      typedef QuadMv<M> Primitive;
+      typedef QuadMIntersectorKMoellerTrumbore<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        for (size_t i=0; i<QuadMv<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(quad.v0,i);
+          const Vec3vf<K> p1 = broadcast<vfloat<K>>(quad.v1,i);
+          const Vec3vf<K> p2 = broadcast<vfloat<K>>(quad.v2,i);
+          const Vec3vf<K> p3 = broadcast<vfloat<K>>(quad.v3,i);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<QuadMv<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(quad.v0,i);
+          const Vec3vf<K> p1 = broadcast<vfloat<K>>(quad.v1,i);
+          const Vec3vf<K> p2 = broadcast<vfloat<K>>(quad.v2,i);
+          const Vec3vf<K> p3 = broadcast<vfloat<K>>(quad.v3,i);
+          if (pre.intersectK(valid0,ray,p0,p1,p2,p3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect1(ray,k,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.occluded1(ray,k,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+    };
+
+    /*! Intersects M quads with 1 ray */
+    template<int M, bool filter>
+    struct QuadMvIntersector1Pluecker
+    {
+      typedef QuadMv<M> Primitive;
+      typedef QuadMIntersector1Pluecker<M,filter> Precalculations;
+        
+      /*! Intersect a ray with the M quads and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+        
+      /*! Test if the ray is occluded by one of M quads. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.occluded(ray,context, quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& quad)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, quad);
+      }
+    };
+
+    /*! Intersects M triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct QuadMvIntersectorKPluecker
+    {
+      typedef QuadMv<M> Primitive;
+      typedef QuadMIntersectorKPluecker<M,K,filter> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        for (size_t i=0; i<QuadMv<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(quad.v0,i);
+          const Vec3vf<K> p1 = broadcast<vfloat<K>>(quad.v1,i);
+          const Vec3vf<K> p2 = broadcast<vfloat<K>>(quad.v2,i);
+          const Vec3vf<K> p3 = broadcast<vfloat<K>>(quad.v3,i);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,IntersectKEpilogM<M,K,filter>(ray,context,quad.geomID(),quad.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<QuadMv<M>::max_size(); i++)
+        {
+          if (!quad.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(quad.v0,i);
+          const Vec3vf<K> p1 = broadcast<vfloat<K>>(quad.v1,i);
+          const Vec3vf<K> p2 = broadcast<vfloat<K>>(quad.v2,i);
+          const Vec3vf<K> p3 = broadcast<vfloat<K>>(quad.v3,i);
+          if (pre.intersectK(valid0,ray,p0,p1,p2,p3,OccludedKEpilogM<M,K,filter>(valid0,ray,context,quad.geomID(),quad.primID(),i)))
+            break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect1(ray,k,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const QuadMv<M>& quad)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.occluded1(ray,k,context,quad.v0,quad.v1,quad.v2,quad.v3,quad.geomID(),quad.primID());
+      }
+    };
+  }
+}
+

thirdparty/embree/kernels/geometry/roundline_intersector.h

@@ -0,0 +1,715 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "curve_intersector_precalculations.h"
+
+
+/*
+  
+  This file implements the intersection of a ray with a round linear
+  curve segment. We define the geometry of such a round linear curve
+  segment from point p0 with radius r0 to point p1 with radius r1
+  using the cone that touches spheres p0/r0 and p1/r1 tangentially
+  plus the sphere p1/r1. We denote the tangentially touching cone from
+  p0/r0 to p1/r1 with cone(p0,r0,p1,r1) and the cone plus the ending
+  sphere with cone_sphere(p0,r0,p1,r1).
+
+  For multiple connected round linear curve segments this construction
+  yield a proper shape when viewed from the outside. Using the
+  following CSG we can also handle the interior in most common cases:
+
+     round_linear_curve(pl,rl,p0,r0,p1,r1,pr,rr) =
+       cone_sphere(p0,r0,p1,r1) - cone(pl,rl,p0,r0) - cone(p1,r1,pr,rr)
+
+  Thus by subtracting the neighboring cone geometries, we cut away
+  parts of the center cone_sphere surface which lie inside the
+  combined curve. This approach works as long as geometry of the
+  current cone_sphere penetrates into direct neighbor segments only,
+  and not into segments further away.
+  
+  To construct a cone that touches two spheres at p0 and p1 with r0
+  and r1, one has to increase the cone radius at r0 and r1 to obtain
+  larger radii w0 and w1, such that the infinite cone properly touches
+  the spheres.  From the paper "Ray Tracing Generalized Tube
+  Primitives: Method and Applications"
+  (https://www.researchgate.net/publication/334378683_Ray_Tracing_Generalized_Tube_Primitives_Method_and_Applications)
+  one can derive the following equations for these increased
+  radii:
+
+     sr = 1.0f / sqrt(1-sqr(dr)/sqr(p1-p0))
+     w0 = sr*r0
+     w1 = sr*r1
+
+  Further, we want the cone to start where it touches the sphere at p0
+  and to end where it touches sphere at p1.  Therefore, we need to
+  construct clipping locations y0 and y1 for the start and end of the
+  cone. These start and end clipping location of the cone can get
+  calculated as:
+
+     Y0 =               - r0 * (r1-r0) / length(p1-p0)
+     Y1 = length(p1-p0) - r1 * (r1-r0) / length(p1-p0)
+
+  Where the cone starts a distance Y0 and ends a distance Y1 away of
+  point p0 along the cone center. The distance between Y1-Y0 can get
+  calculated as:
+
+    dY = length(p1-p0) - (r1-r0)^2 / length(p1-p0)
+
+  In the code below, Y will always be scaled by length(p1-p0) to
+  obtain y and you will find the terms r0*(r1-r0) and
+  (p1-p0)^2-(r1-r0)^2.
+
+ */
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+      struct RoundLineIntersectorHitM
+      {
+        __forceinline RoundLineIntersectorHitM() {}
+        
+        __forceinline RoundLineIntersectorHitM(const vfloat<M>& u, const vfloat<M>& v, const vfloat<M>& t, const Vec3vf<M>& Ng)
+          : vu(u), vv(v), vt(t), vNg(Ng) {}
+	
+        __forceinline void finalize() {}
+	
+        __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+        __forceinline float t  (const size_t i) const { return vt[i]; }
+        __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+
+        __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); }
+        __forceinline vfloat<M> t () const { return vt; }
+        __forceinline Vec3vf<M> Ng() const { return vNg; }
+       
+      public:
+        vfloat<M> vu;
+        vfloat<M> vv;
+        vfloat<M> vt;
+        Vec3vf<M> vNg;
+      };
+    
+    namespace __roundline_internal
+    {
+      template<int M>
+        struct ConeGeometry
+        {
+          ConeGeometry (const Vec4vf<M>& a, const Vec4vf<M>& b)
+          : p0(a.xyz()), p1(b.xyz()), dP(p1-p0), dPdP(dot(dP,dP)), r0(a.w), sqr_r0(sqr(r0)), r1(b.w), dr(r1-r0), drdr(dr*dr), r0dr (r0*dr), g(dPdP - drdr) {}
+          
+          /* 
+             
+             This function tests if a point is accepted by first cone
+             clipping plane.
+
+             First, we need to project the point onto the line p0->p1:
+             
+               Y = (p-p0)*(p1-p0)/length(p1-p0)
+             
+             This value y is the distance to the projection point from
+             p0. The clip distances are calculated as:
+             
+               Y0 =               - r0 * (r1-r0) / length(p1-p0)
+               Y1 = length(p1-p0) - r1 * (r1-r0) / length(p1-p0)
+             
+             Thus to test if the point p is accepted by the first
+             clipping plane we need to test Y > Y0 and to test if it
+             is accepted by the second clipping plane we need to test
+             Y < Y1.
+             
+             By multiplying the calculations with length(p1-p0) these
+             calculation can get simplied to:
+             
+               y = (p-p0)*(p1-p0)
+               y0 =           - r0 * (r1-r0)
+               y1 = (p1-p0)^2 - r1 * (r1-r0)
+
+             and the test y > y0 and y < y1.
+             
+          */
+          
+          __forceinline vbool<M> isClippedByPlane (const vbool<M>& valid_i, const Vec3vf<M>& p) const
+          {
+            const Vec3vf<M> p0p = p - p0;
+            const vfloat<M> y = dot(p0p,dP);
+            const vfloat<M> cap0 = -r0dr;
+            const vbool<M> inside_cone = y > cap0;
+            return valid_i & (p0.x != vfloat<M>(inf)) & (p1.x != vfloat<M>(inf)) & inside_cone;
+          }
+          
+          /* 
+             
+             This function tests whether a point lies inside the capped cone
+             tangential to its ending spheres.
+
+             Therefore one has to check if the point is inside the
+             region defined by the cone clipping planes, which is
+             performed similar as in the previous function.
+             
+             To perform the inside cone test we need to project the
+             point onto the line p0->p1:
+             
+               dP = p1-p0
+               Y = (p-p0)*dP/length(dP)
+                           
+             This value Y is the distance to the projection point from
+             p0. To obtain a parameter value u going from 0 to 1 along
+             the line p0->p1 we calculate:
+             
+               U = Y/length(dP)
+             
+             The radii to use at points p0 and p1 are:
+             
+               w0 = sr * r0
+               w1 = sr * r1
+               dw = w1-w0
+             
+             Using these radii and u one can directly test if the point
+             lies inside the cone using the formula dP*dP < wy*wy with:
+             
+               wy = w0 + u*dw
+               py = p0 + u*dP - p
+                          
+             By multiplying the calculations with length(p1-p0) and
+             inserting the definition of w can obtain simpler equations:
+             
+               y = (p-p0)*dP
+               ry = r0 + y/dP^2 * dr
+               wy = sr*ry        
+               py = p0 + y/dP^2*dP - p
+               y0 =      - r0 * dr
+               y1 = dP^2 - r1 * dr
+             
+             Thus for the in-cone test we get:
+             
+                    py^2 < wy^2
+               <=>  py^2 < sr^2 * ry^2
+               <=>  py^2 * ( dP^2 - dr^2 ) < dP^2 * ry^2
+             
+             This can further get simplified to:
+             
+               (p0-p)^2 * (dP^2 - dr^2) - y^2 < dP^2 * r0^2 + 2.0f*r0*dr*y;            
+                      
+          */
+          
+          __forceinline vbool<M> isInsideCappedCone (const vbool<M>& valid_i, const Vec3vf<M>& p) const
+          {
+            const Vec3vf<M> p0p = p - p0;
+            const vfloat<M> y = dot(p0p,dP);
+            const vfloat<M> cap0 = -r0dr+vfloat<M>(ulp);
+            const vfloat<M> cap1 = -r1*dr + dPdP;
+            
+            vbool<M> inside_cone = valid_i & (p0.x != vfloat<M>(inf)) & (p1.x != vfloat<M>(inf));
+            inside_cone &= y > cap0;  // start clipping plane
+            inside_cone &= y < cap1;  // end clipping plane 
+            inside_cone &= sqr(p0p)*g - sqr(y) < dPdP * sqr_r0 + 2.0f*r0dr*y; // in cone test
+            return inside_cone;
+          }
+          
+        protected:
+          Vec3vf<M> p0;
+          Vec3vf<M> p1;
+          Vec3vf<M> dP;
+          vfloat<M> dPdP;
+          vfloat<M> r0;
+          vfloat<M> sqr_r0;
+          vfloat<M> r1;
+          vfloat<M> dr;
+          vfloat<M> drdr;
+          vfloat<M> r0dr;
+          vfloat<M> g;
+        };
+      
+      template<int M>
+        struct ConeGeometryIntersector : public ConeGeometry<M>
+      {
+        using ConeGeometry<M>::p0;
+        using ConeGeometry<M>::p1;
+        using ConeGeometry<M>::dP;
+        using ConeGeometry<M>::dPdP;
+        using ConeGeometry<M>::r0;
+        using ConeGeometry<M>::sqr_r0;
+        using ConeGeometry<M>::r1;
+        using ConeGeometry<M>::dr;
+        using ConeGeometry<M>::r0dr;
+        using ConeGeometry<M>::g;
+        
+        ConeGeometryIntersector (const Vec3vf<M>& ray_org, const Vec3vf<M>& ray_dir, const vfloat<M>& dOdO, const vfloat<M>& rcp_dOdO, const Vec4vf<M>& a, const Vec4vf<M>& b)
+          : ConeGeometry<M>(a,b), org(ray_org), O(ray_org-p0), dO(ray_dir),  dOdO(dOdO), rcp_dOdO(rcp_dOdO), OdP(dot(dP,O)), dOdP(dot(dP,dO)),  yp(OdP + r0dr) {}
+        
+        /*
+          
+          This function intersects a ray with a cone that touches a
+          start sphere p0/r0 and end sphere p1/r1.
+          
+          To find this ray/cone intersections one could just
+          calculate radii w0 and w1 as described above and use a
+          standard ray/cone intersection routine with these
+          radii. However, it turns out that calculations can get
+          simplified when deriving a specialized ray/cone
+          intersection for this special case. We perform
+          calculations relative to the cone origin p0 and define:
+            
+            O  = ray_org - p0
+            dO = ray_dir
+            dP = p1-p0
+            dr = r1-r0
+            dw = w1-w0
+            
+          For some t we can compute the potential hit point h = O + t*dO and
+          project it onto the cone vector dP to obtain u = (h*dP)/(dP*dP). In
+          case of an intersection, the squared distance from the hit point
+          projected onto the cone center line to the hit point should be equal
+          to the squared cone radius at u:
+            
+            (u*dP - h)^2 = (w0 + u*dw)^2
+           
+          Inserting the definition of h, u, w0, and dw into this formula, then
+          factoring out all terms, and sorting by t^2, t^1, and t^0 terms
+          yields a quadratic equation to solve.
+            
+          Inserting u:
+            ( (h*dP)*dP/dP^2 - h )^2 = ( w0 + (h*dP)*dw/dP^2 )^2
+            
+          Multiplying by dP^4:
+            ( (h*dP)*dP - h*dP^2 )^2 = ( w0*dP^2 + (h*dP)*dw )^2
+            
+          Inserting w0 and dw:
+            ( (h*dP)*dP - h*dP^2 )^2 = ( r0*dP^2 + (h*dP)*dr )^2 / (1-dr^2/dP^2)
+            ( (h*dP)*dP - h*dP^2 )^2 *(dP^2 - dr^2) = dP^2 * ( r0*dP^2 + (h*dP)*dr )^2
+            
+          Now one can insert the definition of h, factor out, and presort by t:
+            ( ((O + t*dO)*dP)*dP - (O + t*dO)*dP^2 )^2 *(dP^2 - dr^2) = dP^2 * ( r0*dP^2 + ((O + t*dO)*dP)*dr )^2
+            ( (O*dP)*dP-O*dP^2 + t*( (dO*dP)*dP - dO*dP^2 ) )^2 *(dP^2 - dr^2) = dP^2 * ( r0*dP^2 + (O*dP)*dr + t*(dO*dP)*dr )^2
+            
+          Factoring out further and sorting by t^2, t^1 and t^0 yields:
+            
+            0 =   t^2 * [ ((dO*dP)*dP - dO-dP^2)^2 * (dP^2 - dr^2) - dP^2*(dO*dP)^2*dr^2 ]
+              + 2*t^1 * [ ((O*dP)*dP - O*dP^2) * ((dO*dP)*dP - dO*dP^2) * (dP^2 - dr^2) - dP^2*(r0*dP^2 + (O*dP)*dr)*(dO*dP)*dr ]
+              +   t^0 * [ ( (O*dP)*dP - O*dP^2)^2 * (dP^2-dr^2) - dP^2*(r0*dP^2 + (O*dP)*dr)^2 ]
+            
+          This can be simplified to:
+            
+             0 =   t^2 * [ (dP^2 - dr^2)*dO^2 - (dO*dP)^2 ]
+               + 2*t^1 * [ (dP^2 - dr^2)*(O*dO) - (dO*dP)*(O*dP + r0*dr) ]
+               +   t^0 * [ (dP^2 - dr^2)*O^2 - (O*dP)^2 - r0^2*dP^2 - 2.0f*r0*dr*(O*dP) ]
+            
+          Solving this quadratic equation yields the values for t at which the
+          ray intersects the cone.
+          
+        */
+        
+        __forceinline bool intersectCone(vbool<M>& valid, vfloat<M>& lower, vfloat<M>& upper)
+        {
+          /* return no hit by default */
+          lower = pos_inf;
+          upper = neg_inf;
+          
+          /* compute quadratic equation A*t^2 + B*t + C = 0 */
+          const vfloat<M> OO = dot(O,O);
+          const vfloat<M> OdO = dot(dO,O);
+          const vfloat<M> A = g * dOdO - sqr(dOdP);
+          const vfloat<M> B = 2.0f * (g*OdO - dOdP*yp);
+          const vfloat<M> C = g*OO - sqr(OdP) - sqr_r0*dPdP - 2.0f*r0dr*OdP;
+          
+          /* we miss the cone if determinant is smaller than zero */
+          const vfloat<M> D = B*B - 4.0f*A*C;
+          valid &= (D >= 0.0f & g > 0.0f);  // if g <= 0 then the cone is inside a sphere end
+          
+          /* When rays are parallel to the cone surface, then the
+           * ray may be inside or outside the cone. We just assume a
+           * miss in that case, which is fine as rays inside the
+           * cone would anyway hit the ending spheres in that
+           * case. */
+          valid &= abs(A) > min_rcp_input;
+          if (unlikely(none(valid))) {
+            return false;
+          }
+          
+          /* compute distance to front and back hit */
+          const vfloat<M> Q = sqrt(D);
+          const vfloat<M> rcp_2A = rcp(2.0f*A);
+          t_cone_front = (-B-Q)*rcp_2A;
+          y_cone_front = yp + t_cone_front*dOdP;
+          lower = select( (y_cone_front > -(float)ulp) & (y_cone_front <= g) & (g > 0.0f), t_cone_front, vfloat<M>(pos_inf));
+#if !defined (EMBREE_BACKFACE_CULLING_CURVES)
+          t_cone_back = (-B+Q)*rcp_2A;
+          y_cone_back  = yp + t_cone_back *dOdP;
+          upper = select( (y_cone_back  > -(float)ulp) & (y_cone_back  <= g) & (g > 0.0f), t_cone_back , vfloat<M>(neg_inf));
+#endif          
+          return true;
+        }
+        
+        /* 
+           This function intersects the ray with the end sphere at
+           p1. We already clip away hits that are inside the
+           neighboring cone segment.
+           
+        */
+        
+        __forceinline void intersectEndSphere(vbool<M>& valid, 
+                                              const ConeGeometry<M>& coneR, 
+                                              vfloat<M>& lower, vfloat<M>& upper)
+        {
+          /* calculate front and back hit with end sphere */
+          const Vec3vf<M> O1 = org - p1;
+          const vfloat<M> O1dO = dot(O1,dO);
+          const vfloat<M> h2 = sqr(O1dO) - dOdO*(sqr(O1) - sqr(r1));
+          const vfloat<M> rhs1 = select( h2 >= 0.0f, sqrt(h2), vfloat<M>(neg_inf) );
+          
+          /* clip away front hit if it is inside next cone segment */
+          t_sph1_front = (-O1dO - rhs1)*rcp_dOdO;
+          const Vec3vf<M> hit_front = org + t_sph1_front*dO;
+          vbool<M> valid_sph1_front = h2 >= 0.0f & yp + t_sph1_front*dOdP > g & !coneR.isClippedByPlane (valid, hit_front);
+          lower = select(valid_sph1_front, t_sph1_front, vfloat<M>(pos_inf));
+          
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          /* clip away back hit if it is inside next cone segment */
+          t_sph1_back  = (-O1dO + rhs1)*rcp_dOdO;
+          const Vec3vf<M> hit_back = org + t_sph1_back*dO;
+          vbool<M> valid_sph1_back  = h2 >= 0.0f & yp + t_sph1_back*dOdP > g & !coneR.isClippedByPlane (valid, hit_back);
+          upper = select(valid_sph1_back, t_sph1_back,  vfloat<M>(neg_inf));
+#else
+          upper = vfloat<M>(neg_inf);
+#endif
+        }
+
+        __forceinline void intersectBeginSphere(const vbool<M>& valid, 
+                                                vfloat<M>& lower, vfloat<M>& upper)
+        {
+          /* calculate front and back hit with end sphere */
+          const Vec3vf<M> O1 = org - p0;
+          const vfloat<M> O1dO = dot(O1,dO);
+          const vfloat<M> h2 = sqr(O1dO) - dOdO*(sqr(O1) - sqr(r0));
+          const vfloat<M> rhs1 = select( h2 >= 0.0f, sqrt(h2), vfloat<M>(neg_inf) );
+          
+          /* clip away front hit if it is inside next cone segment */
+          t_sph0_front = (-O1dO - rhs1)*rcp_dOdO;
+          vbool<M> valid_sph1_front = valid & h2 >= 0.0f & yp + t_sph0_front*dOdP < 0;
+          lower = select(valid_sph1_front, t_sph0_front, vfloat<M>(pos_inf));
+
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          /* clip away back hit if it is inside next cone segment */
+          t_sph0_back  = (-O1dO + rhs1)*rcp_dOdO;
+          vbool<M> valid_sph1_back  = valid & h2 >= 0.0f & yp + t_sph0_back*dOdP < 0;
+          upper = select(valid_sph1_back, t_sph0_back,  vfloat<M>(neg_inf));
+#else   
+          upper = vfloat<M>(neg_inf);
+#endif
+        }
+        
+        /* 
+           
+           This function calculates the geometry normal of some cone hit.
+           
+           For a given hit point h (relative to p0) with a cone
+           starting at p0 with radius w0 and ending at p1 with
+           radius w1 one normally calculates the geometry normal by
+           first calculating the parmetric u hit location along the
+           cone:
+           
+             u = dot(h,dP)/dP^2
+           
+           Using this value one can now directly calculate the
+           geometry normal by bending the connection vector (h-u*dP)
+           from hit to projected hit with some cone dependent value
+           dw/sqrt(dP^2) * normalize(dP):
+           
+             Ng = normalize(h-u*dP) - dw/length(dP) * normalize(dP)
+           
+           The length of the vector (h-u*dP) can also get calculated
+           by interpolating the radii as w0+u*dw which yields:
+           
+             Ng = (h-u*dP)/(w0+u*dw) - dw/dP^2 * dP
+           
+           Multiplying with (w0+u*dw) yield a scaled Ng':
+           
+             Ng' = (h-u*dP) - (w0+u*dw)*dw/dP^2*dP
+           
+           Inserting the definition of w0 and dw and refactoring
+           yield a further scaled Ng'':
+           
+             Ng'' = (dP^2 - dr^2) (h-q) - (r0+u*dr)*dr*dP
+           
+           Now inserting the definition of u gives and multiplying
+           with the denominator yields:
+           
+             Ng''' = (dP^2-dr^2)*(dP^2*h-dot(h,dP)*dP) - (dP^2*r0+dot(h,dP)*dr)*dr*dP
+           
+           Factoring out, cancelling terms, dividing by dP^2, and
+           factoring again yields finally:
+           
+             Ng'''' = (dP^2-dr^2)*h - dP*(dot(h,dP) + r0*dr)
+           
+        */
+        
+        __forceinline Vec3vf<M> Ng_cone(const vbool<M>& front_hit) const
+        {
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          const vfloat<M> y = select(front_hit, y_cone_front, y_cone_back);
+          const vfloat<M> t = select(front_hit, t_cone_front, t_cone_back);
+          const Vec3vf<M> h = O + t*dO;
+          return g*h-dP*y;
+#else
+          const Vec3vf<M> h = O + t_cone_front*dO;
+          return g*h-dP*y_cone_front;
+#endif
+        }
+        
+        /* compute geometry normal of sphere hit as the difference
+         * vector from hit point to sphere center */
+        
+        __forceinline Vec3vf<M> Ng_sphere1(const vbool<M>& front_hit) const
+        {
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          const vfloat<M> t_sph1 = select(front_hit, t_sph1_front, t_sph1_back);
+          return org+t_sph1*dO-p1;
+#else 
+          return org+t_sph1_front*dO-p1;
+#endif
+        }
+
+        __forceinline Vec3vf<M> Ng_sphere0(const vbool<M>& front_hit) const
+        {
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          const vfloat<M> t_sph0 = select(front_hit, t_sph0_front, t_sph0_back);
+          return org+t_sph0*dO-p0;
+#else
+          return org+t_sph0_front*dO-p0;
+#endif
+        }
+        
+        /* 
+           This function calculates the u coordinate of a
+           hit. Therefore we use the hit distance y (which is zero
+           at the first cone clipping plane) and divide by distance
+           g between the clipping planes.
+           
+        */
+        
+        __forceinline vfloat<M> u_cone(const vbool<M>& front_hit) const
+        {
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+          const vfloat<M> y = select(front_hit, y_cone_front, y_cone_back);
+          return clamp(y*rcp(g));
+#else
+          return clamp(y_cone_front*rcp(g));
+#endif
+        }
+        
+      private:
+        Vec3vf<M> org;
+        Vec3vf<M> O;
+        Vec3vf<M> dO;
+        vfloat<M> dOdO;
+        vfloat<M> rcp_dOdO;
+        vfloat<M> OdP;
+        vfloat<M> dOdP;
+        
+        /* for ray/cone intersection */
+      private:
+        vfloat<M> yp;
+        vfloat<M> y_cone_front;
+        vfloat<M> t_cone_front;
+#if !defined (EMBREE_BACKFACE_CULLING_CURVES)
+        vfloat<M> y_cone_back;
+        vfloat<M> t_cone_back;
+#endif
+        
+        /* for ray/sphere intersection */
+      private:
+        vfloat<M> t_sph1_front;
+        vfloat<M> t_sph0_front;
+#if !defined (EMBREE_BACKFACE_CULLING_CURVES)
+        vfloat<M> t_sph1_back;
+        vfloat<M> t_sph0_back;
+#endif
+      };
+      
+      
+      template<int M, typename Epilog, typename ray_tfar_func>
+        static __forceinline bool intersectConeSphere(const vbool<M>& valid_i,
+                                                      const Vec3vf<M>& ray_org_in, const Vec3vf<M>& ray_dir, 
+                                                      const vfloat<M>& ray_tnear, const ray_tfar_func& ray_tfar,
+                                                      const Vec4vf<M>& v0, const Vec4vf<M>& v1,
+                                                      const Vec4vf<M>& vL, const Vec4vf<M>& vR,
+                                                      const Epilog& epilog)
+      {         
+        vbool<M> valid = valid_i;
+        
+        /* move ray origin closer to make calculations numerically stable */
+        const vfloat<M> dOdO = sqr(ray_dir);
+        const vfloat<M> rcp_dOdO = rcp(dOdO);
+        const Vec3vf<M> center = vfloat<M>(0.5f)*(v0.xyz()+v1.xyz());
+        const vfloat<M> dt = dot(center-ray_org_in,ray_dir)*rcp_dOdO;
+        const Vec3vf<M> ray_org = ray_org_in + dt*ray_dir;
+        
+        /* intersect with cone from v0 to v1 */
+        vfloat<M> t_cone_lower, t_cone_upper;
+        ConeGeometryIntersector<M> cone (ray_org, ray_dir, dOdO, rcp_dOdO, v0, v1);
+        vbool<M> validCone = valid;
+        cone.intersectCone(validCone, t_cone_lower, t_cone_upper);
+
+        valid &= (validCone | (cone.g <= 0.0f));  // if cone is entirely in sphere end - check sphere
+        if (unlikely(none(valid)))
+          return false;
+        
+        /* cone hits inside the neighboring capped cones are inside the geometry and thus ignored */
+        const ConeGeometry<M> coneL (v0, vL);
+        const ConeGeometry<M> coneR (v1, vR);
+#if !defined(EMBREE_BACKFACE_CULLING_CURVES)
+        const Vec3vf<M> hit_lower = ray_org + t_cone_lower*ray_dir;
+        const Vec3vf<M> hit_upper = ray_org + t_cone_upper*ray_dir;
+        t_cone_lower = select (!coneL.isInsideCappedCone (validCone, hit_lower) & !coneR.isInsideCappedCone (validCone, hit_lower), t_cone_lower, vfloat<M>(pos_inf));
+        t_cone_upper = select (!coneL.isInsideCappedCone (validCone, hit_upper) & !coneR.isInsideCappedCone (validCone, hit_upper), t_cone_upper, vfloat<M>(neg_inf));
+#endif
+
+        /* intersect ending sphere */
+        vfloat<M> t_sph1_lower, t_sph1_upper;
+        vfloat<M> t_sph0_lower = vfloat<M>(pos_inf);
+        vfloat<M> t_sph0_upper = vfloat<M>(neg_inf);
+        cone.intersectEndSphere(valid, coneR, t_sph1_lower, t_sph1_upper);
+
+        const vbool<M> isBeginPoint = valid & (vL[0] == vfloat<M>(pos_inf));
+        if (unlikely(any(isBeginPoint))) {
+          cone.intersectBeginSphere (isBeginPoint, t_sph0_lower, t_sph0_upper);
+        }
+        
+        /* CSG union of cone and end sphere */
+        vfloat<M> t_sph_lower = min(t_sph0_lower, t_sph1_lower);
+        vfloat<M> t_cone_sphere_lower = min(t_cone_lower, t_sph_lower);
+#if !defined (EMBREE_BACKFACE_CULLING_CURVES)
+        vfloat<M> t_sph_upper = max(t_sph0_upper, t_sph1_upper);
+        vfloat<M> t_cone_sphere_upper = max(t_cone_upper, t_sph_upper);
+        
+        /* filter out hits that are not in tnear/tfar range */
+        const vbool<M> valid_lower = valid & ray_tnear <= dt+t_cone_sphere_lower & dt+t_cone_sphere_lower <= ray_tfar() & t_cone_sphere_lower != vfloat<M>(pos_inf);
+        const vbool<M> valid_upper = valid & ray_tnear <= dt+t_cone_sphere_upper & dt+t_cone_sphere_upper <= ray_tfar() & t_cone_sphere_upper != vfloat<M>(neg_inf);
+        
+        /* check if there is a first hit */
+        const vbool<M> valid_first = valid_lower | valid_upper;
+        if (unlikely(none(valid_first)))
+          return false;
+        
+        /* construct first hit */
+        const vfloat<M> t_first = select(valid_lower, t_cone_sphere_lower, t_cone_sphere_upper);
+        const vbool<M> cone_hit_first = t_first == t_cone_lower | t_first == t_cone_upper;
+        const vbool<M> sph0_hit_first = t_first == t_sph0_lower | t_first == t_sph0_upper;
+        const Vec3vf<M> Ng_first = select(cone_hit_first, cone.Ng_cone(valid_lower), select (sph0_hit_first, cone.Ng_sphere0(valid_lower), cone.Ng_sphere1(valid_lower)));
+        const vfloat<M> u_first  = select(cone_hit_first, cone.u_cone(valid_lower), select (sph0_hit_first, vfloat<M>(zero), vfloat<M>(one)));
+
+        /* invoke intersection filter for first hit */
+        RoundLineIntersectorHitM<M> hit(u_first,zero,dt+t_first,Ng_first);
+        const bool is_hit_first = epilog(valid_first, hit);
+        
+        /* check for possible second hits before potentially accepted hit */
+        const vfloat<M> t_second = t_cone_sphere_upper;
+        const vbool<M> valid_second = valid_lower & valid_upper & (dt+t_cone_sphere_upper <= ray_tfar());
+        if (unlikely(none(valid_second)))
+          return is_hit_first;
+        
+        /* invoke intersection filter for second hit */
+        const vbool<M> cone_hit_second = t_second == t_cone_lower | t_second == t_cone_upper;
+        const vbool<M> sph0_hit_second = t_second == t_sph0_lower | t_second == t_sph0_upper;
+        const Vec3vf<M> Ng_second = select(cone_hit_second, cone.Ng_cone(false), select (sph0_hit_second, cone.Ng_sphere0(false), cone.Ng_sphere1(false)));
+        const vfloat<M> u_second  = select(cone_hit_second, cone.u_cone(false), select (sph0_hit_second, vfloat<M>(zero), vfloat<M>(one)));
+
+        hit = RoundLineIntersectorHitM<M>(u_second,zero,dt+t_second,Ng_second);
+        const bool is_hit_second = epilog(valid_second, hit);
+        
+        return is_hit_first | is_hit_second;
+#else
+        /* filter out hits that are not in tnear/tfar range */
+        const vbool<M> valid_lower = valid & ray_tnear <= dt+t_cone_sphere_lower & dt+t_cone_sphere_lower <= ray_tfar() & t_cone_sphere_lower != vfloat<M>(pos_inf);
+        
+        /* check if there is a valid hit */
+        if (unlikely(none(valid_lower)))
+          return false;
+        
+        /* construct first hit */
+        const vbool<M> cone_hit_first = t_cone_sphere_lower == t_cone_lower | t_cone_sphere_lower == t_cone_upper;
+        const vbool<M> sph0_hit_first = t_cone_sphere_lower == t_sph0_lower | t_cone_sphere_lower == t_sph0_upper;
+        const Vec3vf<M> Ng_first = select(cone_hit_first, cone.Ng_cone(valid_lower), select (sph0_hit_first, cone.Ng_sphere0(valid_lower), cone.Ng_sphere1(valid_lower)));
+        const vfloat<M> u_first  = select(cone_hit_first, cone.u_cone(valid_lower), select (sph0_hit_first, vfloat<M>(zero), vfloat<M>(one)));
+
+        /* invoke intersection filter for first hit */
+        RoundLineIntersectorHitM<M> hit(u_first,zero,dt+t_cone_sphere_lower,Ng_first);
+        const bool is_hit_first = epilog(valid_lower, hit);
+        
+        return is_hit_first;
+#endif
+      }
+      
+    } // end namespace __roundline_internal
+    
+    template<int M>
+      struct RoundLinearCurveIntersector1
+      {
+        typedef CurvePrecalculations1 Precalculations;
+
+        template<typename Ray>
+        struct ray_tfar {
+          Ray& ray;
+          __forceinline ray_tfar(Ray& ray) : ray(ray) {}
+          __forceinline vfloat<M> operator() () const { return ray.tfar; };
+        };
+	
+        template<typename Ray, typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            Ray& ray,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const Vec4vf<M>& vLi, const Vec4vf<M>& vRi,
+                                            const Epilog& epilog)
+        {
+          const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+          const Vec3vf<M> ray_dir(ray.dir.x, ray.dir.y, ray.dir.z);
+          const vfloat<M> ray_tnear(ray.tnear());
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          const Vec4vf<M> vL = enlargeRadiusToMinWidth<M>(context,geom,ray_org,vLi);
+          const Vec4vf<M> vR = enlargeRadiusToMinWidth<M>(context,geom,ray_org,vRi);
+          return  __roundline_internal::intersectConeSphere<M>(valid_i,ray_org,ray_dir,ray_tnear,ray_tfar<Ray>(ray),v0,v1,vL,vR,epilog);
+        }
+      };
+    
+    template<int M, int K>
+      struct RoundLinearCurveIntersectorK
+      {
+        typedef CurvePrecalculationsK<K> Precalculations;
+        
+        struct ray_tfar {
+          RayK<K>& ray;
+          size_t k;
+          __forceinline ray_tfar(RayK<K>& ray, size_t k) : ray(ray), k(k) {}
+          __forceinline vfloat<M> operator() () const { return ray.tfar[k]; };
+        };
+        
+        template<typename Epilog>
+        static __forceinline bool intersect(const vbool<M>& valid_i,
+                                            RayK<K>& ray, size_t k,
+                                            RayQueryContext* context,
+                                            const LineSegments* geom,
+                                            const Precalculations& pre,
+                                            const Vec4vf<M>& v0i, const Vec4vf<M>& v1i,
+                                            const Vec4vf<M>& vLi, const Vec4vf<M>& vRi,
+                                            const Epilog& epilog)
+        {
+          const Vec3vf<M> ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+          const Vec3vf<M> ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]);
+          const vfloat<M> ray_tnear = ray.tnear()[k];
+          const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+          const Vec4vf<M> v1 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v1i);
+          const Vec4vf<M> vL = enlargeRadiusToMinWidth<M>(context,geom,ray_org,vLi);
+          const Vec4vf<M> vR = enlargeRadiusToMinWidth<M>(context,geom,ray_org,vRi);
+          return __roundline_internal::intersectConeSphere<M>(valid_i,ray_org,ray_dir,ray_tnear,ray_tfar(ray,k),v0,v1,vL,vR,epilog);
+        }
+      };
+  }
+}

