| 1 | // Copyright 2009-2021 Intel Corporation |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| 4 | #pragma once |
| 5 | |
| 6 | #include "linearspace2.h" |
| 7 | #include "linearspace3.h" |
| 8 | #include "quaternion.h" |
| 9 | #include "bbox.h" |
| 10 | #include "vec4.h" |
| 11 | |
| 12 | namespace embree |
| 13 | { |
| 14 | #define VectorT typename L::Vector |
| 15 | #define ScalarT typename L::Vector::Scalar |
| 16 | |
| 17 | //////////////////////////////////////////////////////////////////////////////// |
| 18 | // Affine Space |
| 19 | //////////////////////////////////////////////////////////////////////////////// |
| 20 | |
| 21 | template<typename L> |
| 22 | struct AffineSpaceT |
| 23 | { |
| 24 | L l; /*< linear part of affine space */ |
| 25 | VectorT p; /*< affine part of affine space */ |
| 26 | |
| 27 | //////////////////////////////////////////////////////////////////////////////// |
| 28 | // Constructors, Assignment, Cast, Copy Operations |
| 29 | //////////////////////////////////////////////////////////////////////////////// |
| 30 | |
| 31 | __forceinline AffineSpaceT ( ) { } |
| 32 | __forceinline AffineSpaceT ( const AffineSpaceT& other ) { l = other.l; p = other.p; } |
| 33 | __forceinline AffineSpaceT ( const L & other ) { l = other ; p = VectorT(zero); } |
| 34 | __forceinline AffineSpaceT& operator=( const AffineSpaceT& other ) { l = other.l; p = other.p; return *this; } |
| 35 | |
| 36 | __forceinline AffineSpaceT( const VectorT& vx, const VectorT& vy, const VectorT& vz, const VectorT& p ) : l(vx,vy,vz), p(p) {} |
| 37 | __forceinline AffineSpaceT( const L& l, const VectorT& p ) : l(l), p(p) {} |
| 38 | |
| 39 | template<typename L1> __forceinline AffineSpaceT( const AffineSpaceT<L1>& s ) : l(s.l), p(s.p) {} |
| 40 | |
| 41 | //////////////////////////////////////////////////////////////////////////////// |
| 42 | // Constants |
| 43 | //////////////////////////////////////////////////////////////////////////////// |
| 44 | |
| 45 | __forceinline AffineSpaceT( ZeroTy ) : l(zero), p(zero) {} |
| 46 | __forceinline AffineSpaceT( OneTy ) : l(one), p(zero) {} |
| 47 | |
| 48 | /*! return matrix for scaling */ |
| 49 | static __forceinline AffineSpaceT scale(const VectorT& s) { return L::scale(s); } |
| 50 | |
| 51 | /*! return matrix for translation */ |
| 52 | static __forceinline AffineSpaceT translate(const VectorT& p) { return AffineSpaceT(one,p); } |
| 53 | |
| 54 | /*! return matrix for rotation, only in 2D */ |
| 55 | static __forceinline AffineSpaceT rotate(const ScalarT& r) { return L::rotate(r); } |
| 56 | |
| 57 | /*! return matrix for rotation around arbitrary point (2D) or axis (3D) */ |
| 58 | static __forceinline AffineSpaceT rotate(const VectorT& u, const ScalarT& r) { return L::rotate(u,r); } |
| 59 | |
| 60 | /*! return matrix for rotation around arbitrary axis and point, only in 3D */ |
