| 1 | // Copyright 2009-2021 Intel Corporation | 
| 2 | // SPDX-License-Identifier: Apache-2.0 | 
| 3 |  | 
| 4 | #pragma once | 
| 5 |  | 
| 6 | #include "../common/ray.h" | 
| 7 | #include "quad_intersector.h" | 
| 8 | #include "curve_intersector_precalculations.h" | 
| 9 |  | 
| 10 | #define Bezier1Intersector1 RibbonCurve1Intersector1 | 
| 11 | #define Bezier1IntersectorK RibbonCurve1IntersectorK | 
| 12 |  | 
| 13 | namespace embree | 
| 14 | { | 
| 15 |   namespace isa | 
| 16 |   { | 
| 17 |     template<typename NativeCurve3ff, int M> | 
| 18 |     struct RibbonHit | 
| 19 |     { | 
| 20 |       __forceinline RibbonHit() {} | 
| 21 |  | 
| 22 |       __forceinline RibbonHit(const vbool<M>& valid, const vfloat<M>& U, const vfloat<M>& V, const vfloat<M>& T, const int i, const int N, | 
| 23 |                               const NativeCurve3ff& curve3D) | 
| 24 |         : U(U), V(V), T(T), i(i), N(N), curve3D(curve3D), valid(valid) {} | 
| 25 |        | 
| 26 |       __forceinline void finalize()  | 
| 27 |       { | 
| 28 |         vu = (vfloat<M>(step)+U+vfloat<M>(float(i)))*(1.0f/float(N)); | 
| 29 |         vv = V; | 
| 30 |         vt = T; | 
| 31 |       } | 
| 32 |        | 
| 33 |       __forceinline Vec2f uv (const size_t i) const { return Vec2f(vu[i],vv[i]); } | 
| 34 |       __forceinline float t  (const size_t i) const { return vt[i]; } | 
| 35 |       __forceinline Vec3fa Ng(const size_t i) const { return curve3D.eval_du(vu[i]); } | 
| 36 |  | 
| 37 |       __forceinline Vec2vf<M> uv() const { return Vec2vf<M>(vu,vv); } | 
| 38 |       __forceinline vfloat<M> t () const { return vt; } | 
| 39 |       __forceinline Vec3vf<M> Ng() const { return (Vec3vf<M>) curve3D.template veval_du<M>(vu); } | 
| 40 |        | 
| 41 |     public: | 
| 42 |       vfloat<M> U; | 
| 43 |       vfloat<M> V; | 
| 44 |       vfloat<M> T; | 
| 45 |       int i, N; | 
| 46 |       NativeCurve3ff curve3D; | 
| 47 |        | 
| 48 |     public: | 
| 49 |       vbool<M> valid; | 
| 50 |       vfloat<M> vu; | 
| 51 |       vfloat<M> vv; | 
| 52 |       vfloat<M> vt; | 
| 53 |     }; | 
| 54 |  | 
| 55 |     /* calculate squared distance of point p0 to line p1->p2 */ | 
| 56 |     __forceinline std::pair<vfloatx,vfloatx> sqr_point_line_distance(const Vec2vfx& p0, const Vec2vfx& p1, const Vec2vfx& p2) | 
| 57 |     { | 
| 58 |       const vfloatx num = det(a: p2-p1,b: p1-p0); | 
| 59 |       const vfloatx den2 = dot(a: p2-p1,b: p2-p1); | 
| 60 |       return std::make_pair(x: num*num,y: den2); | 
| 61 |     } | 
| 62 |      | 
| 63 |     /* performs culling against a cylinder */ | 
| 64 |     __forceinline vboolx cylinder_culling_test(const Vec2vfx& p0, const Vec2vfx& p1, const Vec2vfx& p2, const vfloatx& r) | 
| 65 |     { | 
| 66 |       const std::pair<vfloatx,vfloatx> d = sqr_point_line_distance(p0,p1,p2); | 
| 67 |       return d.first <= r*r*d.second; | 
| 68 |     } | 
| 69 |  | 
| 70 |     template<typename NativeCurve3ff, typename Epilog> | 
| 71 |     __forceinline bool intersect_ribbon(const Vec3fa& ray_org, const Vec3fa& ray_dir, const float ray_tnear, const float& ray_tfar, | 
| 72 |                                         const LinearSpace3fa& ray_space, const float& depth_scale, | 
| 73 |                                         const NativeCurve3ff& curve3D, const int N, | 
| 74 |                                         const Epilog& epilog) | 
| 75 |     { | 
| 76 |       /* transform control points into ray space */ | 
| 77 |       const NativeCurve3ff curve2D = curve3D.xfm_pr(ray_space,ray_org); | 
| 78 |       float eps = 4.0f*float(ulp)*reduce_max(max(abs(curve2D.v0),abs(curve2D.v1),abs(curve2D.v2),abs(curve2D.v3))); | 
| 79 |        | 
| 80 |       /* evaluate the bezier curve */ | 
| 81 |       bool ishit = false; | 
| 82 |       vboolx valid = vfloatx(step) < vfloatx(float(N)); | 
| 83 |       const Vec4vfx p0 = curve2D.template eval0<VSIZEX>(0,N); | 