thirdparty/embree/kernels/geometry/roundlinei_intersector.h

@@ -0,0 +1,123 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "roundline_intersector.h"
+#include "intersector_epilog.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, bool filter>
+    struct RoundLinearCurveMiIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom);
+        const vbool<M> valid = line.valid();
+        RoundLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,vL,vR,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom);
+        const vbool<M> valid = line.valid();
+        return RoundLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,vL,vR,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, bool filter>
+    struct RoundLinearCurveMiMBIntersector1
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        RoundLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,vL,vR,Intersect1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom,ray.time());
+        const vbool<M> valid = line.valid();
+        return RoundLinearCurveIntersector1<M>::intersect(valid,ray,context,geom,pre,v0,v1,vL,vR,Occluded1EpilogM<M,filter>(ray,context,line.geomID(),line.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& line)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, line);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct RoundLinearCurveMiIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom);
+        const vbool<M> valid = line.valid();
+        RoundLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,vL,vR,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom);
+        const vbool<M> valid = line.valid();
+        return RoundLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,vL,vR,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct RoundLinearCurveMiMBIntersectorK
+    {
+      typedef LineMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context,  const Primitive& line)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        RoundLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,vL,vR,Intersect1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& line)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const LineSegments* geom = context->scene->get<LineSegments>(line.geomID());
+        Vec4vf<M> v0,v1,vL,vR; line.gather(v0,v1,vL,vR,geom,ray.time()[k]);
+        const vbool<M> valid = line.valid();
+        return RoundLinearCurveIntersectorK<M,K>::intersect(valid,ray,k,context,geom,pre,v0,v1,vL,vR,Occluded1KEpilogM<M,K,filter>(ray,k,context,line.geomID(),line.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/sphere_intersector.h

@@ -0,0 +1,235 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/scene_points.h"
+#include "curve_intersector_precalculations.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+    struct SphereIntersectorHitM
+    {
+      __forceinline SphereIntersectorHitM() {}
+
+      __forceinline SphereIntersectorHitM(const vfloat<M>& t, const Vec3vf<M>& Ng)
+        : vt(t), vNg(Ng) {}
+
+      __forceinline void finalize() {}
+
+      __forceinline Vec2f uv(const size_t i) const {
+        return Vec2f(0.0f, 0.0f);
+      }
+      __forceinline Vec2vf<M> uv() const {
+        return Vec2vf<M>(0.0f, 0.0f);
+      }
+      __forceinline float t(const size_t i) const {
+        return vt[i];
+      }
+      __forceinline vfloat<M> t() const {
+        return vt;
+      }
+      __forceinline Vec3fa Ng(const size_t i) const {
+        return Vec3fa(vNg.x[i], vNg.y[i], vNg.z[i]);
+      }
+      __forceinline Vec3vf<M> Ng() const { 
+        return vNg;
+      }
+
+     public:
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+    };
+
+    template<>
+    struct SphereIntersectorHitM<1>
+    {
+      __forceinline SphereIntersectorHitM() {}
+
+      __forceinline SphereIntersectorHitM(const float& t, const Vec3f& Ng)
+        : vt(t), vNg(Ng) {}
+
+      __forceinline void finalize() {}
+
+      __forceinline Vec2f uv() const {
+        return Vec2f(0.0f, 0.0f);
+      }
+
+      __forceinline float t() const {
+        return vt;
+      }
+
+      __forceinline Vec3f Ng() const { 
+        return vNg;
+      }
+
+     public:
+      float vt;
+      Vec3f vNg;
+    };
+
+    template<int M>
+    struct SphereIntersector1
+    {
+      typedef CurvePrecalculations1 Precalculations;
+
+      template<typename Ray, typename Epilog>
+      static __forceinline bool intersect(
+          const vbool<M>& valid_i, Ray& ray,
+          const Precalculations& pre, const Vec4vf<M>& v0, const Epilog& epilog)
+      {
+        vbool<M> valid = valid_i;
+
+        const vfloat<M> rd2    = rcp(dot(ray.dir, ray.dir));
+        const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+        const Vec3vf<M> ray_dir(ray.dir.x, ray.dir.y, ray.dir.z);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+
+        const Vec3vf<M> c0     = center - ray_org;
+        const vfloat<M> projC0 = dot(c0, ray_dir) * rd2;
+        const Vec3vf<M> perp   = c0 - projC0 * ray_dir;
+        const vfloat<M> l2     = dot(perp, perp);
+        const vfloat<M> r2     = radius * radius;
+        valid &= (l2 <= r2);
+        if (unlikely(none(valid)))
+          return false;
+
+        const vfloat<M> td      = sqrt((r2 - l2) * rd2);
+        const vfloat<M> t_front = projC0 - td;
+        const vfloat<M> t_back  = projC0 + td;
+
+        const vbool<M> valid_front = valid & (ray.tnear() <= t_front) & (t_front <= ray.tfar);
+
+        /* check if there is a first hit */
+#if defined (EMBREE_BACKFACE_CULLING_SPHERES)
+        /* check if there is a first hit */
+        const vbool<M> valid_first = valid_front;
+#else
+        const vbool<M> valid_back  = valid & (ray.tnear() <= t_back ) & (t_back  <= ray.tfar);
+        const vbool<M> valid_first = valid_front | valid_back;
+#endif
+        if (unlikely(none(valid_first)))
+          return false;
+
+        /* construct first hit */
+        const vfloat<M> td_front = -td;
+        const vfloat<M> td_back  = +td;
+        const vfloat<M> t_first  = select(valid_front, t_front, t_back);
+        const Vec3vf<M> Ng_first = select(valid_front, td_front, td_back) * ray_dir - perp;
+        SphereIntersectorHitM<M> hit(t_first, Ng_first);
+
+        /* invoke intersection filter for first hit */
+        const bool is_hit_first = epilog(valid_first, hit);
+
+#if defined (EMBREE_BACKFACE_CULLING_SPHERES)
+        return is_hit_first;
+#else
+        /* check for possible second hits before potentially accepted hit */
+        const vfloat<M> t_second = t_back;
+        const vbool<M> valid_second = valid_front & valid_back & (t_second <= ray.tfar);
+        if (unlikely(none(valid_second)))
+          return is_hit_first;
+
+        /* invoke intersection filter for second hit */
+        const Vec3vf<M> Ng_second = td_back * ray_dir - perp;
+        hit = SphereIntersectorHitM<M> (t_second, Ng_second);
+        const bool is_hit_second = epilog(valid_second, hit);
+        return is_hit_first | is_hit_second;
+#endif
+      }
+
+      template<typename Epilog>
+      static __forceinline bool intersect(
+        const vbool<M>& valid_i, Ray& ray, RayQueryContext* context, const Points* geom,
+        const Precalculations& pre, const Vec4vf<M>& v0i, const Epilog& epilog)
+      {
+        const Vec3vf<M> ray_org(ray.org.x, ray.org.y, ray.org.z);
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        return intersect(valid_i,ray,pre,v0,epilog);
+      }
+    };
+
+    template<int M, int K>
+    struct SphereIntersectorK
+    {
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      template<typename Epilog>
+      static __forceinline bool intersect(const vbool<M>& valid_i,
+                                          RayK<K>& ray, size_t k,
+                                          RayQueryContext* context,
+                                          const Points* geom,
+                                          const Precalculations& pre,
+                                          const Vec4vf<M>& v0i,
+                                          const Epilog& epilog)
+      {
+        vbool<M> valid = valid_i;
+
+        const Vec3vf<M> ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]);
+        const Vec3vf<M> ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]);
+        const vfloat<M> rd2 = rcp(dot(ray_dir, ray_dir));
+
+        const Vec4vf<M> v0 = enlargeRadiusToMinWidth<M>(context,geom,ray_org,v0i);
+        const Vec3vf<M> center = v0.xyz();
+        const vfloat<M> radius = v0.w;
+
+        const Vec3vf<M> c0     = center - ray_org;
+        const vfloat<M> projC0 = dot(c0, ray_dir) * rd2;
+        const Vec3vf<M> perp   = c0 - projC0 * ray_dir;
+        const vfloat<M> l2     = dot(perp, perp);
+        const vfloat<M> r2     = radius * radius;
+        valid &= (l2 <= r2);
+        if (unlikely(none(valid)))
+          return false;
+
+        const vfloat<M> td      = sqrt((r2 - l2) * rd2);
+        const vfloat<M> t_front = projC0 - td;
+        const vfloat<M> t_back  = projC0 + td;
+
+        const vbool<M> valid_front = valid & (ray.tnear()[k] <= t_front) & (t_front <= ray.tfar[k]);
+
+        /* check if there is a first hit */
+#if defined (EMBREE_BACKFACE_CULLING_SPHERES)
+        const vbool<M> valid_first = valid_front;
+#else
+        const vbool<M> valid_back  = valid & (ray.tnear()[k] <= t_back ) & (t_back  <= ray.tfar[k]);
+        const vbool<M> valid_first = valid_front | valid_back;
+#endif
+        if (unlikely(none(valid_first)))
+          return false;
+
+        /* construct first hit */
+        const vfloat<M> td_front = -td;
+        const vfloat<M> td_back  = +td;
+        const vfloat<M> t_first  = select(valid_front, t_front, t_back);
+        const Vec3vf<M> Ng_first = select(valid_front, td_front, td_back) * ray_dir - perp;
+        SphereIntersectorHitM<M> hit(t_first, Ng_first);
+
+        /* invoke intersection filter for first hit */
+        const bool is_hit_first = epilog(valid_first, hit);
+
+#if defined (EMBREE_BACKFACE_CULLING_SPHERES)
+        return is_hit_first;
+#else
+        /* check for possible second hits before potentially accepted hit */
+        const vfloat<M> t_second = t_back;
+        const vbool<M> valid_second = valid_front & valid_back & (t_second <= ray.tfar[k]);
+        if (unlikely(none(valid_second)))
+          return is_hit_first;
+
+        /* invoke intersection filter for second hit */
+        const Vec3vf<M> Ng_second = td_back * ray_dir - perp;
+        hit = SphereIntersectorHitM<M> (t_second, Ng_second);
+        const bool is_hit_second = epilog(valid_second, hit);
+        
+        return is_hit_first | is_hit_second;
+#endif
+      }
+    };
+  }  // namespace isa
+}  // namespace embree

thirdparty/embree/kernels/geometry/spherei_intersector.h

@@ -0,0 +1,156 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "intersector_epilog.h"
+#include "pointi.h"
+#include "sphere_intersector.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, bool filter>
+    struct SphereMiIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& sphere)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom);
+        const vbool<M> valid = sphere.valid();
+        SphereIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Intersect1EpilogM<M, filter>(ray, context, sphere.geomID(), sphere.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& sphere)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom);
+        const vbool<M> valid = sphere.valid();
+        return SphereIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Occluded1EpilogM<M, filter>(ray, context, sphere.geomID(), sphere.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query,
+                                           PointQueryContext* context,
+                                           const Primitive& sphere)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, sphere);
+      }
+    };
+
+    template<int M, bool filter>
+    struct SphereMiMBIntersector1
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculations1 Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre,
+                                          RayHit& ray,
+                                          RayQueryContext* context,
+                                          const Primitive& sphere)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom, ray.time());
+        const vbool<M> valid = sphere.valid();
+        SphereIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Intersect1EpilogM<M, filter>(ray, context, sphere.geomID(), sphere.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre,
+                                         Ray& ray,
+                                         RayQueryContext* context,
+                                         const Primitive& sphere)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom, ray.time());
+        const vbool<M> valid = sphere.valid();
+        return SphereIntersector1<M>::intersect(
+          valid, ray, context, geom, pre, v0, Occluded1EpilogM<M, filter>(ray, context, sphere.geomID(), sphere.primID()));
+      }
+
+      static __forceinline bool pointQuery(PointQuery* query,
+                                           PointQueryContext* context,
+                                           const Primitive& sphere)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, sphere);
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct SphereMiIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& sphere)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom);
+        const vbool<M> valid = sphere.valid();
+        SphereIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Intersect1KEpilogM<M, K, filter>(ray, k, context, sphere.geomID(), sphere.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& sphere)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom);
+        const vbool<M> valid = sphere.valid();
+        return SphereIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Occluded1KEpilogM<M, K, filter>(ray, k, context, sphere.geomID(), sphere.primID()));
+      }
+    };
+
+    template<int M, int K, bool filter>
+    struct SphereMiMBIntersectorK
+    {
+      typedef PointMi<M> Primitive;
+      typedef CurvePrecalculationsK<K> Precalculations;
+
+      static __forceinline void intersect(
+          const Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& sphere)
+      {
+        STAT3(normal.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom, ray.time()[k]);
+        const vbool<M> valid = sphere.valid();
+        SphereIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Intersect1KEpilogM<M, K, filter>(ray, k, context, sphere.geomID(), sphere.primID()));
+      }
+
+      static __forceinline bool occluded(
+          const Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& sphere)
+      {
+        STAT3(shadow.trav_prims, 1, 1, 1);
+        const Points* geom = context->scene->get<Points>(sphere.geomID());
+        Vec4vf<M> v0; sphere.gather(v0, geom, ray.time()[k]);
+        const vbool<M> valid = sphere.valid();
+        return SphereIntersectorK<M, K>::intersect(
+          valid, ray, k, context, geom, pre, v0,
+          Occluded1KEpilogM<M, K, filter>(ray, k, context, sphere.geomID(), sphere.primID()));
+      }
+    };
+  }  // namespace isa
+}  // namespace embree

thirdparty/embree/kernels/geometry/subdivpatch1.h

@@ -0,0 +1,38 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../geometry/primitive.h"
+#include "../subdiv/subdivpatch1base.h"
+
+namespace embree
+{
+
+  struct __aligned(64) SubdivPatch1 : public SubdivPatch1Base
+  {
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    
+    static Type type;
+
+  public:
+
+    /*! constructor for cached subdiv patch */
+    SubdivPatch1 (const unsigned int gID,
+                        const unsigned int pID,
+                        const unsigned int subPatch,
+                        const SubdivMesh *const mesh,
+                        const size_t time,
+                        const Vec2f uv[4],
+                        const float edge_level[4],
+                        const int subdiv[4],
+                        const int simd_width) 
+      : SubdivPatch1Base(gID,pID,subPatch,mesh,time,uv,edge_level,subdiv,simd_width) {}
+  };
+}

thirdparty/embree/kernels/geometry/subdivpatch1_intersector.h

@@ -0,0 +1,237 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "subdivpatch1.h"
+#include "grid_soa.h"
+#include "grid_soa_intersector1.h"
+#include "grid_soa_intersector_packet.h"
+#include "../common/ray.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<typename T>
+      class SubdivPatch1Precalculations : public T
+    { 
+    public:
+      __forceinline SubdivPatch1Precalculations (const Ray& ray, const void* ptr)
+        : T(ray,ptr) {}
+    };
+
+    template<int K, typename T>
+      class SubdivPatch1PrecalculationsK : public T
+    { 
+    public:
+      __forceinline SubdivPatch1PrecalculationsK (const vbool<K>& valid, RayK<K>& ray)
+        : T(valid,ray) {}
+    };
+
+    class SubdivPatch1Intersector1
+    {
+    public:
+      typedef GridSOA Primitive;
+      typedef SubdivPatch1Precalculations<GridSOAIntersector1::Precalculations> Precalculations;
+
+      static __forceinline bool processLazyNode(Precalculations& pre, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        lazy_node = prim->root(0);
+        pre.grid = (Primitive*)prim;
+        return false;
+      }
+
+      /*! Intersect a ray with the primitive. */
+      template<int N, bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) 
+      {
+        if (likely(ty == 0)) GridSOAIntersector1::intersect(pre,ray,context,prim,lazy_node);
+        else                 processLazyNode(pre,context,prim,lazy_node);
+      }
+
+      template<int N, bool robust>
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) {
+        intersect(This,pre,ray,context,prim,ty,tray,lazy_node);
+      }
+      
+      /*! Test if the ray is occluded by the primitive */
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAIntersector1::occluded(pre,ray,context,prim,lazy_node);
+        else                 return processLazyNode(pre,context,prim,lazy_node);
+      }
+
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) {
+        return occluded(This,pre,ray,context,prim,ty,tray,lazy_node);
+      }
+      
+      template<int N>
+        static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t ty, const TravPointQuery<N> &tquery, size_t& lazy_node) 
+      {
+          // TODO: PointQuery implement
+          assert(false && "not implemented");
+          return false;
+      }
+
+      template<int N>
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravPointQuery<N> &tquery, size_t& lazy_node) {
+        return pointQuery(This,query,context,prim,ty,tquery,lazy_node);
+      }
+    };
+
+    class SubdivPatch1MBIntersector1
+    {
+    public:
+      typedef SubdivPatch1 Primitive;
+      typedef GridSOAMBIntersector1::Precalculations Precalculations;
+      
+      static __forceinline bool processLazyNode(Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim_i, size_t& lazy_node)
+      {
+        Primitive* prim = (Primitive*) prim_i;
+        GridSOA* grid = nullptr;
+        grid = (GridSOA*) prim->root_ref.get();
+        pre.itime = getTimeSegment(ray.time(), float(grid->time_steps-1), pre.ftime);
+        lazy_node = grid->root(pre.itime);
+        pre.grid = grid;
+        return false;
+      }
+
+      /*! Intersect a ray with the primitive. */
+      template<int N, bool robust>
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) 
+      {
+        if (likely(ty == 0)) GridSOAMBIntersector1::intersect(pre,ray,context,prim,lazy_node);
+        else                 processLazyNode(pre,ray,context,prim,lazy_node);
+      }
+
+      template<int N, bool robust>
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) {
+        intersect(This,pre,ray,context,prim,ty,tray,lazy_node);
+      }
+      
+      /*! Test if the ray is occluded by the primitive */
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAMBIntersector1::occluded(pre,ray,context,prim,lazy_node);
+        else                 return processLazyNode(pre,ray,context,prim,lazy_node);
+      }
+
+      template<int N, bool robust>
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node) {
+        return occluded(This,pre,ray,context,prim,ty,tray,lazy_node);
+      }
+      
+      template<int N>
+        static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t ty, const TravPointQuery<N> &tquery, size_t& lazy_node) 
+      {
+          // TODO: PointQuery implement
+          assert(false && "not implemented");
+          return false;
+      }
+
+      template<int N, bool robust>
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, size_t ty0, const Primitive* prim, size_t ty, const TravPointQuery<N> &tquery, size_t& lazy_node) {
+        return pointQuery(This,query,context,prim,ty,tquery,lazy_node);
+      }
+    };
+
+    template <int K>
+      struct SubdivPatch1IntersectorK
+    {
+      typedef GridSOA Primitive;
+      typedef SubdivPatch1PrecalculationsK<K,typename GridSOAIntersectorK<K>::Precalculations> Precalculations;
+      
+      static __forceinline bool processLazyNode(Precalculations& pre, RayQueryContext* context, const Primitive* prim, size_t& lazy_node)
+      {
+        lazy_node = prim->root(0);
+        pre.grid = (Primitive*)prim;
+        return false;
+      }
+      
+      template<bool robust>        
+      static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) GridSOAIntersectorK<K>::intersect(valid,pre,ray,context,prim,lazy_node);
+        else                 processLazyNode(pre,context,prim,lazy_node);
+      }
+      
+      template<bool robust>        
+      static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAIntersectorK<K>::occluded(valid,pre,ray,context,prim,lazy_node);
+        else                 return processLazyNode(pre,context,prim,lazy_node);
+      }
+      
+      template<int N, bool robust>              
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) GridSOAIntersectorK<K>::intersect(pre,ray,k,context,prim,lazy_node);
+        else                 processLazyNode(pre,context,prim,lazy_node);
+      }
+      
+      template<int N, bool robust>              
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAIntersectorK<K>::occluded(pre,ray,k,context,prim,lazy_node);
+        else                 return processLazyNode(pre,context,prim,lazy_node);
+      }
+    };
+
+    typedef SubdivPatch1IntersectorK<4>  SubdivPatch1Intersector4;
+    typedef SubdivPatch1IntersectorK<8>  SubdivPatch1Intersector8;
+    typedef SubdivPatch1IntersectorK<16> SubdivPatch1Intersector16;
+
+    template <int K>
+      struct SubdivPatch1MBIntersectorK
+    {
+      typedef SubdivPatch1 Primitive;
+      //typedef GridSOAMBIntersectorK<K>::Precalculations Precalculations;
+      typedef SubdivPatch1PrecalculationsK<K,typename GridSOAMBIntersectorK<K>::Precalculations> Precalculations;
+      
+      static __forceinline bool processLazyNode(Precalculations& pre, RayQueryContext* context, const Primitive* prim_i, size_t& lazy_node)
+      {
+        Primitive* prim = (Primitive*) prim_i;
+        GridSOA* grid = (GridSOA*) prim->root_ref.get();
+        lazy_node = grid->troot;
+        pre.grid = grid;
+        return false;
+      }
+
+      template<bool robust>
+      static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) GridSOAMBIntersectorK<K>::intersect(valid,pre,ray,context,prim,lazy_node);
+        else                 processLazyNode(pre,context,prim,lazy_node);
+      }
+
+      template<bool robust>
+      static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRayK<K, robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAMBIntersectorK<K>::occluded(valid,pre,ray,context,prim,lazy_node);
+        else                 return processLazyNode(pre,context,prim,lazy_node);
+      }
+      
+      template<int N, bool robust>      
+      static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) GridSOAMBIntersectorK<K>::intersect(pre,ray,k,context,prim,lazy_node);
+        else                 processLazyNode(pre,context,prim,lazy_node);
+      }
+      
+      template<int N, bool robust>      
+      static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t ty, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        if (likely(ty == 0)) return GridSOAMBIntersectorK<K>::occluded(pre,ray,k,context,prim,lazy_node);
+        else                 return processLazyNode(pre,context,prim,lazy_node);
+      }
+    };
+
+    typedef SubdivPatch1MBIntersectorK<4>  SubdivPatch1MBIntersector4;
+    typedef SubdivPatch1MBIntersectorK<8>  SubdivPatch1MBIntersector8;
+    typedef SubdivPatch1MBIntersectorK<16> SubdivPatch1MBIntersector16;
+  }
+}

