1// Copyright 2009-2021 Intel Corporation
2// SPDX-License-Identifier: Apache-2.0
3
4#pragma once
5
6#include "default.h"
7#include "geometry.h"
8#include "buffer.h"
9
10#include "../subdiv/bezier_curve.h"
11#include "../subdiv/hermite_curve.h"
12#include "../subdiv/bspline_curve.h"
13#include "../subdiv/catmullrom_curve.h"
14#include "../subdiv/linear_bezier_patch.h"
15
16namespace embree
17{
18 /*! represents an array of bicubic bezier curves */
19 struct CurveGeometry : public Geometry
20 {
21 /*! type of this geometry */
22 static const Geometry::GTypeMask geom_type = Geometry::MTY_CURVE4;
23
24 public:
25
26 /*! bezier curve construction */
27 CurveGeometry (Device* device, Geometry::GType gtype);
28
29 public:
30 void setMask(unsigned mask);
31 void setNumTimeSteps (unsigned int numTimeSteps);
32 void setVertexAttributeCount (unsigned int N);
33 void setBuffer(RTCBufferType type, unsigned int slot, RTCFormat format, const Ref<Buffer>& buffer, size_t offset, size_t stride, unsigned int num);
34 void* getBuffer(RTCBufferType type, unsigned int slot);
35 void updateBuffer(RTCBufferType type, unsigned int slot);
36 void commit();
37 bool verify();
38 void setTessellationRate(float N);
39 void setMaxRadiusScale(float s);
40 void addElementsToCount (GeometryCounts & counts) const;
41
42 public:
43
44 /*! returns the number of vertices */
45 __forceinline size_t numVertices() const {
46 return vertices[0].size();
47 }
48
49 /*! returns the i'th curve */
50 __forceinline const unsigned int& curve(size_t i) const {
51 return curves[i];
52 }
53
54 /*! returns i'th vertex of the first time step */
55 __forceinline Vec3ff vertex(size_t i) const {
56 return vertices0[i];
57 }
58
59 /*! returns i'th normal of the first time step */
60 __forceinline Vec3fa normal(size_t i) const {
61 return normals0[i];
62 }
63
64 /*! returns i'th tangent of the first time step */
65 __forceinline Vec3ff tangent(size_t i) const {
66 return tangents0[i];
67 }
68
69 /*! returns i'th normal derivative of the first time step */
70 __forceinline Vec3fa dnormal(size_t i) const {
71 return dnormals0[i];
72 }
73
74 /*! returns i'th radius of the first time step */
75 __forceinline float radius(size_t i) const {
76 return vertices0[i].w;
77 }
78
79 /*! returns i'th vertex of itime'th timestep */
80 __forceinline Vec3ff vertex(size_t i, size_t itime) const {
81 return vertices[itime][i];
82 }
83
84 /*! returns i'th normal of itime'th timestep */
85 __forceinline Vec3fa normal(size_t i, size_t itime) const {
86 return normals[itime][i];
87 }
88
89 /*! returns i'th tangent of itime'th timestep */
90 __forceinline Vec3ff tangent(size_t i, size_t itime) const {
91 return tangents[itime][i];
92 }
93
94 /*! returns i'th normal derivative of itime'th timestep */
95 __forceinline Vec3fa dnormal(size_t i, size_t itime) const {
96 return dnormals[itime][i];
97 }
98
99 /*! returns i'th radius of itime'th timestep */
100 __forceinline float radius(size_t i, size_t itime) const {
101 return vertices[itime][i].w;
102 }
103
104 /*! gathers the curve starting with i'th vertex */
105 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, size_t i) const
106 {
107 p0 = vertex(i+0);
108 p1 = vertex(i+1);
109 p2 = vertex(i+2);
110 p3 = vertex(i+3);
111 }
112
113 /*! gathers the curve starting with i'th vertex of itime'th timestep */