| 84 |       const Vec4vfx p1 = curve2D.template eval1<VSIZEX>(0,N); | 
| 85 |       valid &= cylinder_culling_test(p0: zero,p1: Vec2vfx(p0.x,p0.y),p2: Vec2vfx(p1.x,p1.y),r: max(a: p0.w,b: p1.w)); | 
| 86 |        | 
| 87 |       if (any(b: valid))  | 
| 88 |       { | 
| 89 |         Vec3vfx dp0dt = curve2D.template derivative0<VSIZEX>(0,N); | 
| 90 |         Vec3vfx dp1dt = curve2D.template derivative1<VSIZEX>(0,N); | 
| 91 |         dp0dt = select(s: reduce_max(a: abs(a: dp0dt)) < vfloatx(eps),t: Vec3vfx(p1-p0),f: dp0dt); | 
| 92 |         dp1dt = select(s: reduce_max(a: abs(a: dp1dt)) < vfloatx(eps),t: Vec3vfx(p1-p0),f: dp1dt); | 
| 93 |         const Vec3vfx n0(dp0dt.y,-dp0dt.x,0.0f); | 
| 94 |         const Vec3vfx n1(dp1dt.y,-dp1dt.x,0.0f); | 
| 95 |         const Vec3vfx nn0 = normalize(a: n0); | 
| 96 |         const Vec3vfx nn1 = normalize(a: n1); | 
| 97 |         const Vec3vfx lp0 = madd(a: p0.w,b: nn0,c: Vec3vfx(p0)); | 
| 98 |         const Vec3vfx lp1 = madd(a: p1.w,b: nn1,c: Vec3vfx(p1)); | 
| 99 |         const Vec3vfx up0 = nmadd(a: p0.w,b: nn0,c: Vec3vfx(p0)); | 
| 100 |         const Vec3vfx up1 = nmadd(a: p1.w,b: nn1,c: Vec3vfx(p1)); | 
| 101 |          | 
| 102 |         vfloatx vu,vv,vt; | 
| 103 |         vboolx valid0 = intersect_quad_backface_culling<VSIZEX>(valid0: valid,ray_org: zero,ray_dir: Vec3fa(0,0,1),ray_tnear,ray_tfar,quad_v0: lp0,quad_v1: lp1,quad_v2: up1,quad_v3: up0,u_o&: vu,v_o&: vv,t_o&: vt); | 
| 104 |  | 
| 105 |         if (any(b: valid0)) | 
| 106 |         { | 
| 107 |           /* ignore self intersections */ | 
| 108 |           if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) { | 
| 109 |             vfloatx r = lerp(a: p0.w, b: p1.w, t: vu); | 
| 110 |             valid0 &= vt > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale; | 
| 111 |           } | 
| 112 |            | 
| 113 |           if (any(b: valid0)) | 
| 114 |           { | 
| 115 |             vv = madd(a: 2.0f,b: vv,c: vfloatx(-1.0f)); | 
| 116 |             RibbonHit<NativeCurve3ff,VSIZEX> bhit(valid0,vu,vv,vt,0,N,curve3D); | 
| 117 |             ishit |= epilog(bhit.valid,bhit); | 
| 118 |           } | 
| 119 |         } | 
| 120 |       } | 
| 121 |        | 
| 122 |       if (unlikely(VSIZEX < N))  | 
| 123 |       { | 
| 124 |         /* process SIMD-size many segments per iteration */ | 
| 125 |         for (int i=VSIZEX; i<N; i+=VSIZEX) | 
| 126 |         { | 
| 127 |           /* evaluate the bezier curve */ | 
| 128 |           vboolx valid = vintx(i)+vintx(step) < vintx(N); | 
| 129 |           const Vec4vfx p0 = curve2D.template eval0<VSIZEX>(i,N); | 
| 130 |           const Vec4vfx p1 = curve2D.template eval1<VSIZEX>(i,N); | 
| 131 |           valid &= cylinder_culling_test(p0: zero,p1: Vec2vfx(p0.x,p0.y),p2: Vec2vfx(p1.x,p1.y),r: max(a: p0.w,b: p1.w)); | 
| 132 |           if (none(b: valid)) continue; | 
| 133 |            | 
| 134 |           Vec3vfx dp0dt = curve2D.template derivative0<VSIZEX>(i,N); | 
| 135 |           Vec3vfx dp1dt = curve2D.template derivative1<VSIZEX>(i,N); | 
| 136 |           dp0dt = select(s: reduce_max(a: abs(a: dp0dt)) < vfloatx(eps),t: Vec3vfx(p1-p0),f: dp0dt); | 
| 137 |           dp1dt = select(s: reduce_max(a: abs(a: dp1dt)) < vfloatx(eps),t: Vec3vfx(p1-p0),f: dp1dt); | 
| 138 |           const Vec3vfx n0(dp0dt.y,-dp0dt.x,0.0f); | 
| 139 |           const Vec3vfx n1(dp1dt.y,-dp1dt.x,0.0f); | 
| 140 |           const Vec3vfx nn0 = normalize(a: n0); | 
| 141 |           const Vec3vfx nn1 = normalize(a: n1); | 
| 142 |           const Vec3vfx lp0 = madd(a: p0.w,b: nn0,c: Vec3vfx(p0)); | 
| 143 |           const Vec3vfx lp1 = madd(a: p1.w,b: nn1,c: Vec3vfx(p1)); | 
| 144 |           const Vec3vfx up0 = nmadd(a: p0.w,b: nn0,c: Vec3vfx(p0)); | 
| 145 |           const Vec3vfx up1 = nmadd(a: p1.w,b: nn1,c: Vec3vfx(p1)); | 
| 146 |            | 