thirdparty/embree/kernels/geometry/subgrid.h

@@ -0,0 +1,520 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "../common/ray.h"
+#include "../common/scene_grid_mesh.h"
+#include "../bvh/bvh.h"
+
+namespace embree
+{
+    /* Stores M quads from an indexed face set */
+      struct SubGrid
+      {
+        /* Virtual interface to query information about the quad type */
+        struct Type : public PrimitiveType
+        {
+          const char* name() const;
+          size_t sizeActive(const char* This) const;
+          size_t sizeTotal(const char* This) const;
+          size_t getBytes(const char* This) const;
+        };
+        static Type type;
+
+      public:
+
+        /* primitive supports multiple time segments */
+        static const bool singleTimeSegment = false;
+
+        /* Returns maximum number of stored quads */
+        static __forceinline size_t max_size() { return 1; }
+
+        /* Returns required number of primitive blocks for N primitives */
+        static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+
+      public:
+
+        /* Default constructor */
+        __forceinline SubGrid() {  }
+
+        /* Construction from vertices and IDs */
+        __forceinline SubGrid(const unsigned int x,
+                              const unsigned int y,
+                              const unsigned int geomID,
+                              const unsigned int primID)
+          : _x(x), _y(y), _geomID(geomID), _primID(primID)
+        {
+        }
+
+        __forceinline bool invalid3x3X() const { return (unsigned int)_x & (1<<15); }
+        __forceinline bool invalid3x3Y() const { return (unsigned int)_y & (1<<15); }
+
+        /* Gather the quads */
+        __forceinline void gather(Vec3vf4& p0,
+                                  Vec3vf4& p1,
+                                  Vec3vf4& p2,
+                                  Vec3vf4& p3,
+                                  const GridMesh* const mesh,
+                                  const GridMesh::Grid &g) const
+        {
+          /* first quad always valid */
+          const size_t vtxID00 = g.startVtxID + x() + y() * g.lineVtxOffset;
+          const size_t vtxID01 = vtxID00 + 1;
+          const vfloat4 vtx00  = vfloat4::loadu(mesh->vertexPtr(vtxID00));
+          const vfloat4 vtx01  = vfloat4::loadu(mesh->vertexPtr(vtxID01));
+          const size_t vtxID10 = vtxID00 + g.lineVtxOffset;
+          const size_t vtxID11 = vtxID01 + g.lineVtxOffset;
+          const vfloat4 vtx10  = vfloat4::loadu(mesh->vertexPtr(vtxID10));
+          const vfloat4 vtx11  = vfloat4::loadu(mesh->vertexPtr(vtxID11));
+
+          /* deltaX => vtx02, vtx12 */
+          const size_t deltaX  = invalid3x3X() ? 0 : 1;
+          const size_t vtxID02 = vtxID01 + deltaX;       
+          const vfloat4 vtx02  = vfloat4::loadu(mesh->vertexPtr(vtxID02));
+          const size_t vtxID12 = vtxID11 + deltaX;       
+          const vfloat4 vtx12  = vfloat4::loadu(mesh->vertexPtr(vtxID12));
+
+          /* deltaY => vtx20, vtx21 */
+          const size_t deltaY  = invalid3x3Y() ? 0 : g.lineVtxOffset;
+          const size_t vtxID20 = vtxID10 + deltaY;
+          const size_t vtxID21 = vtxID11 + deltaY;
+          const vfloat4 vtx20  = vfloat4::loadu(mesh->vertexPtr(vtxID20));
+          const vfloat4 vtx21  = vfloat4::loadu(mesh->vertexPtr(vtxID21));
+
+          /* deltaX/deltaY => vtx22 */
+          const size_t vtxID22 = vtxID11 + deltaX + deltaY;       
+          const vfloat4 vtx22  = vfloat4::loadu(mesh->vertexPtr(vtxID22));
+
+          transpose(vtx00,vtx01,vtx11,vtx10,p0.x,p0.y,p0.z);
+          transpose(vtx01,vtx02,vtx12,vtx11,p1.x,p1.y,p1.z);
+          transpose(vtx11,vtx12,vtx22,vtx21,p2.x,p2.y,p2.z);
+          transpose(vtx10,vtx11,vtx21,vtx20,p3.x,p3.y,p3.z);                    
+        }
+
+        template<typename T>
+        __forceinline vfloat4 getVertexMB(const GridMesh* const mesh, const size_t offset, const size_t itime, const float ftime) const
+        {
+          const T v0 = T::loadu(mesh->vertexPtr(offset,itime+0));
+          const T v1 = T::loadu(mesh->vertexPtr(offset,itime+1));
+          return lerp(v0,v1,ftime);
+        }
+
+        /* Gather the quads */
+        __forceinline void gatherMB(Vec3vf4& p0,
+                                    Vec3vf4& p1,
+                                    Vec3vf4& p2,
+                                    Vec3vf4& p3,
+                                    const GridMesh* const mesh,
+                                    const GridMesh::Grid &g,
+                                    const size_t itime, 
+                                    const float ftime) const
+        {
+          /* first quad always valid */
+          const size_t vtxID00 = g.startVtxID + x() + y() * g.lineVtxOffset;
+          const size_t vtxID01 = vtxID00 + 1;
+          const vfloat4 vtx00  = getVertexMB<vfloat4>(mesh,vtxID00,itime,ftime);
+          const vfloat4 vtx01  = getVertexMB<vfloat4>(mesh,vtxID01,itime,ftime);
+          const size_t vtxID10 = vtxID00 + g.lineVtxOffset;
+          const size_t vtxID11 = vtxID01 + g.lineVtxOffset;
+          const vfloat4 vtx10  = getVertexMB<vfloat4>(mesh,vtxID10,itime,ftime);
+          const vfloat4 vtx11  = getVertexMB<vfloat4>(mesh,vtxID11,itime,ftime);
+
+          /* deltaX => vtx02, vtx12 */
+          const size_t deltaX  = invalid3x3X() ? 0 : 1;
+          const size_t vtxID02 = vtxID01 + deltaX;       
+          const vfloat4 vtx02  = getVertexMB<vfloat4>(mesh,vtxID02,itime,ftime);
+          const size_t vtxID12 = vtxID11 + deltaX;       
+          const vfloat4 vtx12  = getVertexMB<vfloat4>(mesh,vtxID12,itime,ftime);
+
+          /* deltaY => vtx20, vtx21 */
+          const size_t deltaY  = invalid3x3Y() ? 0 : g.lineVtxOffset;
+          const size_t vtxID20 = vtxID10 + deltaY;
+          const size_t vtxID21 = vtxID11 + deltaY;
+          const vfloat4 vtx20  = getVertexMB<vfloat4>(mesh,vtxID20,itime,ftime);
+          const vfloat4 vtx21  = getVertexMB<vfloat4>(mesh,vtxID21,itime,ftime);
+
+          /* deltaX/deltaY => vtx22 */
+          const size_t vtxID22 = vtxID11 + deltaX + deltaY;       
+          const vfloat4 vtx22  = getVertexMB<vfloat4>(mesh,vtxID22,itime,ftime);
+
+          transpose(vtx00,vtx01,vtx11,vtx10,p0.x,p0.y,p0.z);
+          transpose(vtx01,vtx02,vtx12,vtx11,p1.x,p1.y,p1.z);
+          transpose(vtx11,vtx12,vtx22,vtx21,p2.x,p2.y,p2.z);
+          transpose(vtx10,vtx11,vtx21,vtx20,p3.x,p3.y,p3.z);                    
+        }
+
+
+
+        /* Gather the quads */
+        __forceinline void gather(Vec3vf4& p0,
+                                  Vec3vf4& p1,
+                                  Vec3vf4& p2,
+                                  Vec3vf4& p3,
+                                  const Scene *const scene) const
+        {
+          const GridMesh* const mesh = scene->get<GridMesh>(geomID());
+          const GridMesh::Grid &g    = mesh->grid(primID());
+          gather(p0,p1,p2,p3,mesh,g);
+        }
+
+        /* Gather the quads in the motion blur case */
+        __forceinline void gatherMB(Vec3vf4& p0,
+                                    Vec3vf4& p1,
+                                    Vec3vf4& p2,
+                                    Vec3vf4& p3,
+                                    const Scene *const scene,
+                                    const size_t itime, 
+                                    const float ftime) const
+        {
+          const GridMesh* const mesh = scene->get<GridMesh>(geomID());
+          const GridMesh::Grid &g    = mesh->grid(primID());
+          gatherMB(p0,p1,p2,p3,mesh,g,itime,ftime);
+        }
+
+        /* Gather the quads */
+        __forceinline void gather(Vec3fa vtx[16], const Scene *const scene) const
+        {
+          const GridMesh* mesh     = scene->get<GridMesh>(geomID());
+          const GridMesh::Grid &g  = mesh->grid(primID());
+
+          /* first quad always valid */
+          const size_t vtxID00 = g.startVtxID + x() + y() * g.lineVtxOffset;
+          const size_t vtxID01 = vtxID00 + 1;
+          const Vec3fa vtx00  = Vec3fa::loadu(mesh->vertexPtr(vtxID00));
+          const Vec3fa vtx01  = Vec3fa::loadu(mesh->vertexPtr(vtxID01));
+          const size_t vtxID10 = vtxID00 + g.lineVtxOffset;
+          const size_t vtxID11 = vtxID01 + g.lineVtxOffset;
+          const Vec3fa vtx10  = Vec3fa::loadu(mesh->vertexPtr(vtxID10));
+          const Vec3fa vtx11  = Vec3fa::loadu(mesh->vertexPtr(vtxID11));
+
+          /* deltaX => vtx02, vtx12 */
+          const size_t deltaX  = invalid3x3X() ? 0 : 1;
+          const size_t vtxID02 = vtxID01 + deltaX;       
+          const Vec3fa vtx02  = Vec3fa::loadu(mesh->vertexPtr(vtxID02));
+          const size_t vtxID12 = vtxID11 + deltaX;       
+          const Vec3fa vtx12  = Vec3fa::loadu(mesh->vertexPtr(vtxID12));
+
+          /* deltaY => vtx20, vtx21 */
+          const size_t deltaY  = invalid3x3Y() ? 0 : g.lineVtxOffset;
+          const size_t vtxID20 = vtxID10 + deltaY;
+          const size_t vtxID21 = vtxID11 + deltaY;
+          const Vec3fa vtx20  = Vec3fa::loadu(mesh->vertexPtr(vtxID20));
+          const Vec3fa vtx21  = Vec3fa::loadu(mesh->vertexPtr(vtxID21));
+
+          /* deltaX/deltaY => vtx22 */
+          const size_t vtxID22 = vtxID11 + deltaX + deltaY;       
+          const Vec3fa vtx22  = Vec3fa::loadu(mesh->vertexPtr(vtxID22));
+
+          vtx[ 0] = vtx00; vtx[ 1] = vtx01; vtx[ 2] = vtx11; vtx[ 3] = vtx10;
+          vtx[ 4] = vtx01; vtx[ 5] = vtx02; vtx[ 6] = vtx12; vtx[ 7] = vtx11;
+          vtx[ 8] = vtx10; vtx[ 9] = vtx11; vtx[10] = vtx21; vtx[11] = vtx20;
+          vtx[12] = vtx11; vtx[13] = vtx12; vtx[14] = vtx22; vtx[15] = vtx21;
+        }
+
+        /* Gather the quads */
+        __forceinline void gatherMB(vfloat4 vtx[16], const Scene *const scene, const size_t itime, const float ftime) const
+        {
+          const GridMesh* mesh     = scene->get<GridMesh>(geomID());
+          const GridMesh::Grid &g  = mesh->grid(primID());
+
+          /* first quad always valid */
+          const size_t vtxID00 = g.startVtxID + x() + y() * g.lineVtxOffset;
+          const size_t vtxID01 = vtxID00 + 1;
+          const vfloat4 vtx00  = getVertexMB<vfloat4>(mesh,vtxID00,itime,ftime);
+          const vfloat4 vtx01  = getVertexMB<vfloat4>(mesh,vtxID01,itime,ftime);
+          const size_t vtxID10 = vtxID00 + g.lineVtxOffset;
+          const size_t vtxID11 = vtxID01 + g.lineVtxOffset;
+          const vfloat4 vtx10  = getVertexMB<vfloat4>(mesh,vtxID10,itime,ftime);
+          const vfloat4 vtx11  = getVertexMB<vfloat4>(mesh,vtxID11,itime,ftime);
+
+          /* deltaX => vtx02, vtx12 */
+          const size_t deltaX  = invalid3x3X() ? 0 : 1;
+          const size_t vtxID02 = vtxID01 + deltaX;       
+          const vfloat4 vtx02  = getVertexMB<vfloat4>(mesh,vtxID02,itime,ftime);
+          const size_t vtxID12 = vtxID11 + deltaX;       
+          const vfloat4 vtx12  = getVertexMB<vfloat4>(mesh,vtxID12,itime,ftime);
+
+          /* deltaY => vtx20, vtx21 */
+          const size_t deltaY  = invalid3x3Y() ? 0 : g.lineVtxOffset;
+          const size_t vtxID20 = vtxID10 + deltaY;
+          const size_t vtxID21 = vtxID11 + deltaY;
+          const vfloat4 vtx20  = getVertexMB<vfloat4>(mesh,vtxID20,itime,ftime);
+          const vfloat4 vtx21  = getVertexMB<vfloat4>(mesh,vtxID21,itime,ftime);
+
+          /* deltaX/deltaY => vtx22 */
+          const size_t vtxID22 = vtxID11 + deltaX + deltaY;       
+          const vfloat4 vtx22  = getVertexMB<vfloat4>(mesh,vtxID22,itime,ftime);
+
+          vtx[ 0] = vtx00; vtx[ 1] = vtx01; vtx[ 2] = vtx11; vtx[ 3] = vtx10;
+          vtx[ 4] = vtx01; vtx[ 5] = vtx02; vtx[ 6] = vtx12; vtx[ 7] = vtx11;
+          vtx[ 8] = vtx10; vtx[ 9] = vtx11; vtx[10] = vtx21; vtx[11] = vtx20;
+          vtx[12] = vtx11; vtx[13] = vtx12; vtx[14] = vtx22; vtx[15] = vtx21;
+        }        
+          
+
+        /* Calculate the bounds of the subgrid */
+        __forceinline const BBox3fa bounds(const Scene *const scene, const size_t itime=0) const
+        {
+          BBox3fa bounds = empty;
+          FATAL("not implemented yet");
+          return bounds;
+        }
+
+        /* Calculate the linear bounds of the primitive */
+        __forceinline LBBox3fa linearBounds(const Scene* const scene, const size_t itime)
+        {
+          return LBBox3fa(bounds(scene,itime+0),bounds(scene,itime+1));
+        }
+
+        __forceinline LBBox3fa linearBounds(const Scene *const scene, size_t itime, size_t numTimeSteps)
+        {
+          LBBox3fa allBounds = empty;
+          FATAL("not implemented yet");
+          return allBounds;
+        }
+
+        __forceinline LBBox3fa linearBounds(const Scene *const scene, const BBox1f time_range)
+        {
+          LBBox3fa allBounds = empty;
+          FATAL("not implemented yet");
+          return allBounds;
+        }
+
+
+        friend embree_ostream operator<<(embree_ostream cout, const SubGrid& sg) {
+          return cout << "SubGrid " << " ( x = " << sg.x() << ", y = " << sg.y() << ", geomID = " << sg.geomID() << ", primID = " << sg.primID() << ", invalid3x3X() " << (int)sg.invalid3x3X() << ", invalid3x3Y() " << (int)sg.invalid3x3Y();
+        }
+
+        __forceinline unsigned int geomID() const { return _geomID; }
+        __forceinline unsigned int primID() const { return _primID; }
+        __forceinline unsigned int x() const { return (unsigned int)_x & 0x7fff; }
+        __forceinline unsigned int y() const { return (unsigned int)_y & 0x7fff; }
+
+      private:
+        unsigned short _x;
+        unsigned short _y;
+        unsigned int _geomID;    // geometry ID of mesh
+        unsigned int _primID;    // primitive ID of primitive inside mesh
+      };
+
+      struct SubGridID {
+        unsigned short x;
+        unsigned short y;
+        unsigned int primID;
+        
+        __forceinline SubGridID() {}
+        __forceinline SubGridID(const unsigned int x, const unsigned int y, const unsigned int primID) :
+        x(x), y(y), primID(primID) {}
+        
+      };
+      
+      /* QuantizedBaseNode as large subgrid leaf */
+      template<int N>
+      struct SubGridQBVHN
+      {
+        /* Virtual interface to query information about the quad type */
+        struct Type : public PrimitiveType
+        {
+          const char* name() const;
+          size_t sizeActive(const char* This) const;
+          size_t sizeTotal(const char* This) const;
+          size_t getBytes(const char* This) const;
+        };
+        static Type type;
+
+      public:
+
+        __forceinline size_t size() const
+        {
+          for (size_t i=0;i<N;i++)
+            if (primID(i) == -1) return i;
+          return N;
+        }
+
+      __forceinline void clear() {
+        for (size_t i=0;i<N;i++)
+          subgridIDs[i] = SubGridID(0,0,(unsigned int)-1);
+        qnode.clear();
+      }
+
+        /* Default constructor */
+        __forceinline SubGridQBVHN() {  }
+
+        /* Construction from vertices and IDs */
+        __forceinline SubGridQBVHN(const unsigned int x[N],
+                                   const unsigned int y[N],
+                                   const unsigned int primID[N],
+                                   const BBox3fa * const subGridBounds,
+                                   const unsigned int geomID,
+                                   const unsigned int items)
+        {
+          clear();
+          _geomID = geomID;
+
+          __aligned(64) typename BVHN<N>::AABBNode node;
+          node.clear();          
+          for (size_t i=0;i<items;i++)
+          {
+            subgridIDs[i] = SubGridID(x[i],y[i],primID[i]);
+            node.setBounds(i,subGridBounds[i]);
+          }
+          qnode.init_dim(node);
+        }
+
+        __forceinline unsigned int geomID() const { return _geomID; }
+        __forceinline unsigned int primID(const size_t i) const { assert(i < N); return subgridIDs[i].primID; }
+        __forceinline unsigned int x(const size_t i) const { assert(i < N); return subgridIDs[i].x; }
+        __forceinline unsigned int y(const size_t i) const { assert(i < N); return subgridIDs[i].y; }
+
+        __forceinline SubGrid subgrid(const size_t i) const {
+          assert(i < N);
+          assert(primID(i) != -1);
+          return SubGrid(x(i),y(i),geomID(),primID(i));
+        }
+
+      public:
+        SubGridID subgridIDs[N];
+
+        typename BVHN<N>::QuantizedBaseNode qnode;
+
+        unsigned int _geomID;    // geometry ID of mesh
+
+
+        friend embree_ostream operator<<(embree_ostream cout, const SubGridQBVHN& sg) {
+          cout << "SubGridQBVHN " << embree_endl;
+          for (size_t i=0;i<N;i++)
+            cout << i << " ( x = " << sg.subgridIDs[i].x << ", y = " << sg.subgridIDs[i].y << ", primID = " << sg.subgridIDs[i].primID << " )" << embree_endl;
+          cout << "geomID " << sg._geomID << embree_endl;
+          cout << "lowerX " << sg.qnode.dequantizeLowerX() << embree_endl;
+          cout << "upperX " << sg.qnode.dequantizeUpperX() << embree_endl;
+          cout << "lowerY " << sg.qnode.dequantizeLowerY() << embree_endl;
+          cout << "upperY " << sg.qnode.dequantizeUpperY() << embree_endl;
+          cout << "lowerZ " << sg.qnode.dequantizeLowerZ() << embree_endl;
+          cout << "upperZ " << sg.qnode.dequantizeUpperZ() << embree_endl;
+          return cout;
+        }
+
+      };
+
+      template<int N>
+        typename SubGridQBVHN<N>::Type SubGridQBVHN<N>::type;
+
+      typedef SubGridQBVHN<4> SubGridQBVH4;
+      typedef SubGridQBVHN<8> SubGridQBVH8;
+
+
+      
+
+
+      /* QuantizedBaseNode as large subgrid leaf */
+      template<int N>
+      struct SubGridMBQBVHN
+      {
+        /* Virtual interface to query information about the quad type */
+        struct Type : public PrimitiveType
+        {
+          const char* name() const;
+          size_t sizeActive(const char* This) const;
+          size_t sizeTotal(const char* This) const;
+          size_t getBytes(const char* This) const;
+        };
+        static Type type;
+
+      public:
+
+        __forceinline size_t size() const
+        {
+          for (size_t i=0;i<N;i++)
+            if (primID(i) == -1) return i;
+          return N;
+        }
+
+      __forceinline void clear() {
+        for (size_t i=0;i<N;i++)
+          subgridIDs[i] = SubGridID(0,0,(unsigned int)-1);
+        qnode.clear();
+      }
+
+        /* Default constructor */
+        __forceinline SubGridMBQBVHN() {  }
+
+        /* Construction from vertices and IDs */
+        __forceinline SubGridMBQBVHN(const unsigned int x[N],
+                                     const unsigned int y[N],
+                                     const unsigned int primID[N],
+                                     const BBox3fa * const subGridBounds0,
+                                     const BBox3fa * const subGridBounds1,
+                                     const unsigned int geomID,
+                                     const float toffset,
+                                     const float tscale,
+                                     const unsigned int items)
+        {
+          clear();
+          _geomID = geomID;
+          time_offset = toffset;
+          time_scale  = tscale;
+
+          __aligned(64) typename BVHN<N>::AABBNode node0,node1;
+          node0.clear();          
+          node1.clear();          
+          for (size_t i=0;i<items;i++)
+          {
+            subgridIDs[i] = SubGridID(x[i],y[i],primID[i]);
+            node0.setBounds(i,subGridBounds0[i]);
+            node1.setBounds(i,subGridBounds1[i]);
+          }
+          qnode.node0.init_dim(node0);
+          qnode.node1.init_dim(node1);
+        }
+
+        __forceinline unsigned int geomID() const { return _geomID; }
+        __forceinline unsigned int primID(const size_t i) const { assert(i < N); return subgridIDs[i].primID; }
+        __forceinline unsigned int x(const size_t i) const { assert(i < N); return subgridIDs[i].x; }
+        __forceinline unsigned int y(const size_t i) const { assert(i < N); return subgridIDs[i].y; }
+
+        __forceinline SubGrid subgrid(const size_t i) const {
+          assert(i < N);
+          assert(primID(i) != -1);
+          return SubGrid(x(i),y(i),geomID(),primID(i));
+        }
+
+        __forceinline float adjustTime(const float t) const { return time_scale * (t-time_offset); }
+
+        template<int K>
+        __forceinline vfloat<K> adjustTime(const vfloat<K> &t) const { return time_scale * (t-time_offset); }
+
+      public:
+        SubGridID subgridIDs[N];
+
+        typename BVHN<N>::QuantizedBaseNodeMB qnode;
+
+        float time_offset;
+        float time_scale;
+        unsigned int _geomID;    // geometry ID of mesh
+
+
+        friend embree_ostream operator<<(embree_ostream cout, const SubGridMBQBVHN& sg) {
+          cout << "SubGridMBQBVHN " << embree_endl;
+          for (size_t i=0;i<N;i++)
+            cout << i << " ( x = " << sg.subgridIDs[i].x << ", y = " << sg.subgridIDs[i].y << ", primID = " << sg.subgridIDs[i].primID << " )" << embree_endl;
+          cout << "geomID      " << sg._geomID << embree_endl;
+          cout << "time_offset " << sg.time_offset << embree_endl;
+          cout << "time_scale  " << sg.time_scale << embree_endl;         
+          cout << "lowerX " << sg.qnode.node0.dequantizeLowerX() << embree_endl;
+          cout << "upperX " << sg.qnode.node0.dequantizeUpperX() << embree_endl;
+          cout << "lowerY " << sg.qnode.node0.dequantizeLowerY() << embree_endl;
+          cout << "upperY " << sg.qnode.node0.dequantizeUpperY() << embree_endl;
+          cout << "lowerZ " << sg.qnode.node0.dequantizeLowerZ() << embree_endl;
+          cout << "upperZ " << sg.qnode.node0.dequantizeUpperZ() << embree_endl;
+          cout << "lowerX " << sg.qnode.node1.dequantizeLowerX() << embree_endl;
+          cout << "upperX " << sg.qnode.node1.dequantizeUpperX() << embree_endl;
+          cout << "lowerY " << sg.qnode.node1.dequantizeLowerY() << embree_endl;
+          cout << "upperY " << sg.qnode.node1.dequantizeUpperY() << embree_endl;
+          cout << "lowerZ " << sg.qnode.node1.dequantizeLowerZ() << embree_endl;
+          cout << "upperZ " << sg.qnode.node1.dequantizeUpperZ() << embree_endl;
+          return cout;
+        }
+
+      };
+}