| 61 | static __forceinline AffineSpaceT rotate(const VectorT& p, const VectorT& u, const ScalarT& r) { return translate(+p) * rotate(u,r) * translate(-p); } |
| 62 | |
| 63 | /*! return matrix for looking at given point, only in 3D */ |
| 64 | static __forceinline AffineSpaceT lookat(const VectorT& eye, const VectorT& point, const VectorT& up) { |
| 65 | VectorT Z = normalize(point-eye); |
| 66 | VectorT U = normalize(cross(up,Z)); |
| 67 | VectorT V = normalize(cross(Z,U)); |
| 68 | return AffineSpaceT(L(U,V,Z),eye); |
| 69 | } |
| 70 | |
| 71 | }; |
| 72 | |
| 73 | // template specialization to get correct identity matrix for type AffineSpace3fa |
| 74 | template<> |
| 75 | __forceinline AffineSpaceT<LinearSpace3ff>::AffineSpaceT( OneTy ) : l(one), p(0.f, 0.f, 0.f, 1.f) {} |
| 76 | |
| 77 | //////////////////////////////////////////////////////////////////////////////// |
| 78 | // Unary Operators |
| 79 | //////////////////////////////////////////////////////////////////////////////// |
| 80 | |
| 81 | template<typename L> __forceinline AffineSpaceT<L> operator -( const AffineSpaceT<L>& a ) { return AffineSpaceT<L>(-a.l,-a.p); } |
| 82 | template<typename L> __forceinline AffineSpaceT<L> operator +( const AffineSpaceT<L>& a ) { return AffineSpaceT<L>(+a.l,+a.p); } |
| 83 | template<typename L> __forceinline AffineSpaceT<L> rcp( const AffineSpaceT<L>& a ) { L il = rcp(a.l); return AffineSpaceT<L>(il,-(il*a.p)); } |
| 84 | |
| 85 | //////////////////////////////////////////////////////////////////////////////// |
| 86 | // Binary Operators |
| 87 | //////////////////////////////////////////////////////////////////////////////// |
| 88 | |
| 89 | template<typename L> __forceinline const AffineSpaceT<L> operator +( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return AffineSpaceT<L>(a.l+b.l,a.p+b.p); } |
| 90 | template<typename L> __forceinline const AffineSpaceT<L> operator -( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return AffineSpaceT<L>(a.l-b.l,a.p-b.p); } |
| 91 | |
| 92 | template<typename L> __forceinline const AffineSpaceT<L> operator *( const ScalarT & a, const AffineSpaceT<L>& b ) { return AffineSpaceT<L>(a*b.l,a*b.p); } |
| 93 | template<typename L> __forceinline const AffineSpaceT<L> operator *( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return AffineSpaceT<L>(a.l*b.l,a.l*b.p+a.p); } |
| 94 | template<typename L> __forceinline const AffineSpaceT<L> operator /( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return a * rcp(b); } |
| 95 | template<typename L> __forceinline const AffineSpaceT<L> operator /( const AffineSpaceT<L>& a, const ScalarT & b ) { return a * rcp(b); } |
| 96 | |
| 97 | template<typename L> __forceinline AffineSpaceT<L>& operator *=( AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return a = a * b; } |
| 98 | template<typename L> __forceinline AffineSpaceT<L>& operator *=( AffineSpaceT<L>& a, const ScalarT & b ) { return a = a * b; } |