| 147 |           vfloatx vu,vv,vt; | 
| 148 |           vboolx valid0 = intersect_quad_backface_culling<VSIZEX>(valid0: valid,ray_org: zero,ray_dir: Vec3fa(0,0,1),ray_tnear,ray_tfar,quad_v0: lp0,quad_v1: lp1,quad_v2: up1,quad_v3: up0,u_o&: vu,v_o&: vv,t_o&: vt); | 
| 149 |  | 
| 150 |           if (any(b: valid0)) | 
| 151 |           { | 
| 152 |             /* ignore self intersections */ | 
| 153 |             if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) { | 
| 154 |               vfloatx r = lerp(a: p0.w, b: p1.w, t: vu); | 
| 155 |               valid0 &= vt > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR)*r*depth_scale; | 
| 156 |             } | 
| 157 |              | 
| 158 |             if (any(b: valid0)) | 
| 159 |             { | 
| 160 |               vv = madd(a: 2.0f,b: vv,c: vfloatx(-1.0f)); | 
| 161 |               RibbonHit<NativeCurve3ff,VSIZEX> bhit(valid0,vu,vv,vt,i,N,curve3D); | 
| 162 |               ishit |= epilog(bhit.valid,bhit); | 
| 163 |             } | 
| 164 |           } | 
| 165 |         } | 
| 166 |       } | 
| 167 |       return ishit; | 
| 168 |     } | 
| 169 |          | 
| 170 |     template<template<typename Ty> class NativeCurve> | 
| 171 |     struct RibbonCurve1Intersector1 | 
| 172 |     { | 
| 173 |       typedef NativeCurve<Vec3ff> NativeCurve3ff; | 
| 174 |        | 
| 175 |       template<typename Epilog> | 
| 176 |       __forceinline bool intersect(const CurvePrecalculations1& pre, Ray& ray, | 
| 177 |                                    IntersectContext* context, | 
| 178 |                                    const CurveGeometry* geom, const unsigned int primID, | 
| 179 |                                    const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3, | 
| 180 |                                    const Epilog& epilog) | 
| 181 |       { | 
| 182 |         const int N = geom->tessellationRate; | 
| 183 |         NativeCurve3ff curve(v0,v1,v2,v3); | 
| 184 |         curve = enlargeRadiusToMinWidth(context,geom,ray.org,curve); | 
| 185 |         return intersect_ribbon<NativeCurve3ff>(ray.org,ray.dir,ray.tnear(),ray.tfar, | 
| 186 |                                                 pre.ray_space,pre.depth_scale, | 
| 187 |                                                 curve,N, | 
| 188 |                                                 epilog); | 
| 189 |       } | 
| 190 |     }; | 
| 191 |      | 
| 192 |     template<template<typename Ty> class NativeCurve, int K> | 
| 193 |     struct RibbonCurve1IntersectorK | 
| 194 |     { | 
| 195 |       typedef NativeCurve<Vec3ff> NativeCurve3ff; | 
| 196 |        | 
| 197 |       template<typename Epilog> | 
| 198 |       __forceinline bool intersect(const CurvePrecalculationsK<K>& pre, RayK<K>& ray, size_t k, | 
| 199 |                                    IntersectContext* context, | 
| 200 |                                    const CurveGeometry* geom, const unsigned int primID, | 
| 201 |                                    const Vec3ff& v0, const Vec3ff& v1, const Vec3ff& v2, const Vec3ff& v3, | 
| 202 |                                    const Epilog& epilog) | 
| 203 |       { | 
| 204 |         const int N = geom->tessellationRate; | 
| 205 |         const Vec3fa ray_org(ray.org.x[k],ray.org.y[k],ray.org.z[k]); | 
| 206 |         const Vec3fa ray_dir(ray.dir.x[k],ray.dir.y[k],ray.dir.z[k]); | 
| 207 |         NativeCurve3ff curve(v0,v1,v2,v3); | 
| 208 |         curve = enlargeRadiusToMinWidth(context,geom,ray_org,curve); | 
| 209 |         return intersect_ribbon<NativeCurve3ff>(ray_org,ray_dir,ray.tnear()[k],ray.tfar[k], | 
| 210 |                                                 pre.ray_space[k],pre.depth_scale[k], | 
| 211 |                                                 curve,N, | 
| 212 |                                                 epilog); | 
| 213 |       } | 
| 214 |     }; | 
| 215 |   } | 
| 216 | } | 
| 217 |  |