thirdparty/embree/kernels/geometry/subgrid_intersector.h

@@ -0,0 +1,515 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "subgrid.h"
+#include "subgrid_intersector_moeller.h"
+#include "subgrid_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+
+    // =======================================================================================
+    // =================================== SubGridIntersectors ===============================
+    // =======================================================================================
+
+
+    template<int N, bool filter>
+    struct SubGridIntersector1Moeller
+    {
+      typedef SubGridQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersector1MoellerTrumbore<4,filter> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(point_query.trav_prims,1,1,1);
+        AccelSet* accel = (AccelSet*)context->scene->get(subgrid.geomID());
+        assert(accel);
+        context->geomID = subgrid.geomID();
+        context->primID = subgrid.primID();
+        return accel->pointQuery(query, context);
+      }
+
+      template<bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+#if defined(__AVX__)
+          STAT3(normal.trav_hit_boxes[popcnt(mask)],1,1,1);
+#endif
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+            if (unlikely(dist[ID] > ray.tfar)) continue;
+            intersect(pre,ray,context,prim[i].subgrid(ID));
+          }
+        }
+      }
+      template<bool robust>        
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+            if (occluded(pre,ray,context,prim[i].subgrid(ID)))
+              return true;
+          }
+        }
+        return false;
+      }
+    
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t num, const TravPointQuery<N> &tquery, size_t& lazy_node)
+      {
+        bool changed = false;
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask;
+          if (likely(context->query_type == POINT_QUERY_TYPE_SPHERE)) {
+            mask = BVHNQuantizedBaseNodePointQuerySphere1<N>::pointQuery(&prim[i].qnode,tquery,dist);
+          } else {
+            mask = BVHNQuantizedBaseNodePointQueryAABB1<N>::pointQuery(&prim[i].qnode,tquery,dist);
+          }
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+            changed |= pointQuery(query, context, prim[i].subgrid(ID));
+          }
+        }
+        return changed;
+      }
+    };
+
+    template<int N, bool filter>
+    struct SubGridIntersector1Pluecker
+    {
+      typedef SubGridQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersector1Pluecker<4,filter> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(point_query.trav_prims,1,1,1);
+        AccelSet* accel = (AccelSet*)context->scene->get(subgrid.geomID());
+        context->geomID = subgrid.geomID();
+        context->primID = subgrid.primID();
+        return accel->pointQuery(query, context);
+      }
+
+      template<bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+#if defined(__AVX__)
+          STAT3(normal.trav_hit_boxes[popcnt(mask)],1,1,1);
+#endif
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+            if (unlikely(dist[ID] > ray.tfar)) continue;
+            intersect(pre,ray,context,prim[i].subgrid(ID));
+          }
+        }
+      }
+
+      template<bool robust>        
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+            if (occluded(pre,ray,context,prim[i].subgrid(ID)))
+              return true;
+          }
+        }
+        return false;
+      }
+      
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t num, const TravPointQuery<N> &tquery, size_t& lazy_node)
+      {
+        bool changed = false;
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          size_t mask;
+          if (likely(context->query_type == POINT_QUERY_TYPE_SPHERE)) {
+            mask = BVHNQuantizedBaseNodePointQuerySphere1<N>::pointQuery(&prim[i].qnode,tquery,dist);
+          } else {
+            mask = BVHNQuantizedBaseNodePointQueryAABB1<N>::pointQuery(&prim[i].qnode,tquery,dist);
+          }
+#if defined(__AVX__)
+          STAT3(point_query.trav_hit_boxes[popcnt(mask)],1,1,1);
+#endif
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+            changed |= pointQuery(query, context, prim[i].subgrid(ID));
+          }
+        }
+        return changed;
+      }
+    };
+
+    template<int N, int K, bool filter>
+    struct SubGridIntersectorKMoeller
+    {
+      typedef SubGridQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersectorKMoellerTrumbore<4,K,filter> Precalculations;
+
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        Vec3fa vtx[16];
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        subgrid.gather(vtx,context->scene);
+        for (unsigned int i=0; i<4; i++)
+        {
+          const Vec3vf<K> p0 = vtx[i*4+0];
+          const Vec3vf<K> p1 = vtx[i*4+1];
+          const Vec3vf<K> p2 = vtx[i*4+2];
+          const Vec3vf<K> p3 = vtx[i*4+3];
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,g,subgrid,i,IntersectKEpilogM<4,K,filter>(ray,context,subgrid.geomID(),subgrid.primID(),i));
+        }
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        vbool<K> valid0 = valid_i;
+        Vec3fa vtx[16];
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        subgrid.gather(vtx,context->scene);
+        for (unsigned int i=0; i<4; i++)
+        {
+          const Vec3vf<K> p0 = vtx[i*4+0];
+          const Vec3vf<K> p1 = vtx[i*4+1];
+          const Vec3vf<K> p2 = vtx[i*4+2];
+          const Vec3vf<K> p3 = vtx[i*4+3];
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          pre.intersectK(valid0,ray,p0,p1,p2,p3,g,subgrid,i,OccludedKEpilogM<4,K,filter>(valid0,ray,context,subgrid.geomID(),subgrid.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+        template<bool robust>
+          static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid & isecK.intersectK(&prim[j].qnode,i,tray,dist))) continue;
+              intersect(valid,pre,ray,context,prim[j].subgrid(i));
+            }
+          }
+        }
+
+        template<bool robust>        
+        static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+          vbool<K> valid0 = valid;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid0 & isecK.intersectK(&prim[j].qnode,i,tray,dist))) continue;
+              valid0 &= !occluded(valid0,pre,ray,context,prim[j].subgrid(i));
+              if (none(valid0)) break;
+            }
+          }
+          return !valid0;
+        }
+        
+        template<bool robust>        
+          static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+              if (unlikely(dist[ID] > ray.tfar[k])) continue;
+              intersect(pre,ray,k,context,prim[i].subgrid(ID));
+            }
+          }
+        }
+        
+        template<bool robust>
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+              if (occluded(pre,ray,k,context,prim[i].subgrid(ID)))
+                return true;
+            }
+          }
+          return false;
+        }
+    };
+
+
+    template<int N, int K, bool filter>
+    struct SubGridIntersectorKPluecker
+    {
+      typedef SubGridQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersectorKPluecker<4,K,filter> Precalculations;
+
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        Vec3fa vtx[16];
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        subgrid.gather(vtx,context->scene);
+        for (unsigned int i=0; i<4; i++)
+        {
+          const Vec3vf<K> p0 = vtx[i*4+0];
+          const Vec3vf<K> p1 = vtx[i*4+1];
+          const Vec3vf<K> p2 = vtx[i*4+2];
+          const Vec3vf<K> p3 = vtx[i*4+3];
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          pre.intersectK(valid_i,ray,p0,p1,p2,p3,g,subgrid,i,IntersectKEpilogM<4,K,filter>(ray,context,subgrid.geomID(),subgrid.primID(),i));
+        }
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        vbool<K> valid0 = valid_i;
+        Vec3fa vtx[16];
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        subgrid.gather(vtx,context->scene);
+        for (unsigned int i=0; i<4; i++)
+        {
+          const Vec3vf<K> p0 = vtx[i*4+0];
+          const Vec3vf<K> p1 = vtx[i*4+1];
+          const Vec3vf<K> p2 = vtx[i*4+2];
+          const Vec3vf<K> p3 = vtx[i*4+3];
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          pre.occludedK(valid0,ray,p0,p1,p2,p3,g,subgrid,i,OccludedKEpilogM<4,K,filter>(valid0,ray,context,subgrid.geomID(),subgrid.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+        Vec3vf4 v0,v1,v2,v3; subgrid.gather(v0,v1,v2,v3,context->scene);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+      
+        template<bool robust>
+          static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid & isecK.intersectK(&prim[j].qnode,i,tray,dist))) continue;
+              intersect(valid,pre,ray,context,prim[j].subgrid(i));
+            }
+          }
+        }
+
+        template<bool robust>        
+        static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+          vbool<K> valid0 = valid;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid0 & isecK.intersectK(&prim[j].qnode,i,tray,dist))) continue;
+              valid0 &= !occluded(valid0,pre,ray,context,prim[j].subgrid(i));
+              if (none(valid0)) break;
+            }
+          }
+          return !valid0;
+        }
+        
+        template<bool robust>        
+          static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+              if (unlikely(dist[ID] > ray.tfar[k])) continue;
+              intersect(pre,ray,k,context,prim[i].subgrid(ID));
+            }
+          }
+        }
+        
+        template<bool robust>
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              assert(((size_t)1 << ID) & movemask(prim[i].qnode.validMask()));
+
+              if (occluded(pre,ray,k,context,prim[i].subgrid(ID)))
+                return true;
+            }
+          }
+          return false;
+        }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/subgrid_intersector_moeller.h

@@ -0,0 +1,382 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "subgrid.h"
+#include "quad_intersector_moeller.h"
+
+namespace embree
+{
+  namespace isa
+  {
+
+    /* ----------------------------- */
+    /* -- single ray intersectors -- */
+    /* ----------------------------- */
+
+    template<int M>
+    __forceinline void interpolateUV(MoellerTrumboreHitM<M,UVIdentity<M>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const vint<M> &stepX, const vint<M> &stepY) 
+    {
+      /* correct U,V interpolation across the entire grid */
+      const vint<M> sx((int)subgrid.x());
+      const vint<M> sy((int)subgrid.y());
+      const vint<M> sxM(sx + stepX); 
+      const vint<M> syM(sy + stepY); 
+      const float inv_resX = rcp((float)((int)g.resX-1));
+      const float inv_resY = rcp((float)((int)g.resY-1));          
+      hit.U = (hit.U + (vfloat<M>)sxM * hit.absDen) * inv_resX;
+      hit.V = (hit.V + (vfloat<M>)syM * hit.absDen) * inv_resY;
+    }
+    
+    template<int M, bool filter>
+      struct SubGridQuadMIntersector1MoellerTrumbore;
+
+    template<int M, bool filter>
+      struct SubGridQuadMIntersector1MoellerTrumbore
+      {
+        __forceinline SubGridQuadMIntersector1MoellerTrumbore() {}
+
+        __forceinline SubGridQuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {}
+
+        __forceinline void intersect(RayHit& ray, RayQueryContext* context,
+                                     const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                     const GridMesh::Grid &g, const SubGrid& subgrid) const
+        {
+          UVIdentity<M> mapUV;
+          MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+          MoellerTrumboreIntersector1<M> intersector(ray,nullptr);
+          Intersect1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID());
+
+          /* intersect first triangle */
+          if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+
+          /* intersect second triangle */
+          if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+          {
+            hit.U = hit.absDen - hit.U;
+            hit.V = hit.absDen - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+        }
+      
+        __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                    const GridMesh::Grid &g, const SubGrid& subgrid) const
+        {
+          UVIdentity<M> mapUV;
+          MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+          MoellerTrumboreIntersector1<M> intersector(ray,nullptr);
+          Occluded1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID());
+          
+          /* intersect first triangle */
+          if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            if (epilog(hit.valid,hit))
+              return true;
+          }
+
+          /* intersect second triangle */
+          if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+          {
+            hit.U = hit.absDen - hit.U;
+            hit.V = hit.absDen - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            if (epilog(hit.valid,hit))
+              return true;
+          }
+          return false;
+        }
+      };
+
+#if defined (__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<bool filter>
+      struct SubGridQuadMIntersector1MoellerTrumbore<4,filter>
+    {
+      __forceinline SubGridQuadMIntersector1MoellerTrumbore() {}
+
+      __forceinline SubGridQuadMIntersector1MoellerTrumbore(const Ray& ray, const void* ptr) {}
+      
+      template<typename Epilog>
+        __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
+      {
+        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
+        UVIdentity<8> mapUV;
+        MoellerTrumboreHitM<8,UVIdentity<8>> hit(mapUV);
+        MoellerTrumboreIntersector1<8> intersector(ray,nullptr);
+        const vbool8 flags(0,0,0,0,1,1,1,1);
+        if (unlikely(intersector.intersect(ray,vtx0,vtx1,vtx2,mapUV,hit)))
+        {
+	  /* correct U,V interpolation across the entire grid */
+	  const vfloat8 U = select(flags,hit.absDen - hit.V,hit.U);	  
+	  const vfloat8 V = select(flags,hit.absDen - hit.U,hit.V);
+	  hit.U = U;
+	  hit.V = V;
+	  hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); 	  
+          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));
+          if (unlikely(epilog(hit.valid,hit)))
+            return true;
+        }
+        return false;
+      }
+      
+      __forceinline bool intersect(RayHit& ray, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const GridMesh::Grid &g, const SubGrid& subgrid) const
+      {
+          return intersect(ray,v0,v1,v2,v3,g,subgrid,Intersect1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID()));
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                  const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                  const GridMesh::Grid &g, const SubGrid& subgrid) const
+      {
+          return intersect(ray,v0,v1,v2,v3,g,subgrid,Occluded1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID()));
+      }
+    };
+
+#endif
+
+    // ============================================================================================================================
+    // ============================================================================================================================
+    // ============================================================================================================================
+
+
+    /* ----------------------------- */
+    /* -- ray packet intersectors -- */
+    /* ----------------------------- */
+
+    template<int K>
+    __forceinline void interpolateUV(const vbool<K>& valid, MoellerTrumboreHitK<K,UVIdentity<K>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const unsigned int i) 
+    {
+      /* correct U,V interpolation across the entire grid */
+      const unsigned int sx = subgrid.x() + (unsigned int)(i % 2);
+      const unsigned int sy = subgrid.y() + (unsigned int)(i >>1);
+      const float inv_resX = rcp((float)(int)(g.resX-1));
+      const float inv_resY = rcp((float)(int)(g.resY-1));      
+      hit.U = select(valid,(hit.U + vfloat<K>((float)sx) * hit.absDen) * inv_resX,hit.U);
+      hit.V = select(valid,(hit.V + vfloat<K>((float)sy) * hit.absDen) * inv_resY,hit.V);
+    }
+        
+    template<int M, int K, bool filter>
+      struct SubGridQuadMIntersectorKMoellerTrumboreBase
+      {
+        __forceinline SubGridQuadMIntersectorKMoellerTrumboreBase(const vbool<K>& valid, const RayK<K>& ray) {}
+
+        template<typename Epilog>
+        __forceinline bool intersectK(const vbool<K>& valid, 
+                                      RayK<K>& ray,
+                                      const Vec3vf<K>& v0,
+                                      const Vec3vf<K>& v1,
+                                      const Vec3vf<K>& v2,
+                                      const Vec3vf<K>& v3,
+                                      const GridMesh::Grid &g, 
+                                      const SubGrid &subgrid,
+                                      const unsigned int i,
+                                      const Epilog& epilog) const
+        {
+	  UVIdentity<K> mapUV;
+	  MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV);
+	  MoellerTrumboreIntersectorK<M,K> intersector;
+
+          const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit);
+	  if (any(valid0))
+	    {
+	      interpolateUV(valid0,hit,g,subgrid,i);
+	      epilog(valid0,hit);
+	    }
+          const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit);
+	  if (any(valid1))
+	    {
+	      hit.U = hit.absDen - hit.U;
+	      hit.V = hit.absDen - hit.V;	      
+	      interpolateUV(valid1,hit,g,subgrid,i);
+	      epilog(valid1,hit);
+	    }
+	  return any(valid0|valid1);	  
+        }
+
+       template<typename Epilog>
+        __forceinline bool occludedK(const vbool<K>& valid, 
+				     RayK<K>& ray,
+				     const Vec3vf<K>& v0,
+				     const Vec3vf<K>& v1,
+				     const Vec3vf<K>& v2,
+				     const Vec3vf<K>& v3,
+				     const GridMesh::Grid &g, 
+				     const SubGrid &subgrid,
+				     const unsigned int i,
+				     const Epilog& epilog) const
+        {
+	  UVIdentity<K> mapUV;
+	  MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV);
+	  MoellerTrumboreIntersectorK<M,K> intersector;
+
+	  vbool<K> valid_final = valid;
+          const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit);
+	  if (any(valid0))
+	    {
+	      interpolateUV(valid0,hit,g,subgrid,i);
+	      epilog(valid0,hit);
+	      valid_final &= !valid0;
+	    }
+	  if (none(valid_final)) return true;	      	  
+          const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit);
+	  if (any(valid1))
+	    {
+	      hit.U = hit.absDen - hit.U;
+	      hit.V = hit.absDen - hit.V;	      
+	      interpolateUV(valid1,hit,g,subgrid,i);
+	      epilog(valid1,hit);
+	      valid_final &= !valid1;	      
+	    }
+	  return none(valid_final);
+        }
+
+        static __forceinline bool intersect1(RayK<K>& ray,
+                                             size_t k,
+                                             const Vec3vf<M>& v0,
+                                             const Vec3vf<M>& v1,
+                                             const Vec3vf<M>& v2,
+                                             MoellerTrumboreHitM<M,UVIdentity<M>> &hit)
+        {
+          const Vec3vf<M> e1 = v0-v1;
+          const Vec3vf<M> e2 = v2-v0;
+	  MoellerTrumboreIntersectorK<8,K> intersector;
+	  UVIdentity<M> mapUV;
+	  return intersector.intersectEdge(ray,k,v0,e1,e2,mapUV,hit);
+        }
+	
+      };
+
+    template<int M, int K, bool filter>
+      struct SubGridQuadMIntersectorKMoellerTrumbore : public SubGridQuadMIntersectorKMoellerTrumboreBase<M,K,filter>
+    {
+      __forceinline SubGridQuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray)
+        : SubGridQuadMIntersectorKMoellerTrumboreBase<M,K,filter>(valid,ray) {}
+
+      __forceinline void intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+	UVIdentity<M> mapUV;
+	MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+        Intersect1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID());
+	MoellerTrumboreIntersectorK<M,K> intersector;
+	/* intersect first triangle */
+	if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+
+	/* intersect second triangle */
+	if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) 
+          {
+	    hit.U = hit.absDen - hit.U;
+	    hit.V = hit.absDen - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+	UVIdentity<M> mapUV;
+        MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+        Occluded1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID());	
+	MoellerTrumboreIntersectorK<M,K> intersector;
+	/* intersect first triangle */
+	if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) 
+        {
+          interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+          if (epilog(hit.valid,hit)) return true;
+        }
+
+	/* intersect second triangle */
+	if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) 
+        {
+          hit.U = hit.absDen - hit.U;
+          hit.V = hit.absDen - hit.V;	  
+          interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+          if (epilog(hit.valid,hit)) return true;
+        }
+        return false;
+      }
+    };
+
+
+#if defined (__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<int K, bool filter>
+      struct SubGridQuadMIntersectorKMoellerTrumbore<4,K,filter> : public SubGridQuadMIntersectorKMoellerTrumboreBase<4,K,filter>
+    {
+      __forceinline SubGridQuadMIntersectorKMoellerTrumbore(const vbool<K>& valid, const RayK<K>& ray)
+        : SubGridQuadMIntersectorKMoellerTrumboreBase<4,K,filter>(valid,ray) {}
+      
+      template<typename Epilog>
+        __forceinline bool intersect1(RayK<K>& ray, size_t k,const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                      const GridMesh::Grid &g, const SubGrid &subgrid, 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);
+
+        UVIdentity<8> mapUV;
+        MoellerTrumboreHitM<8,UVIdentity<8>> hit(mapUV);
+        if (SubGridQuadMIntersectorKMoellerTrumboreBase<8,K,filter>::intersect1(ray,k,vtx0,vtx1,vtx2,hit))
+        {
+	  const vfloat8 U = select(flags,hit.absDen - hit.V,hit.U);	  
+	  const vfloat8 V = select(flags,hit.absDen - hit.U,hit.V);
+	  hit.U = U;
+	  hit.V = V;
+	  hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); 	  	  
+	  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));
+          if (unlikely(epilog(hit.valid,hit)))
+            return true;
+
+        }
+        return false;
+      }
+      
+      __forceinline bool intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Intersect1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID()));
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Occluded1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID()));
+      }
+    };
+
+#endif
+
+
+
+  }
+}

thirdparty/embree/kernels/geometry/subgrid_intersector_pluecker.h

@@ -0,0 +1,367 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "subgrid.h"
+#include "quad_intersector_moeller.h"
+#include "quad_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+
+    template<int M>
+    __forceinline void interpolateUV(PlueckerHitM<M,UVIdentity<M>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const vint<M> &stepX, const vint<M> &stepY) 
+    {
+      /* correct U,V interpolation across the entire grid */
+      const vint<M> sx((int)subgrid.x());
+      const vint<M> sy((int)subgrid.y());
+      const vint<M> sxM(sx + stepX);
+      const vint<M> syM(sy + stepY);
+      const float inv_resX = rcp((float)((int)g.resX-1));
+      const float inv_resY = rcp((float)((int)g.resY-1));          
+      hit.U = (hit.U + vfloat<M>(sxM) * hit.UVW) * inv_resX;
+      hit.V = (hit.V + vfloat<M>(syM) * hit.UVW) * inv_resY;
+    }
+    
+    template<int M, bool filter>
+      struct SubGridQuadMIntersector1Pluecker;
+
+    template<int M, bool filter>
+      struct SubGridQuadMIntersector1Pluecker
+      {
+        __forceinline SubGridQuadMIntersector1Pluecker() {}
+
+        __forceinline SubGridQuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {}
+
+        __forceinline void intersect(RayHit& ray, RayQueryContext* context,
+                                     const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                     const GridMesh::Grid &g, const SubGrid& subgrid) const
+        {
+          UVIdentity<M> mapUV;
+          PlueckerHitM<M,UVIdentity<M>> hit(mapUV);
+          PlueckerIntersector1<M> intersector(ray,nullptr);
+	  
+          Intersect1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID());
+
+          /* intersect first triangle */
+	  if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+
+          /* intersect second triangle */
+	  if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+          {
+	    hit.U = hit.UVW - hit.U;
+	    hit.V = hit.UVW - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+        }
+      
+        __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3,
+                                    const GridMesh::Grid &g, const SubGrid& subgrid) const
+        {
+          UVIdentity<M> mapUV;
+          PlueckerHitM<M,UVIdentity<M>> hit(mapUV);
+          PlueckerIntersector1<M> intersector(ray,nullptr);
+          Occluded1EpilogMU<M,filter> epilog(ray,context,subgrid.geomID(),subgrid.primID());
+
+          /* intersect first triangle */
+	  if (intersector.intersect(ray,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            if (epilog(hit.valid,hit))
+	      return true;
+          }
+
+          /* intersect second triangle */
+	  if (intersector.intersect(ray,v2,v3,v1,mapUV,hit)) 
+          {
+	    hit.U = hit.UVW - hit.U;
+	    hit.V = hit.UVW - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            if (epilog(hit.valid,hit))
+	      return true;
+          }
+          return false;
+        }
+      };
+
+#if defined (__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<bool filter>
+      struct SubGridQuadMIntersector1Pluecker<4,filter>
+    {
+      __forceinline SubGridQuadMIntersector1Pluecker() {}
+
+      __forceinline SubGridQuadMIntersector1Pluecker(const Ray& ray, const void* ptr) {}
+      
+      template<typename Epilog>
+        __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
+      {
+        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
+
+        UVIdentity<8> mapUV;
+        PlueckerHitM<8,UVIdentity<8>> hit(mapUV);
+        PlueckerIntersector1<8> intersector(ray,nullptr);
+        const vbool8 flags(0,0,0,0,1,1,1,1);
+        if (unlikely(intersector.intersect(ray,vtx0,vtx1,vtx2,mapUV,hit)))
+        {
+	  /* correct U,V interpolation across the entire grid */
+	  const vfloat8 U = select(flags,hit.UVW - hit.V,hit.U);	  
+	  const vfloat8 V = select(flags,hit.UVW - hit.U,hit.V);
+	  hit.U = U;
+	  hit.V = V;	  
+	  hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); 
+          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));
+          if (unlikely(epilog(hit.valid,hit)))
+            return true;
+        }
+        return false;
+      }
+      
+      __forceinline bool intersect(RayHit& ray, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                   const GridMesh::Grid &g, const SubGrid& subgrid) const
+      {
+          return intersect(ray,v0,v1,v2,v3,g,subgrid,Intersect1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID()));
+      }
+      
+      __forceinline bool occluded(Ray& ray, RayQueryContext* context,
+                                  const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                  const GridMesh::Grid &g, const SubGrid& subgrid) const
+      {
+          return intersect(ray,v0,v1,v2,v3,g,subgrid,Occluded1EpilogMU<8,filter>(ray,context,subgrid.geomID(),subgrid.primID()));
+      }
+    };
+
+#endif
+
+
+    /* ----------------------------- */
+    /* -- ray packet intersectors -- */
+    /* ----------------------------- */
+
+    template<int K>
+    __forceinline void interpolateUV(const vbool<K>& valid, PlueckerHitK<K,UVIdentity<K>> &hit,const GridMesh::Grid &g, const SubGrid& subgrid, const unsigned int i) 
+    {
+      /* correct U,V interpolation across the entire grid */
+      const unsigned int sx = subgrid.x() + (unsigned int)(i % 2);
+      const unsigned int sy = subgrid.y() + (unsigned int)(i >>1);
+      const float inv_resX = rcp((float)(int)(g.resX-1));
+      const float inv_resY = rcp((float)(int)(g.resY-1));      
+      hit.U = select(valid,(hit.U + vfloat<K>((float)sx) * hit.UVW) * inv_resX,hit.U);
+      hit.V = select(valid,(hit.V + vfloat<K>((float)sy) * hit.UVW) * inv_resY,hit.V);
+    }
+    
+    template<int M, int K, bool filter>
+      struct SubGridQuadMIntersectorKPlueckerBase
+      {
+        __forceinline SubGridQuadMIntersectorKPlueckerBase(const vbool<K>& valid, const RayK<K>& ray) {}
+
+        template<typename Epilog>
+        __forceinline bool intersectK(const vbool<K>& valid, 
+                                      RayK<K>& ray,
+                                      const Vec3vf<K>& v0,
+                                      const Vec3vf<K>& v1,
+                                      const Vec3vf<K>& v2,
+                                      const Vec3vf<K>& v3,
+                                      const GridMesh::Grid &g, 
+                                      const SubGrid &subgrid,
+                                      const unsigned int i,
+                                      const Epilog& epilog) const
+        {
+	  UVIdentity<K> mapUV;
+	  PlueckerHitK<K,UVIdentity<K>> hit(mapUV);
+	  PlueckerIntersectorK<M,K> intersector;
+
+          const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit);
+	  if (any(valid0))
+	    {
+	      interpolateUV(valid0,hit,g,subgrid,i);
+	      epilog(valid0,hit);
+	    }
+          const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit);
+	  if (any(valid1))
+	    {
+	      hit.U = hit.UVW - hit.U;
+	      hit.V = hit.UVW - hit.V;	      
+	      interpolateUV(valid1,hit,g,subgrid,i);
+	      epilog(valid1,hit);
+	    }
+	  return any(valid0|valid1);	  
+        }
+
+       template<typename Epilog>
+        __forceinline bool occludedK(const vbool<K>& valid, 
+				     RayK<K>& ray,
+				     const Vec3vf<K>& v0,
+				     const Vec3vf<K>& v1,
+				     const Vec3vf<K>& v2,
+				     const Vec3vf<K>& v3,
+				     const GridMesh::Grid &g, 
+				     const SubGrid &subgrid,
+				     const unsigned int i,
+				     const Epilog& epilog) const
+        {
+	  UVIdentity<K> mapUV;
+	  PlueckerHitK<K,UVIdentity<K>> hit(mapUV);
+	  PlueckerIntersectorK<M,K> intersector;
+
+	  vbool<K> valid_final = valid;
+          const vbool<K> valid0 = intersector.intersectK(valid,ray,v0,v1,v3,mapUV,hit);
+	  if (any(valid0))
+	    {
+	      interpolateUV(valid0,hit,g,subgrid,i);
+	      epilog(valid0,hit);
+	      valid_final &= !valid0;
+	    }
+	  if (none(valid_final)) return true;	      	  
+          const vbool<K> valid1 = intersector.intersectK(valid,ray,v2,v3,v1,mapUV,hit);
+	  if (any(valid1))
+	    {
+	      hit.U = hit.UVW - hit.U;
+	      hit.V = hit.UVW - hit.V;	      
+	      interpolateUV(valid1,hit,g,subgrid,i);
+	      epilog(valid1,hit);
+	      valid_final &= !valid1;	      
+	    }
+	  return none(valid_final);
+        }
+
+	
+      };
+
+
+    
+
+    template<int M, int K, bool filter>
+      struct SubGridQuadMIntersectorKPluecker : public SubGridQuadMIntersectorKPlueckerBase<M,K,filter>
+    {
+      __forceinline SubGridQuadMIntersectorKPluecker(const vbool<K>& valid, const RayK<K>& ray)
+        : SubGridQuadMIntersectorKPlueckerBase<M,K,filter>(valid,ray) {}
+
+      __forceinline void intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+	UVIdentity<M> mapUV;
+	PlueckerHitM<M,UVIdentity<M>> hit(mapUV);
+        Intersect1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID());
+	PlueckerIntersectorK<M,K> intersector;
+	
+	/* intersect first triangle */
+	if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) 
+          {
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+
+	/* intersect second triangle */
+	if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) 
+          {
+	    hit.U = hit.UVW - hit.U;
+	    hit.V = hit.UVW - hit.V;
+            interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+            epilog(hit.valid,hit);
+          }
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const Vec3vf<M>& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+	UVIdentity<M> mapUV;
+	PlueckerHitM<M,UVIdentity<M>> hit(mapUV);
+        Occluded1KEpilogMU<M,K,filter> epilog(ray,k,context,subgrid.geomID(),subgrid.primID());	
+	PlueckerIntersectorK<M,K> intersector;
+	
+	/* intersect first triangle */
+	if (intersector.intersect(ray,k,v0,v1,v3,mapUV,hit)) 
+        {
+          interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+          if (epilog(hit.valid,hit)) return true;
+        }
+
+	/* intersect second triangle */
+	if (intersector.intersect(ray,k,v2,v3,v1,mapUV,hit)) 
+        {
+	  hit.U = hit.UVW - hit.U;
+	  hit.V = hit.UVW - hit.V;	  
+          interpolateUV<M>(hit,g,subgrid,vint<M>(0,1,1,0),vint<M>(0,0,1,1));
+          if (epilog(hit.valid,hit)) return true;
+        }
+        return false;
+      }
+    };
+
+
+#if defined (__AVX__)
+
+    /*! Intersects 4 quads with 1 ray using AVX */
+    template<int K, bool filter>
+      struct SubGridQuadMIntersectorKPluecker<4,K,filter> : public SubGridQuadMIntersectorKPlueckerBase<4,K,filter>
+    {
+      __forceinline SubGridQuadMIntersectorKPluecker(const vbool<K>& valid, const RayK<K>& ray)
+        : SubGridQuadMIntersectorKPlueckerBase<4,K,filter>(valid,ray) {}
+      
+      template<typename Epilog>
+        __forceinline bool intersect1(RayK<K>& ray, size_t k,const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, 
+                                      const GridMesh::Grid &g, const SubGrid &subgrid, 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
+	UVIdentity<8> mapUV;
+	PlueckerHitM<8,UVIdentity<8>> hit(mapUV);
+	PlueckerIntersectorK<8,K> intersector;
+	const vbool8 flags(0,0,0,0,1,1,1,1);
+        if (unlikely(intersector.intersect(ray,k,vtx0,vtx1,vtx2,mapUV,hit)))	
+        {
+          /* correct U,V interpolation across the entire grid */
+	  const vfloat8 U = select(flags,hit.UVW - hit.V,hit.U);	  
+	  const vfloat8 V = select(flags,hit.UVW - hit.U,hit.V);
+	  hit.U = U;
+	  hit.V = V;	  
+	  hit.vNg *= select(flags,vfloat8(-1.0f),vfloat8(1.0f)); 
+          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));
+          if (unlikely(epilog(hit.valid,hit)))
+            return true;
+        }
+        return false;
+      }
+      
+      __forceinline bool intersect1(RayHitK<K>& ray, size_t k, RayQueryContext* context,
+                                    const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Intersect1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID()));
+      }
+      
+      __forceinline bool occluded1(RayK<K>& ray, size_t k, RayQueryContext* context,
+                                   const Vec3vf4& v0, const Vec3vf4& v1, const Vec3vf4& v2, const Vec3vf4& v3, const GridMesh::Grid &g, const SubGrid &subgrid) const
+      {
+        return intersect1(ray,k,v0,v1,v2,v3,g,subgrid,Occluded1KEpilogMU<8,K,filter>(ray,k,context,subgrid.geomID(),subgrid.primID()));
+      }
+    };
+#endif
+
+    
+  }
+}