| 99 | template<typename L> __forceinline AffineSpaceT<L>& operator /=( AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return a = a / b; } |
| 100 | template<typename L> __forceinline AffineSpaceT<L>& operator /=( AffineSpaceT<L>& a, const ScalarT & b ) { return a = a / b; } |
| 101 | |
| 102 | template<typename L> __forceinline VectorT xfmPoint (const AffineSpaceT<L>& m, const VectorT& p) { return madd(VectorT(p.x),m.l.vx,madd(VectorT(p.y),m.l.vy,madd(VectorT(p.z),m.l.vz,m.p))); } |
| 103 | template<typename L> __forceinline VectorT xfmVector(const AffineSpaceT<L>& m, const VectorT& v) { return xfmVector(m.l,v); } |
| 104 | template<typename L> __forceinline VectorT xfmNormal(const AffineSpaceT<L>& m, const VectorT& n) { return xfmNormal(m.l,n); } |
| 105 | |
| 106 | __forceinline const BBox<Vec3fa> xfmBounds(const AffineSpaceT<LinearSpace3<Vec3fa> >& m, const BBox<Vec3fa>& b) |
| 107 | { |
| 108 | BBox3fa dst = empty; |
| 109 | const Vec3fa p0(b.lower.x,b.lower.y,b.lower.z); dst.extend(xfmPoint(m,p0)); |
| 110 | const Vec3fa p1(b.lower.x,b.lower.y,b.upper.z); dst.extend(xfmPoint(m,p1)); |
| 111 | const Vec3fa p2(b.lower.x,b.upper.y,b.lower.z); dst.extend(xfmPoint(m,p2)); |
| 112 | const Vec3fa p3(b.lower.x,b.upper.y,b.upper.z); dst.extend(xfmPoint(m,p3)); |
| 113 | const Vec3fa p4(b.upper.x,b.lower.y,b.lower.z); dst.extend(xfmPoint(m,p4)); |
| 114 | const Vec3fa p5(b.upper.x,b.lower.y,b.upper.z); dst.extend(xfmPoint(m,p5)); |
| 115 | const Vec3fa p6(b.upper.x,b.upper.y,b.lower.z); dst.extend(xfmPoint(m,p6)); |
| 116 | const Vec3fa p7(b.upper.x,b.upper.y,b.upper.z); dst.extend(xfmPoint(m,p7)); |
| 117 | return dst; |
| 118 | } |
| 119 | |
| 120 | //////////////////////////////////////////////////////////////////////////////// |
| 121 | /// Comparison Operators |
| 122 | //////////////////////////////////////////////////////////////////////////////// |
| 123 | |
| 124 | template<typename L> __forceinline bool operator ==( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return a.l == b.l && a.p == b.p; } |
| 125 | template<typename L> __forceinline bool operator !=( const AffineSpaceT<L>& a, const AffineSpaceT<L>& b ) { return a.l != b.l || a.p != b.p; } |
| 126 | |
| 127 | //////////////////////////////////////////////////////////////////////////////// |
| 128 | /// Select |
| 129 | //////////////////////////////////////////////////////////////////////////////// |
| 130 | |
| 131 | template<typename L> __forceinline AffineSpaceT<L> select ( const typename L::Vector::Scalar::Bool& s, const AffineSpaceT<L>& t, const AffineSpaceT<L>& f ) { |
| 132 | return AffineSpaceT<L>(select(s,t.l,f.l),select(s,t.p,f.p)); |
| 133 | } |
| 134 | |
| 135 | //////////////////////////////////////////////////////////////////////////////// |
| 136 | // Output Operators |
| 137 | //////////////////////////////////////////////////////////////////////////////// |
| 138 | |
| 139 | template<typename L> static embree_ostream operator<<(embree_ostream cout, const AffineSpaceT<L>& m) { |