114 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, size_t i, size_t itime) const
115 {
116 p0 = vertex(i+0,itime);
117 p1 = vertex(i+1,itime);
118 p2 = vertex(i+2,itime);
119 p3 = vertex(i+3,itime);
120 }
121
122 /*! gathers the curve starting with i'th vertex */
123 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i) const
124 {
125 p0 = vertex(i+0);
126 p1 = vertex(i+1);
127 p2 = vertex(i+2);
128 p3 = vertex(i+3);
129 n0 = normal(i+0);
130 n1 = normal(i+1);
131 n2 = normal(i+2);
132 n3 = normal(i+3);
133 }
134
135 /*! gathers the curve starting with i'th vertex of itime'th timestep */
136 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i, size_t itime) const
137 {
138 p0 = vertex(i+0,itime);
139 p1 = vertex(i+1,itime);
140 p2 = vertex(i+2,itime);
141 p3 = vertex(i+3,itime);
142 n0 = normal(i+0,itime);
143 n1 = normal(i+1,itime);
144 n2 = normal(i+2,itime);
145 n3 = normal(i+3,itime);
146 }
147
148 /*! prefetches the curve starting with i'th vertex of itime'th timestep */
149 __forceinline void prefetchL1_vertices(size_t i) const
150 {
151 prefetchL1(vertices0.getPtr(i)+0);
152 prefetchL1(vertices0.getPtr(i)+64);
153 }
154
155 /*! prefetches the curve starting with i'th vertex of itime'th timestep */
156 __forceinline void prefetchL2_vertices(size_t i) const
157 {
158 prefetchL2(vertices0.getPtr(i)+0);
159 prefetchL2(vertices0.getPtr(i)+64);
160 }
161
162 /*! loads curve vertices for specified time */
163 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, size_t i, float time) const
164 {
165 float ftime;
166 const size_t itime = timeSegment(time, ftime);
167
168 const float t0 = 1.0f - ftime;
169 const float t1 = ftime;
170 Vec3ff a0,a1,a2,a3;
171 gather(a0,a1,a2,a3,i,itime);
172 Vec3ff b0,b1,b2,b3;
173 gather(b0,b1,b2,b3,i,itime+1);
174 p0 = madd(Vec3ff(t0),a0,t1*b0);
175 p1 = madd(Vec3ff(t0),a1,t1*b1);
176 p2 = madd(Vec3ff(t0),a2,t1*b2);
177 p3 = madd(Vec3ff(t0),a3,t1*b3);
178 }
179
180 /*! loads curve vertices for specified time */
181 __forceinline void gather(Vec3ff& p0, Vec3ff& p1, Vec3ff& p2, Vec3ff& p3, Vec3fa& n0, Vec3fa& n1, Vec3fa& n2, Vec3fa& n3, size_t i, float time) const
182 {
183 float ftime;
184 const size_t itime = timeSegment(time, ftime);
185
186 const float t0 = 1.0f - ftime;
187 const float t1 = ftime;
188 Vec3ff a0,a1,a2,a3; Vec3fa an0,an1,an2,an3;
189 gather(a0,a1,a2,a3,an0,an1,an2,an3,i,itime);
190 Vec3ff b0,b1,b2,b3; Vec3fa bn0,bn1,bn2,bn3;
191 gather(b0,b1,b2,b3,bn0,bn1,bn2,bn3,i,itime+1);
192 p0 = madd(Vec3ff(t0),a0,t1*b0);
193 p1 = madd(Vec3ff(t0),a1,t1*b1);
194 p2 = madd(Vec3ff(t0),a2,t1*b2);
195 p3 = madd(Vec3ff(t0),a3,t1*b3);
196 n0 = madd(Vec3ff(t0),an0,t1*bn0);
197 n1 = madd(Vec3ff(t0),an1,t1*bn1);
198 n2 = madd(Vec3ff(t0),an2,t1*bn2);
199 n3 = madd(Vec3ff(t0),an3,t1*bn3);
200 }
201
202 template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
203 __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
204 {
205 Vec3ff v0,v1,v2,v3; Vec3fa n0,n1,n2,n3;
206 unsigned int vertexID = curve(primID);
207 gather(v0,v1,v2,v3,n0,n1,n2,n3,vertexID,itime);
208 SourceCurve3ff ccurve(v0,v1,v2,v3);
209 SourceCurve3fa ncurve(n0,n1,n2,n3);
210 ccurve = enlargeRadiusToMinWidth(context,this,ray_org,ccurve);
211 return TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(ccurve,ncurve);