thirdparty/embree/kernels/geometry/subgrid_mb_intersector.h

@@ -0,0 +1,236 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "subgrid_intersector.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int N, bool filter>
+    struct SubGridMBIntersector1Pluecker
+    {
+      typedef SubGridMBQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersector1Pluecker<4,filter> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        float ftime;
+        const int itime = mesh->timeSegment(ray.time(), ftime);
+        Vec3vf4 v0,v1,v2,v3; subgrid.gatherMB(v0,v1,v2,v3,context->scene,itime,ftime);
+        pre.intersect(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        float ftime;
+        const int itime = mesh->timeSegment(ray.time(), ftime);
+
+        Vec3vf4 v0,v1,v2,v3; subgrid.gatherMB(v0,v1,v2,v3,context->scene,itime,ftime);
+        return pre.occluded(ray,context,v0,v1,v2,v3,g,subgrid);
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const SubGrid& subgrid)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, subgrid);
+      }
+
+      template<bool robust>
+        static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+        for (size_t i=0;i<num;i++)
+        {
+          vfloat<N> dist;
+          const float time = prim[i].adjustTime(ray.time());
+
+          assert(time <= 1.0f);
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,time,dist); 
+#if defined(__AVX__)
+          STAT3(normal.trav_hit_boxes[popcnt(mask)],1,1,1);
+#endif
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            if (unlikely(dist[ID] > ray.tfar)) continue;
+            intersect(pre,ray,context,prim[i].subgrid(ID));
+          }
+        }
+      }
+
+      template<bool robust>        
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+      {
+        BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+        for (size_t i=0;i<num;i++)
+        {
+          const float time = prim[i].adjustTime(ray.time());
+          assert(time <= 1.0f);
+          vfloat<N> dist;
+          size_t mask = isec1.intersect(&prim[i].qnode,tray,time,dist); 
+          while(mask != 0)
+          {
+            const size_t ID = bscf(mask); 
+            if (occluded(pre,ray,context,prim[i].subgrid(ID)))
+              return true;
+          }
+        }
+        return false;
+      }
+      
+      static __forceinline bool pointQuery(const Accel::Intersectors* This, PointQuery* query, PointQueryContext* context, const Primitive* prim, size_t num, const TravPointQuery<N> &tquery, size_t& lazy_node)
+      {
+        assert(false && "not implemented");
+        return false;
+      }
+    };
+
+
+    template<int N, int K, bool filter>
+    struct SubGridMBIntersectorKPluecker
+    {
+      typedef SubGridMBQBVHN<N> Primitive;
+      typedef SubGridQuadMIntersectorKPluecker<4,K,filter> Precalculations;
+
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        size_t m_valid = movemask(valid_i);
+        while(m_valid)
+        {
+          size_t ID = bscf(m_valid);
+          intersect(pre,ray,ID,context,subgrid);
+        }
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        vbool<K> valid0 = valid_i;
+        size_t m_valid = movemask(valid_i);
+        while(m_valid)
+        {
+          size_t ID = bscf(m_valid);
+          if (occluded(pre,ray,ID,context,subgrid))
+            clear(valid0,ID);
+        }
+        return !valid0;
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+ 
+        vfloat<K> ftime;
+        const vint<K> itime = mesh->timeSegment<K>(ray.time(), ftime);
+        Vec3vf4 v0,v1,v2,v3; subgrid.gatherMB(v0,v1,v2,v3,context->scene,itime[k],ftime[k]);
+        pre.intersect1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const SubGrid& subgrid)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const GridMesh* mesh    = context->scene->get<GridMesh>(subgrid.geomID());
+        const GridMesh::Grid &g = mesh->grid(subgrid.primID());
+
+        vfloat<K> ftime;
+        const vint<K> itime = mesh->timeSegment<K>(ray.time(), ftime);
+        Vec3vf4 v0,v1,v2,v3; subgrid.gatherMB(v0,v1,v2,v3,context->scene,itime[k],ftime[k]);
+        return pre.occluded1(ray,k,context,v0,v1,v2,v3,g,subgrid);
+      }
+
+        template<bool robust>
+          static __forceinline void intersect(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            const vfloat<K> time = prim[j].template adjustTime<K>(ray.time());
+
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid & isecK.intersectK(&prim[j].qnode,i,tray,time,dist))) continue;
+              intersect(valid,pre,ray,context,prim[j].subgrid(i));
+            }
+          }
+        }
+
+        template<bool robust>        
+        static __forceinline vbool<K> occluded(const vbool<K>& valid, const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive* prim, size_t num, const TravRayK<K, robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersectorK<N,K,robust> isecK;
+
+          vbool<K> valid0 = valid;
+          for (size_t j=0;j<num;j++)
+          {
+            size_t m_valid = movemask(prim[j].qnode.validMask());
+            const vfloat<K> time = prim[j].template adjustTime<K>(ray.time());
+            vfloat<K> dist;
+            while(m_valid)
+            {
+              const size_t i = bscf(m_valid);
+              if (none(valid0 & isecK.intersectK(&prim[j].qnode,i,tray,time,dist))) continue;
+              valid0 &= !occluded(valid0,pre,ray,context,prim[j].subgrid(i));
+              if (none(valid0)) break;
+            }
+          }
+          return !valid0;
+        }
+        
+        template<bool robust>        
+          static __forceinline void intersect(const Accel::Intersectors* This, Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            const float time = prim[i].adjustTime(ray.time()[k]);
+            assert(time <= 1.0f);
+
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,time,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              if (unlikely(dist[ID] > ray.tfar[k])) continue;
+              intersect(pre,ray,k,context,prim[i].subgrid(ID));
+            }
+          }
+        }
+        
+        template<bool robust>
+        static __forceinline bool occluded(const Accel::Intersectors* This, Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive* prim, size_t num, const TravRay<N,robust> &tray, size_t& lazy_node)
+        {
+          BVHNQuantizedBaseNodeIntersector1<N,robust> isec1;
+          
+          for (size_t i=0;i<num;i++)
+          {
+            vfloat<N> dist;
+            const float time = prim[i].adjustTime(ray.time()[k]);
+            assert(time <= 1.0f);
+
+            size_t mask = isec1.intersect(&prim[i].qnode,tray,time,dist); 
+            while(mask != 0)
+            {
+              const size_t ID = bscf(mask); 
+              if (occluded(pre,ray,k,context,prim[i].subgrid(ID)))
+                return true;
+            }
+          }
+          return false;
+        }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/triangle.h

@@ -0,0 +1,162 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  /* Precalculated representation for M triangles. Stores for each
+     triangle a base vertex, two edges, and the geometry normal to
+     speed up intersection calculations */
+  template<int M>
+  struct TriangleM
+  {
+  public:
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+    
+  public:
+
+    /* Returns maximum number of stored triangles */
+    static __forceinline size_t max_size() { return M; }
+    
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+
+  public:
+
+    /* Default constructor */
+    __forceinline TriangleM() {}
+
+    /* Construction from vertices and IDs */
+    __forceinline TriangleM(const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const vuint<M>& geomIDs, const vuint<M>& primIDs)
+      : v0(v0), e1(v0-v1), e2(v2-v0), geomIDs(geomIDs), primIDs(primIDs) {}
+
+    /* Returns a mask that tells which triangles are valid */
+    __forceinline vbool<M> valid() const { return geomIDs != vuint<M>(-1); }
+
+    /* Returns true if the specified triangle is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return geomIDs[i] != -1; }
+    
+    /* Returns the number of stored triangles */
+    __forceinline size_t size() const { return bsf(~movemask(valid()));  }
+
+    /* Returns the geometry IDs */
+    __forceinline       vuint<M>& geomID()       { return geomIDs;  }
+    __forceinline const vuint<M>& geomID() const { return geomIDs;  }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M>& primID()       { return primIDs; }
+    __forceinline const vuint<M>& primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the triangle */
+    __forceinline BBox3fa bounds() const 
+    {
+      Vec3vf<M> p0 = v0;
+      Vec3vf<M> p1 = v0-e1;
+      Vec3vf<M> p2 = v0+e2;
+      Vec3vf<M> lower = min(p0,p1,p2);
+      Vec3vf<M> upper = max(p0,p1,p2);
+      vbool<M> mask = valid();
+      lower.x = select(mask,lower.x,vfloat<M>(pos_inf));
+      lower.y = select(mask,lower.y,vfloat<M>(pos_inf));
+      lower.z = select(mask,lower.z,vfloat<M>(pos_inf));
+      upper.x = select(mask,upper.x,vfloat<M>(neg_inf));
+      upper.y = select(mask,upper.y,vfloat<M>(neg_inf));
+      upper.z = select(mask,upper.z,vfloat<M>(neg_inf));
+      return BBox3fa(Vec3fa(reduce_min(lower.x),reduce_min(lower.y),reduce_min(lower.z)),
+                     Vec3fa(reduce_max(upper.x),reduce_max(upper.y),reduce_max(upper.z)));
+    }
+
+    /* Non temporal store */
+    __forceinline static void store_nt(TriangleM* dst, const TriangleM& src)
+    {
+      vfloat<M>::store_nt(&dst->v0.x,src.v0.x);
+      vfloat<M>::store_nt(&dst->v0.y,src.v0.y);
+      vfloat<M>::store_nt(&dst->v0.z,src.v0.z);
+      vfloat<M>::store_nt(&dst->e1.x,src.e1.x);
+      vfloat<M>::store_nt(&dst->e1.y,src.e1.y);
+      vfloat<M>::store_nt(&dst->e1.z,src.e1.z);
+      vfloat<M>::store_nt(&dst->e2.x,src.e2.x);
+      vfloat<M>::store_nt(&dst->e2.y,src.e2.y);
+      vfloat<M>::store_nt(&dst->e2.z,src.e2.z);
+      vuint<M>::store_nt(&dst->geomIDs,src.geomIDs);
+      vuint<M>::store_nt(&dst->primIDs,src.primIDs);
+    }
+
+    /* Fill triangle from triangle list */
+    __forceinline void fill(const PrimRef* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero;
+      
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+	const PrimRef& prim = prims[begin];
+        const unsigned geomID = prim.geomID();
+        const unsigned primID = prim.primID();
+        const TriangleMesh* __restrict__ const mesh = scene->get<TriangleMesh>(geomID);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+        const Vec3fa& p0 = mesh->vertex(tri.v[0]);
+        const Vec3fa& p1 = mesh->vertex(tri.v[1]);
+        const Vec3fa& p2 = mesh->vertex(tri.v[2]);
+        vgeomID [i] = geomID;
+        vprimID [i] = primID;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+      }
+      TriangleM::store_nt(this,TriangleM(v0,v1,v2,vgeomID,vprimID));
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(TriangleMesh* mesh)
+    {
+      BBox3fa bounds = empty;
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero;
+
+	  for (size_t i=0; i<M; i++)
+      {
+        if (unlikely(geomID(i) == -1)) break;
+        const unsigned geomId = geomID(i);
+        const unsigned primId = primID(i);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primId);
+        const Vec3fa p0 = mesh->vertex(tri.v[0]);
+        const Vec3fa p1 = mesh->vertex(tri.v[1]);
+        const Vec3fa p2 = mesh->vertex(tri.v[2]);
+        bounds.extend(merge(BBox3fa(p0),BBox3fa(p1),BBox3fa(p2)));
+        vgeomID [i] = geomId;
+        vprimID [i] = primId;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+      }
+      TriangleM::store_nt(this,TriangleM(v0,v1,v2,vgeomID,vprimID));
+      return bounds;
+    }
+
+  public:
+    Vec3vf<M> v0;      // base vertex of the triangles
+    Vec3vf<M> e1;      // 1st edge of the triangles (v0-v1)
+    Vec3vf<M> e2;      // 2nd edge of the triangles (v2-v0)
+  private:
+    vuint<M> geomIDs; // geometry IDs
+    vuint<M> primIDs; // primitive IDs
+  };
+
+  template<int M>
+  typename TriangleM<M>::Type TriangleM<M>::type;
+
+  typedef TriangleM<4> Triangle4;
+}

thirdparty/embree/kernels/geometry/triangle_intersector.h

@@ -0,0 +1,96 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "triangle_intersector_moeller.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMIntersector1Moeller
+    {
+      typedef TriangleM<M> Primitive;
+      typedef MoellerTrumboreIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with the M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersectEdge(ray,tri.v0,tri.e1,tri.e2,UVIdentity<M>(),Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersectEdge(ray,tri.v0,tri.e1,tri.e2,UVIdentity<M>(),Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+      
+    };
+
+    /*! Intersects M triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct TriangleMIntersectorKMoeller
+    {
+      typedef TriangleM<M> Primitive;
+      typedef MoellerTrumboreIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        STAT_USER(0,TriangleM<M>::max_size());
+        for (size_t i=0; i<TriangleM<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> e1 = broadcast<vfloat<K>>(tri.e1,i);
+          const Vec3vf<K> e2 = broadcast<vfloat<K>>(tri.e2,i);
+          pre.intersectEdgeK(valid_i,ray,p0,e1,e2,UVIdentity<K>(),IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<TriangleM<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> p0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> e1 = broadcast<vfloat<K>>(tri.e1,i);
+          const Vec3vf<K> e2 = broadcast<vfloat<K>>(tri.e2,i);
+          pre.intersectEdgeK(valid0,ray,p0,e1,e2,UVIdentity<K>(),OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersectEdge(ray,k,tri.v0,tri.e1,tri.e2,UVIdentity<M>(),Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const TriangleM<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersectEdge(ray,k,tri.v0,tri.e1,tri.e2,UVIdentity<M>(),Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/triangle_intersector_moeller.h

@@ -0,0 +1,525 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "intersector_epilog.h"
+
+/*! This intersector implements a modified version of the Moeller
+ *  Trumbore intersector from the paper "Fast, Minimum Storage
+ *  Ray-Triangle Intersection". In contrast to the paper we
+ *  precalculate some factors and factor the calculations differently
+ *  to allow precalculating the cross product e1 x e2. The resulting
+ *  algorithm is similar to the fastest one of the paper "Optimizing
+ *  Ray-Triangle Intersection via Automated Search". */
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M, typename UVMapper>
+    struct MoellerTrumboreHitM
+    {
+      __forceinline MoellerTrumboreHitM(const UVMapper& mapUV) : mapUV(mapUV) {}
+
+      __forceinline MoellerTrumboreHitM(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& T, const vfloat<M>& absDen, const Vec3vf<M>& Ng, const UVMapper& mapUV)
+        : U(U), V(V), T(T), absDen(absDen), mapUV(mapUV), valid(valid), vNg(Ng) {}
+      
+      __forceinline void finalize() 
+      {
+        const vfloat<M> rcpAbsDen = rcp(absDen);
+        vt = T * rcpAbsDen;
+        vu = U * rcpAbsDen;
+        vv = V * rcpAbsDen;
+        mapUV(vu,vv,vNg);
+      }
+
+      __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); }
+      __forceinline vfloat<M> t () const { return vt; }
+      __forceinline Vec3vf<M> Ng() const { return vNg; }
+     
+      __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+      __forceinline float t  (const size_t i) const { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+      
+    public:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> T;
+      vfloat<M> absDen;
+      UVMapper mapUV;
+      
+    public:
+      vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+    };
+    
+    template<int M, bool early_out = true>
+    struct MoellerTrumboreIntersector1
+    {
+      __forceinline MoellerTrumboreIntersector1() {}
+
+      __forceinline MoellerTrumboreIntersector1(const Ray& ray, const void* ptr) {}
+
+      template<typename UVMapper>
+      __forceinline bool intersect(const vbool<M>& valid0,
+                                   Ray& ray,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_e1,
+                                   const Vec3vf<M>& tri_e2,
+                                   const Vec3vf<M>& tri_Ng,
+                                   const UVMapper& mapUV,
+                                   MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        /* calculate denominator */
+        vbool<M> valid = valid0;
+
+        const Vec3vf<M> O = Vec3vf<M>((Vec3fa)ray.org);
+        const Vec3vf<M> D = Vec3vf<M>((Vec3fa)ray.dir);
+        const Vec3vf<M> C = Vec3vf<M>(tri_v0) - O;
+        const Vec3vf<M> R = cross(C,D);
+        const vfloat<M> den = dot(Vec3vf<M>(tri_Ng),D);
+        const vfloat<M> absDen = abs(den);
+        const vfloat<M> sgnDen = signmsk(den);
+
+        /* perform edge tests */
+        const vfloat<M> U = asFloat(asInt(dot(R,Vec3vf<M>(tri_e2))) ^ asInt(sgnDen));
+        const vfloat<M> V = asFloat(asInt(dot(R,Vec3vf<M>(tri_e1))) ^ asInt(sgnDen));
+        
+        /* perform backface culling */        
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= (den < vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#else
+        valid &= (den != vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#endif
+        if (likely(early_out && none(valid))) return false;
+
+        /* perform depth test */
+        const vfloat<M> T = asFloat(asInt(dot(Vec3vf<M>(tri_Ng),C)) ^ asInt(sgnDen));
+
+        valid &= (absDen*vfloat<M>(ray.tnear()) < T) & (T <= absDen*vfloat<M>(ray.tfar));
+        if (likely(early_out && none(valid))) return false;
+           
+        /* update hit information */
+        new (&hit) MoellerTrumboreHitM<M,UVMapper>(valid,U,V,T,absDen,tri_Ng,mapUV);
+
+        return early_out || any(valid);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersectEdge(const vbool<M>& valid,
+                                       Ray& ray,
+                                       const Vec3vf<M>& tri_v0,
+                                       const Vec3vf<M>& tri_e1,
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        const Vec3<vfloat<M>> tri_Ng = cross(tri_e2,tri_e1);
+        return intersect(valid,ray,tri_v0,tri_e1,tri_e2,tri_Ng,mapUV,hit);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersectEdge(Ray& ray,
+                                       const Vec3vf<M>& tri_v0,
+                                       const Vec3vf<M>& tri_e1,
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        vbool<M> valid = true;
+        const Vec3<vfloat<M>> tri_Ng = cross(tri_e2,tri_e1);
+        return intersect(valid,ray,tri_v0,tri_e1,tri_e2,tri_Ng,mapUV,hit);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersect(Ray& ray,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,v0,e1,e2,mapUV,hit);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersect(const vbool<M>& valid,
+                                   Ray& ray,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(valid,ray,v0,e1,e2,mapUV,hit);
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersectEdge(Ray& ray,
+                                       const Vec3vf<M>& v0,
+                                       const Vec3vf<M>& e1,
+                                       const Vec3vf<M>& e2,
+                                       const UVMapper& mapUV,
+                                       const Epilog& epilog) const
+      {
+        MoellerTrumboreHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersectEdge(ray,v0,e1,e2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename UVMapper, typename Epilog>
+        __forceinline bool intersect(Ray& ray,
+                                     const Vec3vf<M>& v0,
+                                     const Vec3vf<M>& v1,
+                                     const Vec3vf<M>& v2,
+                                     const UVMapper& mapUV,
+                                     const Epilog& epilog) const
+      {
+        MoellerTrumboreHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersect(ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename Epilog>
+        __forceinline bool intersect(Ray& ray,
+                                     const Vec3vf<M>& v0,
+                                     const Vec3vf<M>& v1,
+                                     const Vec3vf<M>& v2,
+                                     const Epilog& epilog) const
+      {
+        auto mapUV = UVIdentity<M>();
+        MoellerTrumboreHitM<M,UVIdentity<M>> hit(mapUV);
+        if (likely(intersect(ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersect(const vbool<M>& valid,
+                                   Ray& ray,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   const Epilog& epilog) const
+      {
+        MoellerTrumboreHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersect(valid,ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+    };
+    
+    template<int K, typename UVMapper>
+    struct MoellerTrumboreHitK
+    {
+      __forceinline MoellerTrumboreHitK(const UVMapper& mapUV) : mapUV(mapUV) {}
+      __forceinline MoellerTrumboreHitK(const vfloat<K>& U, const vfloat<K>& V, const vfloat<K>& T, const vfloat<K>& absDen, const Vec3vf<K>& Ng, const UVMapper& mapUV)
+        : U(U), V(V), T(T), absDen(absDen), Ng(Ng), mapUV(mapUV) {}
+      
+      __forceinline std::tuple<vfloat<K>,vfloat<K>,vfloat<K>,Vec3vf<K>> operator() () const
+      {
+        const vfloat<K> rcpAbsDen = rcp(absDen);
+        const vfloat<K> t = T * rcpAbsDen;
+        vfloat<K> u = U * rcpAbsDen;
+        vfloat<K> v = V * rcpAbsDen;
+        Vec3vf<K> vNg = Ng;
+        mapUV(u,v,vNg);
+        return std::make_tuple(u,v,t,vNg);
+      }
+      
+      vfloat<K> U;
+      vfloat<K> V;
+      const vfloat<K> T;
+      const vfloat<K> absDen;
+      const Vec3vf<K> Ng;
+      const UVMapper& mapUV;
+    };
+    
+    template<int M, int K>
+    struct MoellerTrumboreIntersectorK
+    {
+      __forceinline MoellerTrumboreIntersectorK() {}
+      __forceinline MoellerTrumboreIntersectorK(const vbool<K>& valid, const RayK<K>& ray) {}
+      
+      /*! Intersects K rays with one of M triangles. */
+      template<typename UVMapper>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        //RayK<K>& ray,
+                                        const Vec3vf<K>& ray_org,
+                                        const Vec3vf<K>& ray_dir,
+                                        const vfloat<K>& ray_tnear,
+                                        const vfloat<K>& ray_tfar,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_e1,
+                                        const Vec3vf<K>& tri_e2,
+                                        const Vec3vf<K>& tri_Ng,
+                                        const UVMapper& mapUV,
+                                        MoellerTrumboreHitK<K,UVMapper> &hit) const
+      { 
+        /* calculate denominator */
+        vbool<K> valid = valid0;
+        const Vec3vf<K> C = tri_v0 - ray_org;
+        const Vec3vf<K> R = cross(C,ray_dir);
+        const vfloat<K> den = dot(tri_Ng,ray_dir);
+        const vfloat<K> absDen = abs(den);
+        const vfloat<K> sgnDen = signmsk(den);
+        
+        /* test against edge p2 p0 */
+        const vfloat<K> U = dot(tri_e2,R) ^ sgnDen;
+        valid &= U >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p0 p1 */
+        const vfloat<K> V = dot(tri_e1,R) ^ sgnDen;
+        valid &= V >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p1 p2 */
+        const vfloat<K> W = absDen-U-V;
+        valid &= W >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<K> T = dot(tri_Ng,C) ^ sgnDen;
+        valid &= (absDen*ray_tnear < T) & (T <= absDen*ray_tfar);
+        if (unlikely(none(valid))) return false;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= den < vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#else
+        valid &= den != vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#endif
+        
+        /* calculate hit information */
+        new (&hit) MoellerTrumboreHitK<K,UVMapper>(U,V,T,absDen,tri_Ng,mapUV);
+        return valid;
+      }
+
+      /*! Intersects K rays with one of M triangles. */
+      template<typename UVMapper>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0, 
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const UVMapper& mapUV,
+                                        MoellerTrumboreHitK<K,UVMapper> &hit) const
+      {
+        const Vec3vf<K> e1 = tri_v0-tri_v1;
+        const Vec3vf<K> e2 = tri_v2-tri_v0;
+        const Vec3vf<K> Ng = cross(e2,e1);
+        return intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,e1,e2,Ng,mapUV,hit);
+      }
+
+      /*! Intersects K rays with one of M triangles. */
+      template<typename UVMapper, typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0, 
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const UVMapper& mapUV,
+                                        const Epilog& epilog) const
+      {
+        MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV);		
+        const Vec3vf<K> e1 = tri_v0-tri_v1;
+        const Vec3vf<K> e2 = tri_v2-tri_v0;
+        const Vec3vf<K> Ng = cross(e2,e1);
+        const vbool<K> valid = intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,e1,e2,Ng,mapUV,hit);
+	return epilog(valid,hit);
+      }
+
+
+      
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0, 
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const Epilog& epilog) const
+      {
+	UVIdentity<K> mapUV;	
+        MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV);			
+        const Vec3vf<K> e1 = tri_v0-tri_v1;
+        const Vec3vf<K> e2 = tri_v2-tri_v0;
+        const Vec3vf<K> Ng = cross(e2,e1);
+        const vbool<K> valid = intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,e1,e2,Ng,mapUV,hit);
+	return epilog(valid,hit);
+      }
+
+      /*! Intersects K rays with one of M triangles. */
+      template<typename UVMapper, typename Epilog>
+      __forceinline vbool<K> intersectEdgeK(const vbool<K>& valid0, 
+                                            RayK<K>& ray,
+                                            const Vec3vf<K>& tri_v0, 
+                                            const Vec3vf<K>& tri_e1, 
+                                            const Vec3vf<K>& tri_e2,
+                                            const UVMapper& mapUV,
+                                            const Epilog& epilog) const
+      {
+        MoellerTrumboreHitK<K,UVIdentity<K>> hit(mapUV);			
+        const Vec3vf<K> tri_Ng = cross(tri_e2,tri_e1);
+        const vbool<K> valid = intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,tri_e1,tri_e2,tri_Ng,mapUV,hit);
+	return epilog(valid,hit);
+      }
+      
+      /*! Intersect k'th ray from ray packet of size K with M triangles. */
+      template<typename UVMapper>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const Vec3vf<M>& tri_v0,
+                                       const Vec3vf<M>& tri_e1,
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        /* calculate denominator */
+        typedef Vec3vf<M> Vec3vfM;
+        const Vec3vf<M> tri_Ng = cross(tri_e2,tri_e1);
+
+        const Vec3vfM O = broadcast<vfloat<M>>(ray.org,k);
+        const Vec3vfM D = broadcast<vfloat<M>>(ray.dir,k);
+        const Vec3vfM C = Vec3vfM(tri_v0) - O;
+        const Vec3vfM R = cross(C,D);
+        const vfloat<M> den = dot(Vec3vfM(tri_Ng),D);
+        const vfloat<M> absDen = abs(den);
+        const vfloat<M> sgnDen = signmsk(den);
+        
+        /* perform edge tests */
+        const vfloat<M> U = dot(Vec3vf<M>(tri_e2),R) ^ sgnDen;
+        const vfloat<M> V = dot(Vec3vf<M>(tri_e1),R) ^ sgnDen;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        vbool<M> valid = (den < vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#else
+        vbool<M> valid = (den != vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#endif
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<M> T = dot(Vec3vf<M>(tri_Ng),C) ^ sgnDen;
+        valid &= (absDen*vfloat<M>(ray.tnear()[k]) < T) & (T <= absDen*vfloat<M>(ray.tfar[k]));
+        if (likely(none(valid))) return false;
+        
+        /* calculate hit information */
+        new (&hit) MoellerTrumboreHitM<M,UVMapper>(valid,U,V,T,absDen,tri_Ng,mapUV);
+        return true;
+      }
+
+       template<typename UVMapper>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const BBox<vfloat<M>>& time_range,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       MoellerTrumboreHitM<M,UVMapper>& hit) const
+      {
+        if (likely(intersect(ray,k,tri_v0,tri_e1,tri_e2,mapUV,hit))) 
+        {
+          hit.valid &= time_range.lower <= vfloat<M>(ray.time[k]);
+          hit.valid &= vfloat<M>(ray.time[k]) < time_range.upper;
+          return any(hit.valid);
+        }
+        return false;
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const Vec3vf<M>& v0, 
+                                   const Vec3vf<M>& v1, 
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   MoellerTrumboreHitM<M,UVMapper>& hit) const      
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,k,v0,e1,e2,mapUV,hit);
+      }
+      
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       const Epilog& epilog) const
+      {
+        MoellerTrumboreHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersectEdge(ray,k,tri_v0,tri_e1,tri_e2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,                           
+                                       const BBox<vfloat<M>>& time_range,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2,
+                                       const UVMapper& mapUV,
+                                       const Epilog& epilog) const
+      {
+        MoellerTrumboreHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersectEdge(ray,k,time_range,tri_v0,tri_e1,tri_e2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+      
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const Vec3vf<M>& v0, 
+                                   const Vec3vf<M>& v1, 
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   const Epilog& epilog) const      
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,k,v0,e1,e2,mapUV,epilog);
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const Vec3vf<M>& v0, 
+                                   const Vec3vf<M>& v1, 
+                                   const Vec3vf<M>& v2,
+                                   const Epilog& epilog) const      
+      {
+        return intersect(ray,k,v0,v1,v2,UVIdentity<M>(),epilog);
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const BBox<vfloat<M>>& time_range,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   const Epilog& epilog) const
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,k,time_range,v0,e1,e2,mapUV,epilog);
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/triangle_intersector_pluecker.h