| 140 | return cout << "{ l = " << m.l << ", p = " << m.p << " }" ; |
| 141 | } |
| 142 | |
| 143 | //////////////////////////////////////////////////////////////////////////////// |
| 144 | // Template Instantiations |
| 145 | //////////////////////////////////////////////////////////////////////////////// |
| 146 | |
| 147 | typedef AffineSpaceT<LinearSpace2f> AffineSpace2f; |
| 148 | typedef AffineSpaceT<LinearSpace3f> AffineSpace3f; |
| 149 | typedef AffineSpaceT<LinearSpace3fa> AffineSpace3fa; |
| 150 | typedef AffineSpaceT<LinearSpace3fx> AffineSpace3fx; |
| 151 | typedef AffineSpaceT<LinearSpace3ff> AffineSpace3ff; |
| 152 | typedef AffineSpaceT<Quaternion3f > OrthonormalSpace3f; |
| 153 | |
| 154 | template<int N> using AffineSpace3vf = AffineSpaceT<LinearSpace3<Vec3<vfloat<N>>>>; |
| 155 | typedef AffineSpaceT<LinearSpace3<Vec3<vfloat<4>>>> AffineSpace3vf4; |
| 156 | typedef AffineSpaceT<LinearSpace3<Vec3<vfloat<8>>>> AffineSpace3vf8; |
| 157 | typedef AffineSpaceT<LinearSpace3<Vec3<vfloat<16>>>> AffineSpace3vf16; |
| 158 | |
| 159 | template<int N> using AffineSpace3vff = AffineSpaceT<LinearSpace3<Vec4<vfloat<N>>>>; |
| 160 | typedef AffineSpaceT<LinearSpace3<Vec4<vfloat<4>>>> AffineSpace3vfa4; |
| 161 | typedef AffineSpaceT<LinearSpace3<Vec4<vfloat<8>>>> AffineSpace3vfa8; |
| 162 | typedef AffineSpaceT<LinearSpace3<Vec4<vfloat<16>>>> AffineSpace3vfa16; |
| 163 | |
| 164 | ////////////////////////////////////////////////////////////////////////////// |
| 165 | /// Interpolation |
| 166 | ////////////////////////////////////////////////////////////////////////////// |
| 167 | template<typename T, typename R> |
| 168 | __forceinline AffineSpaceT<T> lerp(const AffineSpaceT<T>& M0, |
| 169 | const AffineSpaceT<T>& M1, |
| 170 | const R& t) |
| 171 | { |
| 172 | return AffineSpaceT<T>(lerp(M0.l,M1.l,t),lerp(M0.p,M1.p,t)); |
| 173 | } |
| 174 | |
| 175 | // slerp interprets the 16 floats of the matrix M = D * R * S as components of |
| 176 | // three matrizes (D, R, S) that are interpolated individually. |
| 177 | template<typename T> __forceinline AffineSpaceT<LinearSpace3<Vec3<T>>> |
| 178 | slerp(const AffineSpaceT<LinearSpace3<Vec4<T>>>& M0, |
| 179 | const AffineSpaceT<LinearSpace3<Vec4<T>>>& M1, |
| 180 | const T& t) |
| 181 | { |
| 182 | QuaternionT<T> q0(M0.p.w, M0.l.vx.w, M0.l.vy.w, M0.l.vz.w); |
| 183 | QuaternionT<T> q1(M1.p.w, M1.l.vx.w, M1.l.vy.w, M1.l.vz.w); |
| 184 | QuaternionT<T> q = slerp(q0, q1, t); |
| 185 | |
| 186 | AffineSpaceT<LinearSpace3<Vec3<T>>> S = lerp(M0, M1, t); |
| 187 | AffineSpaceT<LinearSpace3<Vec3<T>>> D(one); |
| 188 | D.p.x = S.l.vx.y; |
| 189 | D.p.y = S.l.vx.z; |
| 190 | D.p.z = S.l.vy.z; |
| 191 | S.l.vx.y = 0; |
| 192 | S.l.vx.z = 0; |
| 193 | S.l.vy.z = 0; |
| 194 | |
| 195 | AffineSpaceT<LinearSpace3<Vec3<T>>> R = LinearSpace3<Vec3<T>>(q); |
| 196 | return D * R * S; |
| 197 | } |
| 198 | |
| 199 | // this is a specialized version for Vec3fa because that does |
| 200 | // not play along nicely with the other templated Vec3/Vec4 types |
| 201 | __forceinline AffineSpace3fa slerp(const AffineSpace3ff& M0, |
| 202 | const AffineSpace3ff& M1, |
| 203 | const float& t) |
| 204 | { |
| 205 | Quaternion3f q0(M0.p.w, M0.l.vx.w, M0.l.vy.w, M0.l.vz.w); |
| 206 | Quaternion3f q1(M1.p.w, M1.l.vx.w, M1.l.vy.w, M1.l.vz.w); |
| 207 | Quaternion3f q = slerp(q0, q1, t); |
| 208 | |
| 209 | AffineSpace3fa S = lerp(M0, M1, t); |
| 210 | AffineSpace3fa D(one); |
| 211 | D.p.x = S.l.vx.y; |
| 212 | D.p.y = S.l.vx.z; |
| 213 | D.p.z = S.l.vy.z; |
| 214 | S.l.vx.y = 0; |
| 215 | S.l.vx.z = 0; |
| 216 | S.l.vy.z = 0; |
| 217 | |
| 218 | AffineSpace3fa R = LinearSpace3fa(q); |
| 219 | return D * R * S; |
| 220 | } |
| 221 | |
| 222 | __forceinline AffineSpace3fa quaternionDecompositionToAffineSpace(const AffineSpace3ff& qd) |
| 223 | { |
| 224 | // compute affine transform from quaternion decomposition |
| 225 | Quaternion3f q(qd.p.w, qd.l.vx.w, qd.l.vy.w, qd.l.vz.w); |
| 226 | AffineSpace3fa M = qd; |
| 227 | AffineSpace3fa D(one); |
| 228 | D.p.x = M.l.vx.y; |
| 229 | D.p.y = M.l.vx.z; |
| 230 | D.p.z = M.l.vy.z; |
| 231 | M.l.vx.y = 0; |
| 232 | M.l.vx.z = 0; |
| 233 | M.l.vy.z = 0; |
| 234 | AffineSpace3fa R = LinearSpace3fa(q); |
| 235 | return D * R * M; |
| 236 | } |
| 237 | |
| 238 | __forceinline void quaternionDecomposition(const AffineSpace3ff& qd, Vec3fa& T, Quaternion3f& q, AffineSpace3fa& S) |
| 239 | { |
| 240 | q = Quaternion3f(qd.p.w, qd.l.vx.w, qd.l.vy.w, qd.l.vz.w); |
| 241 | S = qd; |
| 242 | T.x = qd.l.vx.y; |
| 243 | T.y = qd.l.vx.z; |
| 244 | T.z = qd.l.vy.z; |
| 245 | S.l.vx.y = 0; |
| 246 | S.l.vx.z = 0; |
| 247 | S.l.vy.z = 0; |
| 248 | } |
| 249 | |
| 250 | __forceinline AffineSpace3fx quaternionDecomposition(Vec3fa const& T, Quaternion3f const& q, AffineSpace3fa const& S) |
| 251 | { |
| 252 | AffineSpace3ff M = S; |
| 253 | M.l.vx.w = q.i; |
| 254 | M.l.vy.w = q.j; |
| 255 | M.l.vz.w = q.k; |
| 256 | M.p.w = q.r; |
| 257 | M.l.vx.y = T.x; |
| 258 | M.l.vx.z = T.y; |
| 259 | M.l.vy.z = T.z; |
| 260 | return M; |
| 261 | } |
| 262 | |
| 263 | struct __aligned(16) QuaternionDecomposition |
| 264 | { |
| 265 | float scale_x = 1.f; |
| 266 | float scale_y = 1.f; |
| 267 | float scale_z = 1.f; |
| 268 | float skew_xy = 0.f; |
| 269 | float skew_xz = 0.f; |
| 270 | float skew_yz = 0.f; |
| 271 | float shift_x = 0.f; |
| 272 | float shift_y = 0.f; |
| 273 | float shift_z = 0.f; |
| 274 | float quaternion_r = 1.f; |
| 275 | float quaternion_i = 0.f; |
| 276 | float quaternion_j = 0.f; |
| 277 | float quaternion_k = 0.f; |
| 278 | float translation_x = 0.f; |
| 279 | float translation_y = 0.f; |
| 280 | float translation_z = 0.f; |
| 281 | }; |
| 282 | |
| 283 | __forceinline QuaternionDecomposition quaternionDecomposition(AffineSpace3ff const& M) |
| 284 | { |
| 285 | QuaternionDecomposition qd; |
| 286 | qd.scale_x = M.l.vx.x; |