@@ -0,0 +1,407 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "trianglev.h"
+#include "trianglev_mb.h"
+#include "intersector_epilog.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
+  {
+    template<int M, typename UVMapper>
+    struct PlueckerHitM
+    {
+      __forceinline PlueckerHitM(const UVMapper& mapUV) : mapUV(mapUV) {}
+      
+      __forceinline PlueckerHitM(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& UVW, const vfloat<M>& t, const Vec3vf<M>& Ng, const UVMapper& mapUV)
+        :  U(U), V(V), UVW(UVW), mapUV(mapUV), valid(valid), vt(t), vNg(Ng) {}
+      
+      __forceinline void finalize() 
+      {
+        const vbool<M> invalid = abs(UVW) < min_rcp_input;
+        const vfloat<M> rcpUVW = select(invalid,vfloat<M>(0.0f),rcp(UVW));
+        vu = min(U * rcpUVW,1.0f);
+        vv = min(V * rcpUVW,1.0f);	
+        mapUV(vu,vv,vNg);
+      }
+
+      __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); }
+      __forceinline vfloat<M> t () const { return vt; }
+      __forceinline Vec3vf<M> Ng() const { return vNg; }
+    
+      __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+      __forceinline float t  (const size_t i) const { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+      
+    public:
+      vfloat<M> U;
+      vfloat<M> V;
+      vfloat<M> UVW;
+      const UVMapper& mapUV;
+      
+    public:
+      vbool<M> valid;      
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+    };
+
+    template<int M, bool early_out = true>
+    struct PlueckerIntersector1
+    {
+      __forceinline PlueckerIntersector1() {}
+
+      __forceinline PlueckerIntersector1(const Ray& ray, const void* ptr) {}
+
+      template<typename UVMapper>
+      __forceinline bool intersect(const vbool<M>& valid0,
+                                   Ray& ray,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_v1,
+                                   const Vec3vf<M>& tri_v2,
+                                   const UVMapper& mapUV,
+				   PlueckerHitM<M,UVMapper>& hit) const
+      {
+        vbool<M> valid = valid0;
+        
+        /* calculate vertices relative to ray origin */
+        const Vec3vf<M> O = Vec3vf<M>((Vec3fa)ray.org);
+        const Vec3vf<M> D = Vec3vf<M>((Vec3fa)ray.dir);
+        const Vec3vf<M> v0 = tri_v0-O;
+        const Vec3vf<M> v1 = tri_v1-O;
+        const Vec3vf<M> v2 = tri_v2-O;
+
+        /* calculate triangle edges */
+        const Vec3vf<M> e0 = v2-v0;
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v1-v2;
+
+        /* perform edge tests */
+        const vfloat<M> U = dot(cross(e0,v2+v0),D);
+        const vfloat<M> V = dot(cross(e1,v0+v1),D);
+        const vfloat<M> W = dot(cross(e2,v1+v2),D);
+        const vfloat<M> UVW = U+V+W;
+        const vfloat<M> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= max(U,V,W) <= eps;
+#else
+        valid &= (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif
+        if (unlikely(early_out && none(valid))) return false;
+
+        /* calculate geometry normal and denominator */
+        const Vec3vf<M> Ng = stable_triangle_normal(e0,e1,e2);
+        const vfloat<M> den = twice(dot(Ng,D));
+        
+        /* perform depth test */
+        const vfloat<M> T = twice(dot(v0,Ng));
+        const vfloat<M> t = rcp(den)*T;
+        valid &= vfloat<M>(ray.tnear()) <= t & t <= vfloat<M>(ray.tfar);
+        valid &= den != vfloat<M>(zero);
+        if (unlikely(early_out && none(valid))) return false;
+
+        /* update hit information */
+        new (&hit) PlueckerHitM<M,UVMapper>(valid,U,V,UVW,t,Ng,mapUV);
+        return early_out || any(valid);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersectEdge(const vbool<M>& valid,
+				       Ray& ray,
+				       const Vec3vf<M>& tri_v0,
+				       const Vec3vf<M>& tri_v1,
+				       const Vec3vf<M>& tri_v2,
+				       const UVMapper& mapUV,
+				       PlueckerHitM<M,UVMapper>& hit) const
+      {
+        return intersect(valid,ray,tri_v0,tri_v1,tri_v2,mapUV,hit);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersectEdge(Ray& ray,
+				       const Vec3vf<M>& tri_v0,
+				       const Vec3vf<M>& tri_v1,
+				       const Vec3vf<M>& tri_v2,
+				       const UVMapper& mapUV,				       
+				       PlueckerHitM<M,UVMapper>& hit) const
+      {
+	vbool<M> valid = true;
+        return intersect(valid,ray,tri_v0,tri_v1,tri_v2,mapUV,hit);
+      }
+
+      template<typename UVMapper>
+      __forceinline bool intersect(Ray& ray,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_v1,
+                                   const Vec3vf<M>& tri_v2,
+                                   const UVMapper& mapUV,				   
+                                   PlueckerHitM<M,UVMapper>& hit) const
+      {
+        return intersectEdge(ray,tri_v0,tri_v1,tri_v2,mapUV,hit);
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersectEdge(Ray& ray,
+                                       const Vec3vf<M>& v0,
+                                       const Vec3vf<M>& e1,
+                                       const Vec3vf<M>& e2,
+                                       const UVMapper& mapUV,
+                                       const Epilog& epilog) const
+      {
+        PlueckerHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersectEdge(ray,v0,e1,e2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename UVMapper, typename Epilog>
+        __forceinline bool intersect(Ray& ray,
+                                     const Vec3vf<M>& v0,
+                                     const Vec3vf<M>& v1,
+                                     const Vec3vf<M>& v2,
+                                     const UVMapper& mapUV,
+                                     const Epilog& epilog) const
+      {
+        PlueckerHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersect(ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename Epilog>
+        __forceinline bool intersect(Ray& ray,
+                                     const Vec3vf<M>& v0,
+                                     const Vec3vf<M>& v1,
+                                     const Vec3vf<M>& v2,
+                                     const Epilog& epilog) const
+      {
+        auto mapUV = UVIdentity<M>();
+        PlueckerHitM<M,UVIdentity<M>> hit(mapUV);
+        if (likely(intersect(ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersect(const vbool<M>& valid,
+                                   Ray& ray,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const UVMapper& mapUV,
+                                   const Epilog& epilog) const
+      {
+        PlueckerHitM<M,UVMapper> hit(mapUV);
+        if (likely(intersect(valid,ray,v0,v1,v2,mapUV,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+      
+    };
+
+    template<int K, typename UVMapper>
+    struct PlueckerHitK
+    {
+      __forceinline PlueckerHitK(const UVMapper& mapUV) : mapUV(mapUV) {}
+      
+      __forceinline PlueckerHitK(const vfloat<K>& U, const vfloat<K>& V, const vfloat<K>& UVW, const vfloat<K>& t, const Vec3vf<K>& Ng, const UVMapper& mapUV)
+        :  U(U), V(V), UVW(UVW), t(t), Ng(Ng), mapUV(mapUV) {}
+      
+      __forceinline std::tuple<vfloat<K>,vfloat<K>,vfloat<K>,Vec3vf<K>> operator() () const
+      {
+        const vbool<K> invalid = abs(UVW) < min_rcp_input;
+        const vfloat<K> rcpUVW = select(invalid,vfloat<K>(0.0f),rcp(UVW));
+        vfloat<K> u = min(U * rcpUVW,1.0f);
+        vfloat<K> v = min(V * rcpUVW,1.0f);
+        Vec3vf<K> vNg = Ng;
+        mapUV(u,v,vNg);
+        return std::make_tuple(u,v,t,vNg);
+      }
+      vfloat<K> U;
+      vfloat<K> V;
+      const vfloat<K> UVW;
+      const vfloat<K> t;
+      const Vec3vf<K> Ng;
+      const UVMapper& mapUV;
+    };
+    
+    template<int M, int K>
+    struct PlueckerIntersectorK
+    {
+      __forceinline PlueckerIntersectorK() {}      
+      __forceinline PlueckerIntersectorK(const vbool<K>& valid, const RayK<K>& ray) {}
+
+      /*! Intersects K rays with one of M triangles. */
+      template<typename UVMapper>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+				    RayK<K>& ray,
+				    const Vec3vf<K>& tri_v0,
+				    const Vec3vf<K>& tri_v1,
+				    const Vec3vf<K>& tri_v2,
+				    const UVMapper& mapUV,
+				    PlueckerHitK<K,UVMapper> &hit) const
+      {
+        /* calculate vertices relative to ray origin */
+        vbool<K> valid = valid0;
+        const Vec3vf<K> O = ray.org;
+        const Vec3vf<K> D = ray.dir;
+        const Vec3vf<K> v0 = tri_v0-O;
+        const Vec3vf<K> v1 = tri_v1-O;
+        const Vec3vf<K> v2 = tri_v2-O;
+
+        /* calculate triangle edges */
+        const Vec3vf<K> e0 = v2-v0;
+        const Vec3vf<K> e1 = v0-v1;
+        const Vec3vf<K> e2 = v1-v2;
+
+        /* perform edge tests */
+        const vfloat<K> U = dot(Vec3vf<K>(cross(e0,v2+v0)),D);
+        const vfloat<K> V = dot(Vec3vf<K>(cross(e1,v0+v1)),D);
+        const vfloat<K> W = dot(Vec3vf<K>(cross(e2,v1+v2)),D);
+        const vfloat<K> UVW = U+V+W;
+        const vfloat<K> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= max(U,V,W) <= eps;
+#else
+        valid &= (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif
+        if (unlikely(none(valid))) return valid;
+
+         /* calculate geometry normal and denominator */
+        const Vec3vf<K> Ng = stable_triangle_normal(e0,e1,e2);
+        const vfloat<K> den = twice(dot(Vec3vf<K>(Ng),D));
+
+        /* perform depth test */
+        const vfloat<K> T = twice(dot(v0,Vec3vf<K>(Ng)));
+        const vfloat<K> t = rcp(den)*T;
+        valid &= ray.tnear() <= t & t <= ray.tfar;
+        valid &= den != vfloat<K>(zero);
+        if (unlikely(none(valid))) return valid;
+        
+        /* calculate hit information */
+        new (&hit) PlueckerHitK<K,UVMapper>(U,V,UVW,t,Ng,mapUV);
+        return valid;
+      }
+
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const Epilog& epilog) const
+      {
+	UVIdentity<K> mapUV;	
+        PlueckerHitK<K,UVIdentity<K>> hit(mapUV);		
+        const vbool<K> valid = intersectK(valid0,ray,tri_v0,tri_v1,tri_v2,mapUV,hit);
+	return epilog(valid,hit);
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+					const UVMapper& mapUV,
+                                        const Epilog& epilog) const
+      {
+        PlueckerHitK<K,UVMapper> hit(mapUV);		
+        const vbool<K> valid = intersectK(valid0,ray,tri_v0,tri_v1,tri_v2,mapUV,hit);
+	return epilog(valid,hit);
+      }
+      
+      /*! Intersect k'th ray from ray packet of size K with M triangles. */
+      template<typename UVMapper>
+      __forceinline bool intersect(RayK<K>& ray, size_t k,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_v1,
+                                   const Vec3vf<M>& tri_v2,
+                                   const UVMapper& mapUV,
+				   PlueckerHitM<M,UVMapper> &hit) const
+      {
+        /* calculate vertices relative to ray origin */
+        const Vec3vf<M> O = broadcast<vfloat<M>>(ray.org,k);
+        const Vec3vf<M> D = broadcast<vfloat<M>>(ray.dir,k);
+        const Vec3vf<M> v0 = tri_v0-O;
+        const Vec3vf<M> v1 = tri_v1-O;
+        const Vec3vf<M> v2 = tri_v2-O;
+
+        /* calculate triangle edges */
+        const Vec3vf<M> e0 = v2-v0;
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v1-v2;
+
+	
+        /* perform edge tests */
+        const vfloat<M> U = dot(cross(e0,v2+v0),D);
+        const vfloat<M> V = dot(cross(e1,v0+v1),D);
+        const vfloat<M> W = dot(cross(e2,v1+v2),D);
+	
+        const vfloat<M> UVW = U+V+W;
+        const vfloat<M> eps = float(ulp)*abs(UVW);
+#if defined(EMBREE_BACKFACE_CULLING)
+        vbool<M> valid = max(U,V,W) <= eps;
+#else
+        vbool<M> valid = (min(U,V,W) >= -eps) | (max(U,V,W) <= eps);
+#endif
+        if (unlikely(none(valid))) return false;
+
+        /* calculate geometry normal and denominator */
+        const Vec3vf<M> Ng = stable_triangle_normal(e0,e1,e2);
+        const vfloat<M> den = twice(dot(Ng,D));
+        
+        /* perform depth test */
+        const vfloat<M> T = twice(dot(v0,Ng));
+        const vfloat<M> t = rcp(den)*T;
+        valid &= vfloat<M>(ray.tnear()[k]) <= t & t <= vfloat<M>(ray.tfar[k]);
+        if (unlikely(none(valid))) return false;
+
+        /* avoid division by 0 */
+        valid &= den != vfloat<M>(zero);
+        if (unlikely(none(valid))) return false;
+
+        /* update hit information */
+        new (&hit) PlueckerHitM<M,UVMapper>(valid,U,V,UVW,t,Ng,mapUV);
+        return true;
+      }
+
+      template<typename UVMapper, typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray, size_t k,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_v1,
+                                   const Vec3vf<M>& tri_v2,
+                                   const UVMapper& mapUV,				   
+                                   const Epilog& epilog) const
+      {
+        PlueckerHitM<M,UVMapper> hit(mapUV);	
+        if (intersect(ray,k,tri_v0,tri_v1,tri_v2,mapUV,hit))
+	  return epilog(hit.valid,hit);
+	return false;
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray, size_t k,
+                                   const Vec3vf<M>& tri_v0,
+                                   const Vec3vf<M>& tri_v1,
+                                   const Vec3vf<M>& tri_v2,
+                                   const Epilog& epilog) const
+      {
+	UVIdentity<M> mapUV;	
+        PlueckerHitM<M,UVIdentity<M>> hit(mapUV);	
+        if (intersect(ray,k,tri_v0,tri_v1,tri_v2,mapUV,hit))
+	  return epilog(hit.valid,hit);
+	return false;
+      }
+      
+    };
+  }
+}

thirdparty/embree/kernels/geometry/triangle_intersector_woop.h

@@ -0,0 +1,418 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "intersector_epilog.h"
+
+/*! This intersector implements a modified version of the Woop's ray-triangle intersection test */
+
+namespace embree
+{
+  namespace isa
+  {
+    template<int M>
+    struct WoopHitM
+    {
+      __forceinline WoopHitM() {}
+
+      __forceinline WoopHitM(const vbool<M>& valid, 
+                             const vfloat<M>& U, 
+                             const vfloat<M>& V, 
+                             const vfloat<M>& T, 
+                             const vfloat<M>& inv_det,                              
+                             const Vec3vf<M>& Ng)
+        : U(U), V(V), T(T), inv_det(inv_det), valid(valid), vNg(Ng) {}
+      
+      __forceinline void finalize() 
+      {
+        vt = T;
+        vu = U*inv_det;
+        vv = V*inv_det;
+      }
+
+      __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); }
+      __forceinline float t  (const size_t i) const { return vt[i]; }
+      __forceinline Vec3fa Ng(const size_t i) const { return Vec3fa(vNg.x[i],vNg.y[i],vNg.z[i]); }
+      
+    private:
+      const vfloat<M> U;
+      const vfloat<M> V;
+      const vfloat<M> T;
+      const vfloat<M> inv_det;
+      
+    public:
+      const vbool<M> valid;
+      vfloat<M> vu;
+      vfloat<M> vv;
+      vfloat<M> vt;
+      Vec3vf<M> vNg;
+    };
+
+    template<int M>
+    struct WoopPrecalculations1
+    {
+      unsigned int kx,ky,kz;
+      Vec3vf<M> org;
+      Vec3fa S;
+      __forceinline WoopPrecalculations1() {}
+
+      __forceinline WoopPrecalculations1(const Ray& ray, const void* ptr)
+      {
+        kz = maxDim(abs(ray.dir));
+        kx = (kz+1) % 3;
+        ky = (kx+1) % 3;
+        const float inv_dir_kz = rcp(ray.dir[kz]);
+        if (ray.dir[kz] < 0.0f) std::swap(kx,ky);
+        S.x = ray.dir[kx] * inv_dir_kz;
+        S.y = ray.dir[ky] * inv_dir_kz;
+        S.z = inv_dir_kz;
+        org = Vec3vf<M>(ray.org[kx],ray.org[ky],ray.org[kz]);
+      }
+    };
+
+    
+    template<int M>
+    struct WoopIntersector1
+    {
+
+        typedef WoopPrecalculations1<M> Precalculations;
+
+      __forceinline WoopIntersector1() {}
+
+      __forceinline WoopIntersector1(const Ray& ray, const void* ptr) {}
+
+      static __forceinline bool intersect(const vbool<M>& valid0,
+                                          Ray& ray,
+                                          const Precalculations& pre,
+                                          const Vec3vf<M>& tri_v0,
+                                          const Vec3vf<M>& tri_v1,
+                                          const Vec3vf<M>& tri_v2,
+                                          WoopHitM<M>& hit)
+      {       
+        vbool<M> valid = valid0;
+
+        /* vertices relative to ray origin */
+        const Vec3vf<M> org = Vec3vf<M>(pre.org.x,pre.org.y,pre.org.z);
+        const Vec3vf<M> A = Vec3vf<M>(tri_v0[pre.kx],tri_v0[pre.ky],tri_v0[pre.kz]) - org;
+        const Vec3vf<M> B = Vec3vf<M>(tri_v1[pre.kx],tri_v1[pre.ky],tri_v1[pre.kz]) - org;
+        const Vec3vf<M> C = Vec3vf<M>(tri_v2[pre.kx],tri_v2[pre.ky],tri_v2[pre.kz]) - org;
+
+        /* shear and scale vertices */
+        const vfloat<M> Ax = nmadd(A.z,pre.S.x,A.x);
+        const vfloat<M> Ay = nmadd(A.z,pre.S.y,A.y);
+        const vfloat<M> Bx = nmadd(B.z,pre.S.x,B.x);
+        const vfloat<M> By = nmadd(B.z,pre.S.y,B.y);
+        const vfloat<M> Cx = nmadd(C.z,pre.S.x,C.x);
+        const vfloat<M> Cy = nmadd(C.z,pre.S.y,C.y);
+
+        /* scaled barycentric */
+        const vfloat<M> U0 = Cx*By;
+        const vfloat<M> U1 = Cy*Bx;
+        const vfloat<M> V0 = Ax*Cy;
+        const vfloat<M> V1 = Ay*Cx;
+        const vfloat<M> W0 = Bx*Ay;
+        const vfloat<M> W1 = By*Ax;
+#if !defined(__AVX512F__)
+        valid &= (U0 >= U1) & (V0 >= V1) & (W0 >= W1) |
+          (U0 <= U1) & (V0 <= V1) & (W0 <= W1);
+#else
+        valid &= ge(ge(U0 >= U1,V0,V1),W0,W1) | le(le(U0 <= U1,V0,V1),W0,W1);
+#endif
+
+        if (likely(none(valid))) return false;
+        const vfloat<M> U = U0-U1;
+        const vfloat<M> V = V0-V1;
+        const vfloat<M> W = W0-W1;
+
+        const vfloat<M> det = U+V+W;
+
+        valid &= det != 0.0f;
+        const vfloat<M> inv_det = rcp(det);
+
+        const vfloat<M> Az = pre.S.z * A.z;
+        const vfloat<M> Bz = pre.S.z * B.z;
+        const vfloat<M> Cz = pre.S.z * C.z;
+        const vfloat<M> T  = madd(U,Az,madd(V,Bz,W*Cz)); 
+        const vfloat<M> t  = T * inv_det;
+        /* perform depth test */
+        valid &= (vfloat<M>(ray.tnear()) < t) & (t <= vfloat<M>(ray.tfar));
+        if (likely(none(valid))) return false;
+        
+        const Vec3vf<M> tri_Ng = cross(tri_v2-tri_v0,tri_v0-tri_v1);
+
+        /* update hit information */
+        new (&hit) WoopHitM<M>(valid,U,V,t,inv_det,tri_Ng);
+        return true;
+      }
+      
+      static __forceinline bool intersect(Ray& ray,
+                                   const Precalculations& pre,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   WoopHitM<M>& hit)
+      {
+        vbool<M> valid = true;
+        return intersect(valid,ray,pre,v0,v1,v2,hit);
+      }
+
+
+      template<typename Epilog>
+      static __forceinline bool intersect(Ray& ray,
+                                     const Precalculations& pre,
+                                     const Vec3vf<M>& v0,
+                                     const Vec3vf<M>& v1,
+                                     const Vec3vf<M>& v2,
+                                     const Epilog& epilog)
+      {
+        WoopHitM<M> hit;
+        if (likely(intersect(ray,pre,v0,v1,v2,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename Epilog>
+      static __forceinline bool intersect(const vbool<M>& valid,
+                                   Ray& ray,
+                                   const Precalculations& pre,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const Epilog& epilog)
+      {
+        WoopHitM<M> hit;
+        if (likely(intersect(valid,ray,pre,v0,v1,v2,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+    };
+    
+#if 0
+    template<int K>
+    struct WoopHitK
+    {
+      __forceinline WoopHitK(const vfloat<K>& U, const vfloat<K>& V, const vfloat<K>& T, const vfloat<K>& absDen, const Vec3vf<K>& Ng)
+        : U(U), V(V), T(T), absDen(absDen), Ng(Ng) {}
+      
+      __forceinline std::tuple<vfloat<K>,vfloat<K>,vfloat<K>,Vec3vf<K>> operator() () const
+      {
+        const vfloat<K> rcpAbsDen = rcp(absDen);
+        const vfloat<K> t = T * rcpAbsDen;
+        const vfloat<K> u = U * rcpAbsDen;
+        const vfloat<K> v = V * rcpAbsDen;
+        return std::make_tuple(u,v,t,Ng);
+      }
+      
+    private:
+      const vfloat<K> U;
+      const vfloat<K> V;
+      const vfloat<K> T;
+      const vfloat<K> absDen;
+      const Vec3vf<K> Ng;
+    };
+    
+    template<int M, int K>
+    struct WoopIntersectorK
+    {
+      __forceinline WoopIntersectorK(const vbool<K>& valid, const RayK<K>& ray) {}
+      
+      /*! Intersects K rays with one of M triangles. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0,
+                                        //RayK<K>& ray,
+                                        const Vec3vf<K>& ray_org,
+                                        const Vec3vf<K>& ray_dir,
+                                        const vfloat<K>& ray_tnear,
+                                        const vfloat<K>& ray_tfar,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_e1,
+                                        const Vec3vf<K>& tri_e2,
+                                        const Vec3vf<K>& tri_Ng,
+                                        const Epilog& epilog) const
+      { 
+        /* calculate denominator */
+        vbool<K> valid = valid0;
+        const Vec3vf<K> C = tri_v0 - ray_org;
+        const Vec3vf<K> R = cross(C,ray_dir);
+        const vfloat<K> den = dot(tri_Ng,ray_dir);
+        const vfloat<K> absDen = abs(den);
+        const vfloat<K> sgnDen = signmsk(den);
+        
+        /* test against edge p2 p0 */
+        const vfloat<K> U = dot(tri_e2,R) ^ sgnDen;
+        valid &= U >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p0 p1 */
+        const vfloat<K> V = dot(tri_e1,R) ^ sgnDen;
+        valid &= V >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* test against edge p1 p2 */
+        const vfloat<K> W = absDen-U-V;
+        valid &= W >= 0.0f;
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<K> T = dot(tri_Ng,C) ^ sgnDen;
+        valid &= (absDen*ray_tnear < T) & (T <= absDen*ray_tfar);
+        if (unlikely(none(valid))) return false;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        valid &= den < vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#else
+        valid &= den != vfloat<K>(zero);
+        if (unlikely(none(valid))) return false;
+#endif
+        
+        /* calculate hit information */
+        WoopHitK<K> hit(U,V,T,absDen,tri_Ng);
+        return epilog(valid,hit);
+      }
+      
+      /*! Intersects K rays with one of M triangles. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectK(const vbool<K>& valid0, 
+                                        RayK<K>& ray,
+                                        const Vec3vf<K>& tri_v0,
+                                        const Vec3vf<K>& tri_v1,
+                                        const Vec3vf<K>& tri_v2,
+                                        const Epilog& epilog) const
+      {
+        const Vec3vf<K> e1 = tri_v0-tri_v1;
+        const Vec3vf<K> e2 = tri_v2-tri_v0;
+        const Vec3vf<K> Ng = cross(e2,e1);
+        return intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,e1,e2,Ng,epilog);
+      }
+
+      /*! Intersects K rays with one of M triangles. */
+      template<typename Epilog>
+      __forceinline vbool<K> intersectEdgeK(const vbool<K>& valid0, 
+                                            RayK<K>& ray,
+                                            const Vec3vf<K>& tri_v0, 
+                                            const Vec3vf<K>& tri_e1, 
+                                            const Vec3vf<K>& tri_e2, 
+                                            const Epilog& epilog) const
+      {
+        const Vec3vf<K> tri_Ng = cross(tri_e2,tri_e1);
+        return intersectK(valid0,ray.org,ray.dir,ray.tnear(),ray.tfar,tri_v0,tri_e1,tri_e2,tri_Ng,epilog);
+      }
+      
+      /*! Intersect k'th ray from ray packet of size K with M triangles. */
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const Vec3vf<M>& tri_v0,
+                                       const Vec3vf<M>& tri_e1,
+                                       const Vec3vf<M>& tri_e2,
+                                       WoopHitM<M>& hit) const
+      {
+        /* calculate denominator */
+        typedef Vec3vf<M> Vec3vfM;
+        const Vec3vf<M> tri_Ng = cross(tri_e2,tri_e1);
+
+        const Vec3vfM O = broadcast<vfloat<M>>(ray.org,k);
+        const Vec3vfM D = broadcast<vfloat<M>>(ray.dir,k);
+        const Vec3vfM C = Vec3vfM(tri_v0) - O;
+        const Vec3vfM R = cross(C,D);
+        const vfloat<M> den = dot(Vec3vfM(tri_Ng),D);
+        const vfloat<M> absDen = abs(den);
+        const vfloat<M> sgnDen = signmsk(den);
+        
+        /* perform edge tests */
+        const vfloat<M> U = dot(Vec3vf<M>(tri_e2),R) ^ sgnDen;
+        const vfloat<M> V = dot(Vec3vf<M>(tri_e1),R) ^ sgnDen;
+        
+        /* perform backface culling */
+#if defined(EMBREE_BACKFACE_CULLING)
+        vbool<M> valid = (den < vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#else
+        vbool<M> valid = (den != vfloat<M>(zero)) & (U >= 0.0f) & (V >= 0.0f) & (U+V<=absDen);
+#endif
+        if (likely(none(valid))) return false;
+        
+        /* perform depth test */
+        const vfloat<M> T = dot(Vec3vf<M>(tri_Ng),C) ^ sgnDen;
+        valid &= (absDen*vfloat<M>(ray.tnear()[k]) < T) & (T <= absDen*vfloat<M>(ray.tfar[k]));
+        if (likely(none(valid))) return false;
+        
+        /* calculate hit information */
+        new (&hit) WoopHitM<M>(valid,U,V,T,absDen,tri_Ng);
+        return true;
+      }
+
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const BBox<vfloat<M>>& time_range,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2, 
+                                       WoopHitM<M>& hit) const
+      {
+        if (likely(intersect(ray,k,tri_v0,tri_e1,tri_e2,hit))) 
+        {
+          hit.valid &= time_range.lower <= vfloat<M>(ray.time[k]);
+          hit.valid &= vfloat<M>(ray.time[k]) < time_range.upper;
+          return any(hit.valid);
+        }
+        return false;
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2, 
+                                       const Epilog& epilog) const
+      {
+        WoopHitM<M> hit;
+        if (likely(intersectEdge(ray,k,tri_v0,tri_e1,tri_e2,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersectEdge(RayK<K>& ray,
+                                       size_t k,                           
+                                       const BBox<vfloat<M>>& time_range,
+                                       const Vec3vf<M>& tri_v0, 
+                                       const Vec3vf<M>& tri_e1, 
+                                       const Vec3vf<M>& tri_e2, 
+                                       const Epilog& epilog) const
+      {
+        WoopHitM<M> hit;
+        if (likely(intersectEdge(ray,k,time_range,tri_v0,tri_e1,tri_e2,hit))) return epilog(hit.valid,hit);
+        return false;
+      }
+      
+      template<typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const Vec3vf<M>& v0, 
+                                   const Vec3vf<M>& v1, 
+                                   const Vec3vf<M>& v2, 
+                                   const Epilog& epilog) const      
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,k,v0,e1,e2,epilog);
+      }
+
+      template<typename Epilog>
+      __forceinline bool intersect(RayK<K>& ray,
+                                   size_t k,
+                                   const BBox<vfloat<M>>& time_range,
+                                   const Vec3vf<M>& v0,
+                                   const Vec3vf<M>& v1,
+                                   const Vec3vf<M>& v2,
+                                   const Epilog& epilog) const
+      {
+        const Vec3vf<M> e1 = v0-v1;
+        const Vec3vf<M> e2 = v2-v0;
+        return intersectEdge(ray,k,time_range,v0,e1,e2,epilog);
+      }
+    };
+#endif
+  }
+}