| 287 | qd.scale_y = M.l.vy.y; |
| 288 | qd.scale_z = M.l.vz.z; |
| 289 | qd.shift_x = M.p.x; |
| 290 | qd.shift_y = M.p.y; |
| 291 | qd.shift_z = M.p.z; |
| 292 | qd.translation_x = M.l.vx.y; |
| 293 | qd.translation_y = M.l.vx.z; |
| 294 | qd.translation_z = M.l.vy.z; |
| 295 | qd.skew_xy = M.l.vy.x; |
| 296 | qd.skew_xz = M.l.vz.x; |
| 297 | qd.skew_yz = M.l.vz.y; |
| 298 | qd.quaternion_r = M.p.w; |
| 299 | qd.quaternion_i = M.l.vx.w; |
| 300 | qd.quaternion_j = M.l.vy.w; |
| 301 | qd.quaternion_k = M.l.vz.w; |
| 302 | return qd; |
| 303 | } |
| 304 | |
| 305 | //////////////////////////////////////////////////////////////////////////////// |
| 306 | /* |
| 307 | * ! Template Specialization for 2D: return matrix for rotation around point |
| 308 | * (rotation around arbitrarty vector is not meaningful in 2D) |
| 309 | */ |
| 310 | template<> __forceinline |
| 311 | AffineSpace2f AffineSpace2f::rotate(const Vec2f& p, const float& r) { |
| 312 | return translate(+p)*AffineSpace2f(LinearSpace2f::rotate(r))*translate(-p); |
| 313 | } |
| 314 | |
| 315 | //////////////////////////////////////////////////////////////////////////////// |
| 316 | // Similarity Transform |
| 317 | // |
| 318 | // checks, if M is a similarity transformation, i.e if there exists a factor D |
| 319 | // such that for all x,y: distance(Mx, My) = D * distance(x, y) |
| 320 | //////////////////////////////////////////////////////////////////////////////// |
| 321 | __forceinline bool similarityTransform(const AffineSpace3fa& M, float* D) |
| 322 | { |
| 323 | if (D) *D = 0.f; |
| 324 | if (abs(dot(M.l.vx, M.l.vy)) > 1e-5f) return false; |
| 325 | if (abs(dot(M.l.vx, M.l.vz)) > 1e-5f) return false; |
| 326 | if (abs(dot(M.l.vy, M.l.vz)) > 1e-5f) return false; |
| 327 | |
| 328 | const float D_x = dot(M.l.vx, M.l.vx); |
| 329 | const float D_y = dot(M.l.vy, M.l.vy); |
| 330 | const float D_z = dot(M.l.vz, M.l.vz); |
| 331 | |
| 332 | if (abs(D_x - D_y) > 1e-5f || |
| 333 | abs(D_x - D_z) > 1e-5f || |
| 334 | abs(D_y - D_z) > 1e-5f) |
| 335 | return false; |
| 336 | |
| 337 | if (D) *D = sqrtf(D_x); |
| 338 | return true; |
| 339 | } |
| 340 | |
| 341 | __forceinline void AffineSpace3fa_store_unaligned(const AffineSpace3fa &source, AffineSpace3fa* ptr) |
| 342 | { |
| 343 | Vec3fa::storeu(&ptr->l.vx, source.l.vx); |
| 344 | Vec3fa::storeu(&ptr->l.vy, source.l.vy); |
| 345 | Vec3fa::storeu(&ptr->l.vz, source.l.vz); |
| 346 | Vec3fa::storeu(&ptr->p, source.p); |
| 347 | } |
| 348 | |
| 349 | __forceinline AffineSpace3fa AffineSpace3fa_load_unaligned(AffineSpace3fa* ptr) |
| 350 | { |
| 351 | AffineSpace3fa space; |
| 352 | space.l.vx = Vec3fa::loadu(&ptr->l.vx); |
| 353 | space.l.vy = Vec3fa::loadu(&ptr->l.vy); |
| 354 | space.l.vz = Vec3fa::loadu(&ptr->l.vz); |
| 355 | space.p = Vec3fa::loadu(&ptr->p); |
| 356 | return space; |
| 357 | } |
| 358 | |
| 359 | #undef VectorT |
| 360 | #undef ScalarT |
| 361 | } |
| 362 | |