thirdparty/embree/kernels/geometry/triangle_triangle_intersector.h

@@ -0,0 +1,132 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#include "primitive.h"
+
+namespace embree
+{ 
+  namespace isa
+  {
+    struct TriangleTriangleIntersector
+    {
+      __forceinline static float T(float pa0, float pa1, float da0, float da1) {
+        return pa0 + (pa1-pa0)*da0/(da0-da1);
+      }
+      
+      __forceinline static bool point_line_side(const Vec2f& p, const Vec2f& a0, const Vec2f& a1) {
+        return det(p-a0,a0-a1) >= 0.0f;
+      }
+      
+      __forceinline static bool point_inside_triangle(const Vec2f& p, const Vec2f& a, const Vec2f& b, const Vec2f& c) 
+      {
+        const bool pab = point_line_side(p,a,b); 
+        const bool pbc = point_line_side(p,b,c);
+        const bool pca = point_line_side(p,c,a);
+        return pab == pbc && pab == pca;
+      }
+      
+      __forceinline static bool intersect_line_line(const Vec2f& a0, const Vec2f& a1, const Vec2f& b0, const Vec2f& b1)
+      {
+        const bool different_sides0 = point_line_side(b0,a0,a1) != point_line_side(b1,a0,a1);
+        const bool different_sides1 = point_line_side(a0,b0,b1) != point_line_side(a1,b0,b1);
+        return different_sides0 && different_sides1;
+      }
+      
+      __forceinline static bool intersect_triangle_triangle (const Vec2f& a0, const Vec2f& a1, const Vec2f& a2, 
+                                                             const Vec2f& b0, const Vec2f& b1, const Vec2f& b2)
+      {
+        const bool a01_b01 = intersect_line_line(a0,a1,b0,b1); 
+        if (a01_b01) return true;
+        const bool a01_b12 = intersect_line_line(a0,a1,b1,b2);
+        if (a01_b12) return true;
+        const bool a01_b20 = intersect_line_line(a0,a1,b2,b0);
+        if (a01_b20) return true;
+        const bool a12_b01 = intersect_line_line(a1,a2,b0,b1);
+        if (a12_b01) return true;
+        const bool a12_b12 = intersect_line_line(a1,a2,b1,b2);
+        if (a12_b12) return true;
+        const bool a12_b20 = intersect_line_line(a1,a2,b2,b0);
+        if (a12_b20) return true;
+        const bool a20_b01 = intersect_line_line(a2,a0,b0,b1);
+        if (a20_b01) return true;
+        const bool a20_b12 = intersect_line_line(a2,a0,b1,b2);
+        if (a20_b12) return true;
+        const bool a20_b20 = intersect_line_line(a2,a0,b2,b0);
+        if (a20_b20) return true;
+        
+        bool a_in_b = point_inside_triangle(a0,b0,b1,b2) && point_inside_triangle(a1,b0,b1,b2) && point_inside_triangle(a2,b0,b1,b2);
+        if (a_in_b) return true;
+        
+        bool b_in_a = point_inside_triangle(b0,a0,a1,a2) && point_inside_triangle(b1,a0,a1,a2) && point_inside_triangle(b2,a0,a1,a2);
+        if (b_in_a) return true;
+        
+        return false;
+      }
+      
+      static bool intersect_triangle_triangle (const Vec3fa& a0, const Vec3fa& a1, const Vec3fa& a2,
+                                               const Vec3fa& b0, const Vec3fa& b1, const Vec3fa& b2)
+      {
+        const float eps = 1E-5f;
+        
+        /* calculate triangle planes */
+        const Vec3fa Na = cross(a1-a0,a2-a0);
+        const float  Ca = dot(Na,a0);
+        const Vec3fa Nb = cross(b1-b0,b2-b0);
+        const float  Cb = dot(Nb,b0);
+        
+        /* project triangle A onto plane B */
+        const float da0 = dot(Nb,a0)-Cb;
+        const float da1 = dot(Nb,a1)-Cb;
+        const float da2 = dot(Nb,a2)-Cb;
+        if (max(da0,da1,da2) < -eps) return false;
+        if (min(da0,da1,da2) > +eps) return false;
+        //CSTAT(bvh_collide_prim_intersections4++);
+        
+        /* project triangle B onto plane A */
+        const float db0 = dot(Na,b0)-Ca;
+        const float db1 = dot(Na,b1)-Ca;
+        const float db2 = dot(Na,b2)-Ca;
+        if (max(db0,db1,db2) < -eps) return false;
+        if (min(db0,db1,db2) > +eps) return false;
+        //CSTAT(bvh_collide_prim_intersections5++);
+        
+        if (unlikely((std::fabs(da0) < eps && std::fabs(da1) < eps && std::fabs(da2) < eps) ||
+                     (std::fabs(db0) < eps && std::fabs(db1) < eps && std::fabs(db2) < eps)))
+        {
+          const size_t dz = maxDim(Na);
+          const size_t dx = (dz+1)%3;
+          const size_t dy = (dx+1)%3;
+          const Vec2f A0(a0[dx],a0[dy]);
+          const Vec2f A1(a1[dx],a1[dy]);
+          const Vec2f A2(a2[dx],a2[dy]);
+          const Vec2f B0(b0[dx],b0[dy]);
+          const Vec2f B1(b1[dx],b1[dy]);
+          const Vec2f B2(b2[dx],b2[dy]);
+          return intersect_triangle_triangle(A0,A1,A2,B0,B1,B2);
+        }
+        
+        const Vec3fa D = cross(Na,Nb);
+        const float pa0 = dot(D,a0);
+        const float pa1 = dot(D,a1);
+        const float pa2 = dot(D,a2);
+        const float pb0 = dot(D,b0);
+        const float pb1 = dot(D,b1);
+        const float pb2 = dot(D,b2);
+        
+        BBox1f ba = empty;
+        if (min(da0,da1) <= 0.0f && max(da0,da1) >= 0.0f && abs(da0-da1) > 0.0f) ba.extend(T(pa0,pa1,da0,da1));
+        if (min(da1,da2) <= 0.0f && max(da1,da2) >= 0.0f && abs(da1-da2) > 0.0f) ba.extend(T(pa1,pa2,da1,da2));
+        if (min(da2,da0) <= 0.0f && max(da2,da0) >= 0.0f && abs(da2-da0) > 0.0f) ba.extend(T(pa2,pa0,da2,da0));
+        
+        BBox1f bb = empty;
+        if (min(db0,db1) <= 0.0f && max(db0,db1) >= 0.0f && abs(db0-db1) > 0.0f) bb.extend(T(pb0,pb1,db0,db1));
+        if (min(db1,db2) <= 0.0f && max(db1,db2) >= 0.0f && abs(db1-db2) > 0.0f) bb.extend(T(pb1,pb2,db1,db2));
+        if (min(db2,db0) <= 0.0f && max(db2,db0) >= 0.0f && abs(db2-db0) > 0.0f) bb.extend(T(pb2,pb0,db2,db0));
+        
+        return conjoint(ba,bb);
+      }
+    };
+  }
+}
+
+  

thirdparty/embree/kernels/geometry/trianglei.h

@@ -0,0 +1,442 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+#include "../common/scene.h"
+
+namespace embree
+{
+  /* Stores M triangles from an indexed face set */
+  template <int M>
+  struct TriangleMi
+  {
+    /* Virtual interface to query information about the triangle type */
+    struct Type : public PrimitiveType
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* primitive supports multiple time segments */
+    static const bool singleTimeSegment = false;
+
+    /* Returns maximum number of stored triangles */
+    static __forceinline size_t max_size() { return M; }
+
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+
+  public:
+
+    /* Default constructor */
+    __forceinline TriangleMi() {  }
+
+    /* Construction from vertices and IDs */
+    __forceinline TriangleMi(const vuint<M>& v0,
+                             const vuint<M>& v1,
+                             const vuint<M>& v2,
+                             const vuint<M>& geomIDs,
+                             const vuint<M>& primIDs)
+#if defined(EMBREE_COMPACT_POLYS)
+      : geomIDs(geomIDs), primIDs(primIDs) {}
+#else
+    : v0_(v0), v1_(v1), v2_(v2), geomIDs(geomIDs), primIDs(primIDs) {}
+#endif
+
+    /* Returns a mask that tells which triangles are valid */
+    __forceinline vbool<M> valid() const { return primIDs != vuint<M>(-1); }
+
+    /* Returns if the specified triangle is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return primIDs[i] != -1; }
+
+    /* Returns the number of stored triangles */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    __forceinline vuint<M> geomID() const { return geomIDs; }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline vuint<M> primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the triangles */
+    __forceinline const BBox3fa bounds(const Scene *const scene, const size_t itime=0) const
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M && valid(i); i++) {
+        const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(i));
+        bounds.extend(mesh->bounds(primID(i),itime));
+      }
+      return bounds;
+    }
+
+    /* Calculate the linear bounds of the primitive */
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, size_t itime) {
+      return LBBox3fa(bounds(scene,itime+0),bounds(scene,itime+1));
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, size_t itime, size_t numTimeSteps)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(i));
+        allBounds.extend(mesh->linearBounds(primID(i), itime, numTimeSteps));
+      }
+      return allBounds;
+    }
+
+    __forceinline LBBox3fa linearBounds(const Scene *const scene, const BBox1f time_range)
+    {
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && valid(i); i++)
+      {
+        const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(i));
+        allBounds.extend(mesh->linearBounds(primID(i), time_range));
+      }
+      return allBounds;
+    }
+    
+    /* Non-temporal store */
+    __forceinline static void store_nt(TriangleMi* dst, const TriangleMi& src)
+    {
+#if !defined(EMBREE_COMPACT_POLYS)
+      vuint<M>::store_nt(&dst->v0_,src.v0_);
+      vuint<M>::store_nt(&dst->v1_,src.v1_);
+      vuint<M>::store_nt(&dst->v2_,src.v2_);
+#endif
+      vuint<M>::store_nt(&dst->geomIDs,src.geomIDs);
+      vuint<M>::store_nt(&dst->primIDs,src.primIDs);
+    }
+
+    /* Fill triangle from triangle list */
+    template<typename PrimRefT>
+    __forceinline void fill(const PrimRefT* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      vuint<M> v0 = zero, v1 = zero, v2 = zero;
+      vuint<M> geomID = -1, primID = -1;
+      const PrimRefT* prim = &prims[begin];
+
+      for (size_t i=0; i<M; i++)
+      {
+        if (begin<end) {
+          geomID[i] = prim->geomID();
+          primID[i] = prim->primID();
+#if !defined(EMBREE_COMPACT_POLYS)
+          const TriangleMesh* mesh = scene->get<TriangleMesh>(prim->geomID());
+          const TriangleMesh::Triangle& tri = mesh->triangle(prim->primID());
+          unsigned int int_stride = mesh->vertices0.getStride()/4;
+          v0[i] = tri.v[0] * int_stride;
+          v1[i] = tri.v[1] * int_stride;
+          v2[i] = tri.v[2] * int_stride;
+#endif
+          begin++;
+        } else {
+          assert(i);
+          if (likely(i > 0)) {
+            geomID[i] = geomID[0];
+            primID[i] = -1;
+            v0[i] = v0[0];
+            v1[i] = v0[0];
+            v2[i] = v0[0];
+          }
+        }
+        if (begin<end) prim = &prims[begin];
+      }
+      new (this) TriangleMi(v0,v1,v2,geomID,primID); // FIXME: use non temporal store
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& begin, size_t end, Scene* scene, size_t itime)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, itime);
+    }
+
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& begin, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      fill(prims, begin, end, scene);
+      return linearBounds(scene, time_range);
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(TriangleMesh* mesh)
+    {
+      BBox3fa bounds = empty;
+      for (size_t i=0; i<M; i++)
+      {
+        if (primID(i) == -1) break;
+        const unsigned int primId = primID(i);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primId);
+        const Vec3fa p0 = mesh->vertex(tri.v[0]);
+        const Vec3fa p1 = mesh->vertex(tri.v[1]);
+        const Vec3fa p2 = mesh->vertex(tri.v[2]);
+        bounds.extend(merge(BBox3fa(p0),BBox3fa(p1),BBox3fa(p2)));
+      }
+      return bounds;
+    }
+
+  protected:
+#if !defined(EMBREE_COMPACT_POLYS)
+    vuint<M> v0_;         // 4 byte offset of 1st vertex
+    vuint<M> v1_;         // 4 byte offset of 2nd vertex
+    vuint<M> v2_;         // 4 byte offset of 3rd vertex
+#endif
+    vuint<M> geomIDs;    // geometry ID of mesh
+    vuint<M> primIDs;    // primitive ID of primitive inside mesh
+  };
+
+  namespace isa
+  {
+    
+  template<int M>
+    struct TriangleMi : public embree::TriangleMi<M>
+  {
+#if !defined(EMBREE_COMPACT_POLYS)
+    using embree::TriangleMi<M>::v0_;
+    using embree::TriangleMi<M>::v1_;
+    using embree::TriangleMi<M>::v2_;
+#endif
+    using embree::TriangleMi<M>::geomIDs;
+    using embree::TriangleMi<M>::primIDs;
+    using embree::TriangleMi<M>::geomID;
+    using embree::TriangleMi<M>::primID;
+    using embree::TriangleMi<M>::valid;
+        
+    /* loads a single vertex */
+    template<int vid>
+    __forceinline Vec3f getVertex(const size_t index, const Scene *const scene) const
+    {
+#if defined(EMBREE_COMPACT_POLYS)
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(index));
+      const TriangleMesh::Triangle& tri = mesh->triangle(primID(index));
+      return (Vec3f) mesh->vertices[0][tri.v[vid]];
+#else
+      const vuint<M>& v = getVertexOffset<vid>();
+      const float* vertices = scene->vertices[geomID(index)];
+      return (Vec3f&) vertices[v[index]];
+#endif
+    }
+
+    template<int vid, typename T>
+    __forceinline Vec3<T> getVertex(const size_t index, const Scene *const scene, const size_t itime, const T& ftime) const
+    {
+#if defined(EMBREE_COMPACT_POLYS)
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(index));
+      const TriangleMesh::Triangle& tri = mesh->triangle(primID(index));
+      const Vec3fa v0 = mesh->vertices[itime+0][tri.v[vid]];
+      const Vec3fa v1 = mesh->vertices[itime+1][tri.v[vid]];
+#else
+      const vuint<M>& v = getVertexOffset<vid>();
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(index));
+      const float* vertices0 = (const float*) mesh->vertexPtr(0,itime+0);
+      const float* vertices1 = (const float*) mesh->vertexPtr(0,itime+1);
+      const Vec3fa v0 = Vec3fa::loadu(vertices0+v[index]);
+      const Vec3fa v1 = Vec3fa::loadu(vertices1+v[index]);
+#endif
+      const Vec3<T> p0(v0.x,v0.y,v0.z);
+      const Vec3<T> p1(v1.x,v1.y,v1.z);
+      return lerp(p0,p1,ftime);
+    }
+
+    template<int vid, int K, typename T>
+    __forceinline Vec3<T> getVertex(const vbool<K>& valid, const size_t index, const Scene *const scene, const vint<K>& itime, const T& ftime) const
+    {
+      Vec3<T> p0, p1;
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(index));
+      
+      for (size_t mask=movemask(valid), i=bsf(mask); mask; mask=btc(mask,i), i=bsf(mask))
+      {
+#if defined(EMBREE_COMPACT_POLYS)
+        const TriangleMesh::Triangle& tri = mesh->triangle(primID(index));
+        const Vec3fa v0 = mesh->vertices[itime[i]+0][tri.v[vid]];
+        const Vec3fa v1 = mesh->vertices[itime[i]+1][tri.v[vid]];
+#else
+        const vuint<M>& v = getVertexOffset<vid>();
+        const float* vertices0 = (const float*) mesh->vertexPtr(0,itime[i]+0);
+        const float* vertices1 = (const float*) mesh->vertexPtr(0,itime[i]+1);
+        const Vec3fa v0 = Vec3fa::loadu(vertices0+v[index]);
+        const Vec3fa v1 = Vec3fa::loadu(vertices1+v[index]);
+#endif
+        p0.x[i] = v0.x; p0.y[i] = v0.y; p0.z[i] = v0.z;
+        p1.x[i] = v1.x; p1.y[i] = v1.y; p1.z[i] = v1.z;
+      }
+      return (T(one)-ftime)*p0 + ftime*p1;
+    }
+
+    struct Triangle {
+      vfloat4 v0,v1,v2;
+    };
+    
+#if defined(EMBREE_COMPACT_POLYS)
+    
+    __forceinline Triangle loadTriangle(const int i, const Scene* const scene) const 
+    {
+      const unsigned int geomID = geomIDs[i];
+      const unsigned int primID = primIDs[i];
+      if (unlikely(primID == -1)) return { zero, zero, zero };
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID);
+      const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+      const vfloat4 v0 = (vfloat4) mesh->vertices0[tri.v[0]];
+      const vfloat4 v1 = (vfloat4) mesh->vertices0[tri.v[1]];
+      const vfloat4 v2 = (vfloat4) mesh->vertices0[tri.v[2]];
+      return { v0, v1, v2 };
+    }
+
+    __forceinline Triangle loadTriangle(const int i, const int itime, const TriangleMesh* const mesh) const 
+    {
+      const unsigned int primID = primIDs[i];
+      if (unlikely(primID == -1)) return { zero, zero, zero };
+      const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+      const vfloat4 v0 = (vfloat4) mesh->vertices[itime][tri.v[0]];
+      const vfloat4 v1 = (vfloat4) mesh->vertices[itime][tri.v[1]];
+      const vfloat4 v2 = (vfloat4) mesh->vertices[itime][tri.v[2]];
+      return { v0, v1, v2 };
+    }
+    
+#else
+
+    __forceinline Triangle loadTriangle(const int i, const Scene* const scene) const 
+    {
+      const float* vertices = scene->vertices[geomID(i)];
+      const vfloat4 v0 = vfloat4::loadu(vertices + v0_[i]);
+      const vfloat4 v1 = vfloat4::loadu(vertices + v1_[i]);
+      const vfloat4 v2 = vfloat4::loadu(vertices + v2_[i]);
+      return { v0, v1, v2 };
+    }
+
+    __forceinline Triangle loadTriangle(const int i, const int itime, const TriangleMesh* const mesh) const 
+    {
+      const float* vertices = (const float*) mesh->vertexPtr(0,itime);
+      const vfloat4 v0 = vfloat4::loadu(vertices + v0_[i]);
+      const vfloat4 v1 = vfloat4::loadu(vertices + v1_[i]);
+      const vfloat4 v2 = vfloat4::loadu(vertices + v2_[i]);
+      return { v0, v1, v2 };
+    }
+    
+#endif
+
+    /* Gather the triangles */
+    __forceinline void gather(Vec3vf<M>& p0, Vec3vf<M>& p1, Vec3vf<M>& p2, const Scene* const scene) const;
+
+    template<int K>
+#if defined(__INTEL_COMPILER) && (__INTEL_COMPILER < 2000) // workaround for compiler bug in ICC 2019
+    __noinline
+#else
+    __forceinline
+#endif
+    void gather(const vbool<K>& valid,
+                Vec3vf<K>& p0,
+                Vec3vf<K>& p1,
+                Vec3vf<K>& p2,
+                const size_t index,
+                const Scene* const scene,
+                const vfloat<K>& time) const
+    {
+      const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(index));
+
+      vfloat<K> ftime;
+      const vint<K> itime = mesh->timeSegment<K>(time, ftime);
+
+      const size_t first = bsf(movemask(valid));
+      if (likely(all(valid,itime[first] == itime)))
+      {
+        p0 = getVertex<0>(index, scene, itime[first], ftime);
+        p1 = getVertex<1>(index, scene, itime[first], ftime);
+        p2 = getVertex<2>(index, scene, itime[first], ftime);
+      } else {
+        p0 = getVertex<0,K>(valid, index, scene, itime, ftime);
+        p1 = getVertex<1,K>(valid, index, scene, itime, ftime);
+        p2 = getVertex<2,K>(valid, index, scene, itime, ftime);
+      }
+    }
+
+    __forceinline void gather(Vec3vf<M>& p0,
+                              Vec3vf<M>& p1,
+                              Vec3vf<M>& p2,
+                              const TriangleMesh* mesh,
+                              const Scene *const scene,
+                              const int itime) const;
+
+    __forceinline void gather(Vec3vf<M>& p0,
+                              Vec3vf<M>& p1,
+                              Vec3vf<M>& p2,
+                              const Scene *const scene,
+                              const float time) const;
+
+
+#if !defined(EMBREE_COMPACT_POLYS)
+    template<int N> const vuint<M>& getVertexOffset() const;
+#endif
+  };
+
+#if !defined(EMBREE_COMPACT_POLYS)
+  template<> template<> __forceinline const vuint<4>& TriangleMi<4>::getVertexOffset<0>() const { return v0_; }
+  template<> template<> __forceinline const vuint<4>& TriangleMi<4>::getVertexOffset<1>() const { return v1_; }
+  template<> template<> __forceinline const vuint<4>& TriangleMi<4>::getVertexOffset<2>() const { return v2_; }
+#endif
+  
+  template<>
+  __forceinline void TriangleMi<4>::gather(Vec3vf4& p0,
+                                           Vec3vf4& p1,
+                                           Vec3vf4& p2,
+                                           const Scene* const scene) const
+  {
+    const Triangle tri0 = loadTriangle(0,scene);
+    const Triangle tri1 = loadTriangle(1,scene);
+    const Triangle tri2 = loadTriangle(2,scene);
+    const Triangle tri3 = loadTriangle(3,scene);
+    transpose(tri0.v0,tri1.v0,tri2.v0,tri3.v0,p0.x,p0.y,p0.z);
+    transpose(tri0.v1,tri1.v1,tri2.v1,tri3.v1,p1.x,p1.y,p1.z);
+    transpose(tri0.v2,tri1.v2,tri2.v2,tri3.v2,p2.x,p2.y,p2.z);
+  }
+
+  template<>
+  __forceinline void TriangleMi<4>::gather(Vec3vf4& p0,
+                                           Vec3vf4& p1,
+                                           Vec3vf4& p2,
+                                           const TriangleMesh* mesh,
+                                           const Scene *const scene,
+                                           const int itime) const
+  {
+    const Triangle tri0 = loadTriangle(0,itime,mesh);
+    const Triangle tri1 = loadTriangle(1,itime,mesh);
+    const Triangle tri2 = loadTriangle(2,itime,mesh);
+    const Triangle tri3 = loadTriangle(3,itime,mesh);
+    transpose(tri0.v0,tri1.v0,tri2.v0,tri3.v0,p0.x,p0.y,p0.z);
+    transpose(tri0.v1,tri1.v1,tri2.v1,tri3.v1,p1.x,p1.y,p1.z);
+    transpose(tri0.v2,tri1.v2,tri2.v2,tri3.v2,p2.x,p2.y,p2.z);
+  }
+
+  template<>
+  __forceinline void TriangleMi<4>::gather(Vec3vf4& p0,
+                                           Vec3vf4& p1,
+                                           Vec3vf4& p2,
+                                           const Scene *const scene,
+                                           const float time) const
+  {
+    const TriangleMesh* mesh = scene->get<TriangleMesh>(geomID(0)); // in mblur mode all geometries are identical
+
+    float ftime;
+    const int itime = mesh->timeSegment(time, ftime);
+
+    Vec3vf4 a0,a1,a2; gather(a0,a1,a2,mesh,scene,itime);
+    Vec3vf4 b0,b1,b2; gather(b0,b1,b2,mesh,scene,itime+1);
+    p0 = lerp(a0,b0,vfloat4(ftime));
+    p1 = lerp(a1,b1,vfloat4(ftime));
+    p2 = lerp(a2,b2,vfloat4(ftime));
+  }
+  }
+
+  template<int M>
+  typename TriangleMi<M>::Type TriangleMi<M>::type;
+
+  typedef TriangleMi<4> Triangle4i;
+}

thirdparty/embree/kernels/geometry/trianglei_intersector.h

@@ -0,0 +1,336 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "trianglei.h"
+#include "triangle_intersector_moeller.h"
+#include "triangle_intersector_pluecker.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMiIntersector1Moeller
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef MoellerTrumboreIntersector1<M> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        pre.intersect(ray,v0,v1,v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        return pre.intersect(ray,v0,v1,v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+    /*! Intersects M triangles with K rays */
+    template<int M, int K, bool filter>
+    struct TriangleMiIntersectorKMoeller
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef MoellerTrumboreIntersectorK<M,K> Precalculations;
+
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        const Scene* scene = context->scene;
+        for (size_t i=0; i<Primitive::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),RayHitK<K>::size());
+          const Vec3vf<K> v0 = tri.template getVertex<0>(i,scene);
+          const Vec3vf<K> v1 = tri.template getVertex<1>(i,scene);
+          const Vec3vf<K> v2 = tri.template getVertex<2>(i,scene);
+          pre.intersectK(valid_i,ray,v0,v1,v2,IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        vbool<K> valid0 = valid_i;
+        const Scene* scene = context->scene;
+
+        for (size_t i=0; i<Primitive::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid_i),RayHitK<K>::size());
+          const Vec3vf<K> v0 = tri.template getVertex<0>(i,scene);
+          const Vec3vf<K> v1 = tri.template getVertex<1>(i,scene);
+          const Vec3vf<K> v2 = tri.template getVertex<2>(i,scene);
+          pre.intersectK(valid0,ray,v0,v1,v2,OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        pre.intersect(ray,k,v0,v1,v2,Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        return pre.intersect(ray,k,v0,v1,v2,Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+
+    /*! Intersects M triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMiIntersector1Pluecker
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef PlueckerIntersector1<M> Precalculations;
+
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        pre.intersect(ray,v0,v1,v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        return pre.intersect(ray,v0,v1,v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+    /*! Intersects M triangles with K rays */
+    template<int M, int K, bool filter>
+    struct TriangleMiIntersectorKPluecker
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef PlueckerIntersectorK<M,K> Precalculations;
+
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        const Scene* scene = context->scene;
+        for (size_t i=0; i<Primitive::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),RayHitK<K>::size());
+          const Vec3vf<K> v0 = tri.template getVertex<0>(i,scene);
+          const Vec3vf<K> v1 = tri.template getVertex<1>(i,scene);
+          const Vec3vf<K> v2 = tri.template getVertex<2>(i,scene);
+          pre.intersectK(valid_i,ray,v0,v1,v2,IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        vbool<K> valid0 = valid_i;
+        const Scene* scene = context->scene;
+
+        for (size_t i=0; i<Primitive::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid_i),RayHitK<K>::size());
+          const Vec3vf<K> v0 = tri.template getVertex<0>(i,scene);
+          const Vec3vf<K> v1 = tri.template getVertex<1>(i,scene);
+          const Vec3vf<K> v2 = tri.template getVertex<2>(i,scene);
+          pre.intersectK(valid0,ray,v0,v1,v2,OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        pre.intersect(ray,k,v0,v1,v2,Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0, v1, v2; tri.gather(v0,v1,v2,context->scene);
+        return pre.intersect(ray,k,v0,v1,v2,Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+
+    /*! Intersects M motion blur triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMiMBIntersector1Moeller
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef MoellerTrumboreIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with the M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time());
+        pre.intersect(ray,v0,v1,v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time());
+        return pre.intersect(ray,v0,v1,v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+    /*! Intersects M motion blur triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct TriangleMiMBIntersectorKMoeller
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef MoellerTrumboreIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        for (size_t i=0; i<TriangleMi<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          Vec3vf<K> v0,v1,v2; tri.template gather<K>(valid_i,v0,v1,v2,i,context->scene,ray.time());
+          pre.intersectK(valid_i,ray,v0,v1,v2,IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<TriangleMi<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          Vec3vf<K> v0,v1,v2; tri.template gather<K>(valid_i,v0,v1,v2,i,context->scene,ray.time());
+          pre.intersectK(valid0,ray,v0,v1,v2,OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time()[k]);
+        pre.intersect(ray,k,v0,v1,v2,Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time()[k]);
+        return pre.intersect(ray,k,v0,v1,v2,Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+
+    /*! Intersects M motion blur triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMiMBIntersector1Pluecker
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef PlueckerIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with the M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time());
+        pre.intersect(ray,v0,v1,v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time());
+        return pre.intersect(ray,v0,v1,v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+    /*! Intersects M motion blur triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct TriangleMiMBIntersectorKPluecker
+    {
+      typedef TriangleMi<M> Primitive;
+      typedef PlueckerIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        for (size_t i=0; i<TriangleMi<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          Vec3vf<K> v0,v1,v2; tri.template gather<K>(valid_i,v0,v1,v2,i,context->scene,ray.time());
+          pre.intersectK(valid_i,ray,v0,v1,v2,IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        vbool<K> valid0 = valid_i;
+        for (size_t i=0; i<TriangleMi<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          Vec3vf<K> v0,v1,v2; tri.template gather<K>(valid_i,v0,v1,v2,i,context->scene,ray.time());
+          pre.intersectK(valid0,ray,v0,v1,v2,OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time()[k]);
+        pre.intersect(ray,k,v0,v1,v2,Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const TriangleMi<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        Vec3vf<M> v0,v1,v2; tri.gather(v0,v1,v2,context->scene,ray.time()[k]);
+        return pre.intersect(ray,k,v0,v1,v2,Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/trianglev.h

@@ -0,0 +1,157 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  /* Stores the vertices of M triangles in struct of array layout */
+  template <int M>
+  struct TriangleMv
+  { 
+  public:
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+    static Type type;
+
+  public:
+
+    /* Returns maximum number of stored triangles */
+    static __forceinline size_t max_size() { return M; }
+    
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+   
+  public:
+
+    /* Default constructor */
+    __forceinline TriangleMv() {}
+
+    /* Construction from vertices and IDs */
+    __forceinline TriangleMv(const Vec3vf<M>& v0, const Vec3vf<M>& v1, const Vec3vf<M>& v2, const vuint<M>& geomIDs, const vuint<M>& primIDs)
+      : v0(v0), v1(v1), v2(v2), geomIDs(geomIDs), primIDs(primIDs) {}
+    
+    /* Returns a mask that tells which triangles are valid */
+    __forceinline vbool<M> valid() const { return geomIDs != vuint<M>(-1); }
+
+    /* Returns true if the specified triangle is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return geomIDs[i] != -1; }
+
+    /* Returns the number of stored triangles */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    __forceinline       vuint<M>& geomID()       { return geomIDs; }
+    __forceinline const vuint<M>& geomID() const { return geomIDs; }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M>& primID()       { return primIDs; }
+    __forceinline const vuint<M>& primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the triangles */
+    __forceinline BBox3fa bounds() const 
+    {
+      Vec3vf<M> lower = min(v0,v1,v2);
+      Vec3vf<M> upper = max(v0,v1,v2);
+      vbool<M> mask = valid();
+      lower.x = select(mask,lower.x,vfloat<M>(pos_inf));
+      lower.y = select(mask,lower.y,vfloat<M>(pos_inf));
+      lower.z = select(mask,lower.z,vfloat<M>(pos_inf));
+      upper.x = select(mask,upper.x,vfloat<M>(neg_inf));
+      upper.y = select(mask,upper.y,vfloat<M>(neg_inf));
+      upper.z = select(mask,upper.z,vfloat<M>(neg_inf));
+      return BBox3fa(Vec3fa(reduce_min(lower.x),reduce_min(lower.y),reduce_min(lower.z)),
+                     Vec3fa(reduce_max(upper.x),reduce_max(upper.y),reduce_max(upper.z)));
+    }
+    
+    /* Non temporal store */
+    __forceinline static void store_nt(TriangleMv* dst, const TriangleMv& src)
+    {
+      vfloat<M>::store_nt(&dst->v0.x,src.v0.x);
+      vfloat<M>::store_nt(&dst->v0.y,src.v0.y);
+      vfloat<M>::store_nt(&dst->v0.z,src.v0.z);
+      vfloat<M>::store_nt(&dst->v1.x,src.v1.x);
+      vfloat<M>::store_nt(&dst->v1.y,src.v1.y);
+      vfloat<M>::store_nt(&dst->v1.z,src.v1.z);
+      vfloat<M>::store_nt(&dst->v2.x,src.v2.x);
+      vfloat<M>::store_nt(&dst->v2.y,src.v2.y);
+      vfloat<M>::store_nt(&dst->v2.z,src.v2.z);
+      vuint<M>::store_nt(&dst->geomIDs,src.geomIDs);
+      vuint<M>::store_nt(&dst->primIDs,src.primIDs);
+    }
+
+    /* Fill triangle from triangle list */
+    __forceinline void fill(const PrimRef* prims, size_t& begin, size_t end, Scene* scene)
+    {
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero;
+      
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+	const PrimRef& prim = prims[begin];
+        const unsigned geomID = prim.geomID();
+        const unsigned primID = prim.primID();
+        const TriangleMesh* __restrict__ const mesh = scene->get<TriangleMesh>(geomID);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+        const Vec3fa& p0 = mesh->vertex(tri.v[0]);
+        const Vec3fa& p1 = mesh->vertex(tri.v[1]);
+        const Vec3fa& p2 = mesh->vertex(tri.v[2]);
+        vgeomID [i] = geomID;
+        vprimID [i] = primID;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+      }
+      TriangleMv::store_nt(this,TriangleMv(v0,v1,v2,vgeomID,vprimID));
+    }
+
+    /* Updates the primitive */
+    __forceinline BBox3fa update(TriangleMesh* mesh)
+    {
+      BBox3fa bounds = empty;
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> v0 = zero, v1 = zero, v2 = zero;
+      
+      for (size_t i=0; i<M; i++)
+      {
+        if (primID(i) == -1) break;
+        const unsigned geomId = geomID(i);
+        const unsigned primId = primID(i);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primId);
+        const Vec3fa p0 = mesh->vertex(tri.v[0]);
+        const Vec3fa p1 = mesh->vertex(tri.v[1]);
+        const Vec3fa p2 = mesh->vertex(tri.v[2]);
+        bounds.extend(merge(BBox3fa(p0),BBox3fa(p1),BBox3fa(p2)));
+        vgeomID [i] = geomId;
+        vprimID [i] = primId;
+        v0.x[i] = p0.x; v0.y[i] = p0.y; v0.z[i] = p0.z;
+        v1.x[i] = p1.x; v1.y[i] = p1.y; v1.z[i] = p1.z;
+        v2.x[i] = p2.x; v2.y[i] = p2.y; v2.z[i] = p2.z;
+      }
+      new (this) TriangleMv(v0,v1,v2,vgeomID,vprimID);
+      return bounds;
+    }
+   
+  public:
+    Vec3vf<M> v0;      // 1st vertex of the triangles
+    Vec3vf<M> v1;      // 2nd vertex of the triangles
+    Vec3vf<M> v2;      // 3rd vertex of the triangles
+  private:
+    vuint<M> geomIDs; // geometry ID
+    vuint<M> primIDs; // primitive ID
+  };
+
+  template<int M>
+  typename TriangleMv<M>::Type TriangleMv<M>::type;
+
+  typedef TriangleMv<4> Triangle4v;
+}

thirdparty/embree/kernels/geometry/trianglev_intersector.h

@@ -0,0 +1,206 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "triangle_intersector_pluecker.h"
+#include "triangle_intersector_moeller.h"
+#include "triangle_intersector_woop.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMvIntersector1Moeller
+    {
+      typedef TriangleMv<M> Primitive;
+      typedef MoellerTrumboreIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,tri.v0,tri.v1,tri.v2,/*UVIdentity<M>(),*/Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersect(ray,tri.v0,tri.v1,tri.v2,/*UVIdentity<M>(),*/Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+
+    template<int M, bool filter>
+    struct TriangleMvIntersector1Woop
+    {
+      typedef TriangleMv<M> Primitive;
+      typedef WoopIntersector1<M> intersec;
+      typedef WoopPrecalculations1<M> Precalculations;
+
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        intersec::intersect(ray,pre,tri.v0,tri.v1,tri.v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return intersec::intersect(ray,pre,tri.v0,tri.v1,tri.v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+
+    /*! Intersects M triangles with K rays */
+    template<int M, int K, bool filter>
+    struct TriangleMvIntersectorKMoeller
+    {
+      typedef TriangleMv<M> Primitive;
+      typedef MoellerTrumboreIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        for (size_t i=0; i<M; i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> v0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> v1 = broadcast<vfloat<K>>(tri.v1,i);
+          const Vec3vf<K> v2 = broadcast<vfloat<K>>(tri.v2,i);
+          pre.intersectK(valid_i,ray,v0,v1,v2,/*UVIdentity<K>(),*/IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<M; i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> v0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> v1 = broadcast<vfloat<K>>(tri.v1,i);
+          const Vec3vf<K> v2 = broadcast<vfloat<K>>(tri.v2,i);
+          pre.intersectK(valid0,ray,v0,v1,v2,/*UVIdentity<K>(),*/OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,k,tri.v0,tri.v1,tri.v2,/*UVIdentity<M>(),*/Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID())); //FIXME: M
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersect(ray,k,tri.v0,tri.v1,tri.v2,/*UVIdentity<M>(),*/Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID())); //FIXME: M
+      }
+    };
+
+    /*! Intersects M triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMvIntersector1Pluecker
+    {
+      typedef TriangleMv<M> Primitive;
+      typedef PlueckerIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHit& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,tri.v0,tri.v1,tri.v2,UVIdentity<M>(),Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersect(ray,tri.v0,tri.v1,tri.v2,UVIdentity<M>(),Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+
+    /*! Intersects M triangles with K rays */
+    template<int M, int K, bool filter>
+    struct TriangleMvIntersectorKPluecker
+    {
+      typedef TriangleMv<M> Primitive;
+      typedef PlueckerIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        for (size_t i=0; i<M; i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> v0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> v1 = broadcast<vfloat<K>>(tri.v1,i);
+          const Vec3vf<K> v2 = broadcast<vfloat<K>>(tri.v2,i);
+          pre.intersectK(valid_i,ray,v0,v1,v2,UVIdentity<K>(),IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const Primitive& tri)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<M; i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> v0 = broadcast<vfloat<K>>(tri.v0,i);
+          const Vec3vf<K> v1 = broadcast<vfloat<K>>(tri.v1,i);
+          const Vec3vf<K> v2 = broadcast<vfloat<K>>(tri.v2,i);
+          pre.intersectK(valid0,ray,v0,v1,v2,UVIdentity<K>(),OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        pre.intersect(ray,k,tri.v0,tri.v1,tri.v2,UVIdentity<M>(),Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const Primitive& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        return pre.intersect(ray,k,tri.v0,tri.v1,tri.v2,UVIdentity<M>(),Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+  }
+}

thirdparty/embree/kernels/geometry/trianglev_mb.h

@@ -0,0 +1,200 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "primitive.h"
+
+namespace embree
+{
+  /* Stores the vertices of M triangles in struct of array layout */
+  template<int M>
+  struct TriangleMvMB
+  {
+  public:
+    struct Type : public PrimitiveType 
+    {
+      const char* name() const;
+      size_t sizeActive(const char* This) const;
+      size_t sizeTotal(const char* This) const;
+      size_t getBytes(const char* This) const;
+    };
+
+    static Type type;
+
+  public:
+
+    /* primitive supports single time segments */
+    static const bool singleTimeSegment = true;
+
+    /* Returns maximum number of stored triangles */
+    static __forceinline size_t max_size() { return M; }
+    
+    /* Returns required number of primitive blocks for N primitives */
+    static __forceinline size_t blocks(size_t N) { return (N+max_size()-1)/max_size(); }
+   
+  public:
+
+    /* Default constructor */
+    __forceinline TriangleMvMB() {}
+
+    /* Construction from vertices and IDs */
+    __forceinline TriangleMvMB(const Vec3vf<M>& a0, const Vec3vf<M>& a1,
+                               const Vec3vf<M>& b0, const Vec3vf<M>& b1,
+                               const Vec3vf<M>& c0, const Vec3vf<M>& c1,
+                               const vuint<M>& geomIDs, const vuint<M>& primIDs)
+      : v0(a0), v1(b0), v2(c0), dv0(a1-a0), dv1(b1-b0), dv2(c1-c0), geomIDs(geomIDs), primIDs(primIDs) {}
+
+    /* Returns a mask that tells which triangles are valid */
+    __forceinline vbool<M> valid() const { return geomIDs != vuint<M>(-1); }
+
+    /* Returns if the specified triangle is valid */
+    __forceinline bool valid(const size_t i) const { assert(i<M); return geomIDs[i] != -1; }
+
+    /* Returns the number of stored triangles */
+    __forceinline size_t size() const { return bsf(~movemask(valid())); }
+
+    /* Returns the geometry IDs */
+    __forceinline       vuint<M>& geomID()       { return geomIDs; }
+    __forceinline const vuint<M>& geomID() const { return geomIDs; }
+    __forceinline unsigned int geomID(const size_t i) const { assert(i<M); return geomIDs[i]; }
+
+    /* Returns the primitive IDs */
+    __forceinline       vuint<M>& primID()       { return primIDs; }
+    __forceinline const vuint<M>& primID() const { return primIDs; }
+    __forceinline unsigned int primID(const size_t i) const { assert(i<M); return primIDs[i]; }
+
+    /* Calculate the bounds of the triangles at t0 */
+    __forceinline BBox3fa bounds0() const 
+    {
+      Vec3vf<M> lower = min(v0,v1,v2);
+      Vec3vf<M> upper = max(v0,v1,v2);
+      const vbool<M> mask = valid();
+      lower.x = select(mask,lower.x,vfloat<M>(pos_inf));
+      lower.y = select(mask,lower.y,vfloat<M>(pos_inf));
+      lower.z = select(mask,lower.z,vfloat<M>(pos_inf));
+      upper.x = select(mask,upper.x,vfloat<M>(neg_inf));
+      upper.y = select(mask,upper.y,vfloat<M>(neg_inf));
+      upper.z = select(mask,upper.z,vfloat<M>(neg_inf));
+      return BBox3fa(Vec3fa(reduce_min(lower.x),reduce_min(lower.y),reduce_min(lower.z)),
+		     Vec3fa(reduce_max(upper.x),reduce_max(upper.y),reduce_max(upper.z)));
+    }
+
+    /* Calculate the bounds of the triangles at t1 */
+    __forceinline BBox3fa bounds1() const 
+    {
+      const Vec3vf<M> p0 = v0+dv0;
+      const Vec3vf<M> p1 = v1+dv1;
+      const Vec3vf<M> p2 = v2+dv2;
+      Vec3vf<M> lower = min(p0,p1,p2);
+      Vec3vf<M> upper = max(p0,p1,p2);
+      const vbool<M> mask = valid();
+      lower.x = select(mask,lower.x,vfloat<M>(pos_inf));
+      lower.y = select(mask,lower.y,vfloat<M>(pos_inf));
+      lower.z = select(mask,lower.z,vfloat<M>(pos_inf));
+      upper.x = select(mask,upper.x,vfloat<M>(neg_inf));
+      upper.y = select(mask,upper.y,vfloat<M>(neg_inf));
+      upper.z = select(mask,upper.z,vfloat<M>(neg_inf));
+      return BBox3fa(Vec3fa(reduce_min(lower.x),reduce_min(lower.y),reduce_min(lower.z)),
+		     Vec3fa(reduce_max(upper.x),reduce_max(upper.y),reduce_max(upper.z)));
+    }
+
+    /* Calculate the linear bounds of the primitive */
+    __forceinline LBBox3fa linearBounds() const {
+      return LBBox3fa(bounds0(),bounds1());
+    }
+
+    /* Fill triangle from triangle list */
+    __forceinline LBBox3fa fillMB(const PrimRef* prims, size_t& begin, size_t end, Scene* scene, size_t itime)
+    {
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> va0 = zero, vb0 = zero, vc0 = zero;
+      Vec3vf<M> va1 = zero, vb1 = zero, vc1 = zero;
+
+      BBox3fa bounds0 = empty;
+      BBox3fa bounds1 = empty;
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+	const PrimRef& prim = prims[begin];
+        const unsigned geomID = prim.geomID();
+        const unsigned primID = prim.primID();
+        const TriangleMesh* __restrict__ const mesh = scene->get<TriangleMesh>(geomID);
+        const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+        const Vec3fa& a0 = mesh->vertex(tri.v[0],size_t(itime+0)); bounds0.extend(a0);
+        const Vec3fa& a1 = mesh->vertex(tri.v[0],size_t(itime+1)); bounds1.extend(a1);
+        const Vec3fa& b0 = mesh->vertex(tri.v[1],size_t(itime+0)); bounds0.extend(b0);
+        const Vec3fa& b1 = mesh->vertex(tri.v[1],size_t(itime+1)); bounds1.extend(b1);
+        const Vec3fa& c0 = mesh->vertex(tri.v[2],size_t(itime+0)); bounds0.extend(c0);
+        const Vec3fa& c1 = mesh->vertex(tri.v[2],size_t(itime+1)); bounds1.extend(c1);
+        vgeomID [i] = geomID;
+        vprimID [i] = primID;
+        va0.x[i] = a0.x; va0.y[i] = a0.y; va0.z[i] = a0.z;
+	va1.x[i] = a1.x; va1.y[i] = a1.y; va1.z[i] = a1.z;
+	vb0.x[i] = b0.x; vb0.y[i] = b0.y; vb0.z[i] = b0.z;
+	vb1.x[i] = b1.x; vb1.y[i] = b1.y; vb1.z[i] = b1.z;
+	vc0.x[i] = c0.x; vc0.y[i] = c0.y; vc0.z[i] = c0.z;
+	vc1.x[i] = c1.x; vc1.y[i] = c1.y; vc1.z[i] = c1.z;
+      }
+      new (this) TriangleMvMB(va0,va1,vb0,vb1,vc0,vc1,vgeomID,vprimID);
+      return LBBox3fa(bounds0,bounds1);
+    }
+
+    /* Fill triangle from triangle list */
+    __forceinline LBBox3fa fillMB(const PrimRefMB* prims, size_t& begin, size_t end, Scene* scene, const BBox1f time_range)
+    {
+      vuint<M> vgeomID = -1, vprimID = -1;
+      Vec3vf<M> va0 = zero, vb0 = zero, vc0 = zero;
+      Vec3vf<M> va1 = zero, vb1 = zero, vc1 = zero;
+
+      LBBox3fa allBounds = empty;
+      for (size_t i=0; i<M && begin<end; i++, begin++)
+      {
+        const PrimRefMB& prim = prims[begin];
+        const unsigned geomID = prim.geomID();
+        const unsigned primID = prim.primID();
+        const TriangleMesh* const mesh = scene->get<TriangleMesh>(geomID);
+        const range<int> itime_range = mesh->timeSegmentRange(time_range);
+        assert(itime_range.size() == 1);
+        const int ilower = itime_range.begin();
+        const TriangleMesh::Triangle& tri = mesh->triangle(primID);
+        allBounds.extend(mesh->linearBounds(primID, time_range));
+        const Vec3fa& a0 = mesh->vertex(tri.v[0],size_t(ilower+0));
+        const Vec3fa& a1 = mesh->vertex(tri.v[0],size_t(ilower+1));
+        const Vec3fa& b0 = mesh->vertex(tri.v[1],size_t(ilower+0));
+        const Vec3fa& b1 = mesh->vertex(tri.v[1],size_t(ilower+1));
+        const Vec3fa& c0 = mesh->vertex(tri.v[2],size_t(ilower+0));
+        const Vec3fa& c1 = mesh->vertex(tri.v[2],size_t(ilower+1));
+        const BBox1f time_range_v(mesh->timeStep(ilower+0),mesh->timeStep(ilower+1));
+        auto a01 = globalLinear(std::make_pair(a0,a1),time_range_v);
+        auto b01 = globalLinear(std::make_pair(b0,b1),time_range_v);
+        auto c01 = globalLinear(std::make_pair(c0,c1),time_range_v);
+        vgeomID [i] = geomID;
+        vprimID [i] = primID;
+        va0.x[i] = a01.first .x; va0.y[i] = a01.first .y; va0.z[i] = a01.first .z;
+	va1.x[i] = a01.second.x; va1.y[i] = a01.second.y; va1.z[i] = a01.second.z;
+	vb0.x[i] = b01.first .x; vb0.y[i] = b01.first .y; vb0.z[i] = b01.first .z;
+	vb1.x[i] = b01.second.x; vb1.y[i] = b01.second.y; vb1.z[i] = b01.second.z;
+	vc0.x[i] = c01.first .x; vc0.y[i] = c01.first .y; vc0.z[i] = c01.first .z;
+	vc1.x[i] = c01.second.x; vc1.y[i] = c01.second.y; vc1.z[i] = c01.second.z;
+      }
+      new (this) TriangleMvMB(va0,va1,vb0,vb1,vc0,vc1,vgeomID,vprimID);
+      return allBounds;
+    }
+
+  public:
+    Vec3vf<M> v0;      // 1st vertex of the triangles
+    Vec3vf<M> v1;      // 2nd vertex of the triangles
+    Vec3vf<M> v2;      // 3rd vertex of the triangles
+    Vec3vf<M> dv0;     // difference vector between time steps t0 and t1 for first vertex
+    Vec3vf<M> dv1;     // difference vector between time steps t0 and t1 for second vertex
+    Vec3vf<M> dv2;     // difference vector between time steps t0 and t1 for third vertex
+  private:
+    vuint<M> geomIDs; // geometry ID
+    vuint<M> primIDs; // primitive ID
+  };
+
+  template<int M>
+  typename TriangleMvMB<M>::Type TriangleMvMB<M>::type;
+
+  typedef TriangleMvMB<4> Triangle4vMB;
+}

thirdparty/embree/kernels/geometry/trianglev_mb_intersector.h

@@ -0,0 +1,211 @@
+// Copyright 2009-2021 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include "triangle.h"
+#include "intersector_epilog.h"
+
+namespace embree
+{
+  namespace isa
+  {
+    /*! Intersects M motion blur triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMvMBIntersector1Moeller
+    {
+      typedef TriangleMvMB<M> Primitive;
+      typedef MoellerTrumboreIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with the M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time());
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        pre.intersect(ray,v0,v1,v2,Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time());
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        return pre.intersect(ray,v0,v1,v2,Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+    
+    /*! Intersects M motion blur triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct TriangleMvMBIntersectorKMoeller
+    {
+      typedef TriangleMvMB<M> Primitive;
+      typedef MoellerTrumboreIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        for (size_t i=0; i<TriangleMvMB<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> time(ray.time());
+          const Vec3vf<K> v0 = madd(time,broadcast<vfloat<K>>(tri.dv0,i),broadcast<vfloat<K>>(tri.v0,i));
+          const Vec3vf<K> v1 = madd(time,broadcast<vfloat<K>>(tri.dv1,i),broadcast<vfloat<K>>(tri.v1,i));
+          const Vec3vf<K> v2 = madd(time,broadcast<vfloat<K>>(tri.dv2,i),broadcast<vfloat<K>>(tri.v2,i));
+          pre.intersectK(valid_i,ray,v0,v1,v2,IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<TriangleMvMB<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> time(ray.time());
+          const Vec3vf<K> v0 = madd(time,broadcast<vfloat<K>>(tri.dv0,i),broadcast<vfloat<K>>(tri.v0,i));
+          const Vec3vf<K> v1 = madd(time,broadcast<vfloat<K>>(tri.dv1,i),broadcast<vfloat<K>>(tri.v1,i));
+          const Vec3vf<K> v2 = madd(time,broadcast<vfloat<K>>(tri.dv2,i),broadcast<vfloat<K>>(tri.v2,i));
+          pre.intersectK(valid0,ray,v0,v1,v2,OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+      
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time()[k]);
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        pre.intersect(ray,k,v0,v1,v2,Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time()[k]);
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        return pre.intersect(ray,k,v0,v1,v2,Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+
+    /*! Intersects M motion blur triangles with 1 ray */
+    template<int M, bool filter>
+    struct TriangleMvMBIntersector1Pluecker
+    {
+      typedef TriangleMvMB<M> Primitive;
+      typedef PlueckerIntersector1<M> Precalculations;
+
+      /*! Intersect a ray with the M triangles and updates the hit. */
+      static __forceinline void intersect(const Precalculations& pre, RayHit& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time());
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        pre.intersect(ray,v0,v1,v2,UVIdentity<M>(),Intersect1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of M triangles. */
+      static __forceinline bool occluded(const Precalculations& pre, Ray& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time());
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        return pre.intersect(ray,v0,v1,v2,UVIdentity<M>(),Occluded1EpilogM<M,filter>(ray,context,tri.geomID(),tri.primID()));
+      }
+      
+      static __forceinline bool pointQuery(PointQuery* query, PointQueryContext* context, const Primitive& tri)
+      {
+        return PrimitivePointQuery1<Primitive>::pointQuery(query, context, tri);
+      }
+    };
+    
+    /*! Intersects M motion blur triangles with K rays. */
+    template<int M, int K, bool filter>
+    struct TriangleMvMBIntersectorKPluecker
+    {
+      typedef TriangleMvMB<M> Primitive;
+      typedef PlueckerIntersectorK<M,K> Precalculations;
+
+      /*! Intersects K rays with M triangles. */
+      static __forceinline void intersect(const vbool<K>& valid_i, Precalculations& pre, RayHitK<K>& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        for (size_t i=0; i<TriangleMvMB<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(normal.trav_prims,1,popcnt(valid_i),K);
+          const Vec3vf<K> time(ray.time());
+          const Vec3vf<K> v0 = madd(time,broadcast<vfloat<K>>(tri.dv0,i),broadcast<vfloat<K>>(tri.v0,i));
+          const Vec3vf<K> v1 = madd(time,broadcast<vfloat<K>>(tri.dv1,i),broadcast<vfloat<K>>(tri.v1,i));
+          const Vec3vf<K> v2 = madd(time,broadcast<vfloat<K>>(tri.dv2,i),broadcast<vfloat<K>>(tri.v2,i));
+          pre.intersectK(valid_i,ray,v0,v1,v2,UVIdentity<K>(),IntersectKEpilogM<M,K,filter>(ray,context,tri.geomID(),tri.primID(),i));
+        }
+      }
+
+      /*! Test for K rays if they are occluded by any of the M triangles. */
+      static __forceinline vbool<K> occluded(const vbool<K>& valid_i, Precalculations& pre, RayK<K>& ray, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        vbool<K> valid0 = valid_i;
+
+        for (size_t i=0; i<TriangleMvMB<M>::max_size(); i++)
+        {
+          if (!tri.valid(i)) break;
+          STAT3(shadow.trav_prims,1,popcnt(valid0),K);
+          const Vec3vf<K> time(ray.time());
+          const Vec3vf<K> v0 = madd(time,broadcast<vfloat<K>>(tri.dv0,i),broadcast<vfloat<K>>(tri.v0,i));
+          const Vec3vf<K> v1 = madd(time,broadcast<vfloat<K>>(tri.dv1,i),broadcast<vfloat<K>>(tri.v1,i));
+          const Vec3vf<K> v2 = madd(time,broadcast<vfloat<K>>(tri.dv2,i),broadcast<vfloat<K>>(tri.v2,i));
+          pre.intersectK(valid0,ray,v0,v1,v2,UVIdentity<K>(),OccludedKEpilogM<M,K,filter>(valid0,ray,context,tri.geomID(),tri.primID(),i));
+          if (none(valid0)) break;
+        }
+        return !valid0;
+      }
+
+      /*! Intersect a ray with M triangles and updates the hit. */
+      static __forceinline void intersect(Precalculations& pre, RayHitK<K>& ray, size_t k, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(normal.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time()[k]);
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        pre.intersect(ray,k,v0,v1,v2,UVIdentity<M>(),Intersect1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+
+      /*! Test if the ray is occluded by one of the M triangles. */
+      static __forceinline bool occluded(Precalculations& pre, RayK<K>& ray, size_t k, RayQueryContext* context, const TriangleMvMB<M>& tri)
+      {
+        STAT3(shadow.trav_prims,1,1,1);
+        const Vec3vf<M> time(ray.time()[k]);
+        const Vec3vf<M> v0 = madd(time,Vec3vf<M>(tri.dv0),Vec3vf<M>(tri.v0));
+        const Vec3vf<M> v1 = madd(time,Vec3vf<M>(tri.dv1),Vec3vf<M>(tri.v1));
+        const Vec3vf<M> v2 = madd(time,Vec3vf<M>(tri.dv2),Vec3vf<M>(tri.v2));
+        return pre.intersect(ray,k,v0,v1,v2,UVIdentity<M>(),Occluded1KEpilogM<M,K,filter>(ray,k,context,tri.geomID(),tri.primID()));
+      }
+    };
+  }
+}