212 }
213
214 template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
215 __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
216 {
217 float ftime;
218 const size_t itime = timeSegment(time, ftime);
219 const TensorLinearCubicBezierSurface3fa curve0 = getNormalOrientedCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context,ray_org,primID,itime+0);
220 const TensorLinearCubicBezierSurface3fa curve1 = getNormalOrientedCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context,ray_org,primID,itime+1);
221 return clerp(curve0,curve1,ftime);
222 }
223
224 /*! gathers the hermite curve starting with i'th vertex */
225 __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3ff& p1, Vec3ff& t1, size_t i) const
226 {
227 p0 = vertex (i+0);
228 p1 = vertex (i+1);
229 t0 = tangent(i+0);
230 t1 = tangent(i+1);
231 }
232
233 /*! gathers the hermite curve starting with i'th vertex of itime'th timestep */
234 __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3ff& p1, Vec3ff& t1, size_t i, size_t itime) const
235 {
236 p0 = vertex (i+0,itime);
237 p1 = vertex (i+1,itime);
238 t0 = tangent(i+0,itime);
239 t1 = tangent(i+1,itime);
240 }
241
242 /*! loads curve vertices for specified time */
243 __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3ff& p1, Vec3ff& t1, size_t i, float time) const
244 {
245 float ftime;
246 const size_t itime = timeSegment(time, ftime);
247 const float f0 = 1.0f - ftime, f1 = ftime;
248 Vec3ff ap0,at0,ap1,at1;
249 gather_hermite(ap0,at0,ap1,at1,i,itime);
250 Vec3ff bp0,bt0,bp1,bt1;
251 gather_hermite(bp0,bt0,bp1,bt1,i,itime+1);
252 p0 = madd(Vec3ff(f0),ap0,f1*bp0);
253 t0 = madd(Vec3ff(f0),at0,f1*bt0);
254 p1 = madd(Vec3ff(f0),ap1,f1*bp1);
255 t1 = madd(Vec3ff(f0),at1,f1*bt1);
256 }
257
258 /*! gathers the hermite curve starting with i'th vertex */
259 __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3ff& t1, Vec3fa& n1, Vec3fa& dn1, size_t i) const
260 {
261 p0 = vertex (i+0);
262 p1 = vertex (i+1);
263 t0 = tangent(i+0);
264 t1 = tangent(i+1);
265 n0 = normal(i+0);
266 n1 = normal(i+1);
267 dn0 = dnormal(i+0);
268 dn1 = dnormal(i+1);
269 }
270
271 /*! gathers the hermite curve starting with i'th vertex of itime'th timestep */
272 __forceinline void gather_hermite(Vec3ff& p0, Vec3ff& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3ff& t1, Vec3fa& n1, Vec3fa& dn1, size_t i, size_t itime) const
273 {
274 p0 = vertex (i+0,itime);
275 p1 = vertex (i+1,itime);
276 t0 = tangent(i+0,itime);
277 t1 = tangent(i+1,itime);
278 n0 = normal(i+0,itime);
279 n1 = normal(i+1,itime);
280 dn0 = dnormal(i+0,itime);
281 dn1 = dnormal(i+1,itime);
282 }
283
284 /*! loads curve vertices for specified time */
285 __forceinline void gather_hermite(Vec3ff& p0, Vec3fa& t0, Vec3fa& n0, Vec3fa& dn0, Vec3ff& p1, Vec3fa& t1, Vec3fa& n1, Vec3fa& dn1, size_t i, float time) const
286 {
287 float ftime;
288 const size_t itime = timeSegment(time, ftime);
289 const float f0 = 1.0f - ftime, f1 = ftime;
290 Vec3ff ap0,at0,ap1,at1; Vec3fa an0,adn0,an1,adn1;
291 gather_hermite(ap0,at0,an0,adn0,ap1,at1,an1,adn1,i,itime);
292 Vec3ff bp0,bt0,bp1,bt1; Vec3fa bn0,bdn0,bn1,bdn1;
293 gather_hermite(bp0,bt0,bn0,bdn0,bp1,bt1,bn1,bdn1,i,itime+1);
294 p0 = madd(Vec3ff(f0),ap0,f1*bp0);
295 t0 = madd(Vec3ff(f0),at0,f1*bt0);
296 n0 = madd(Vec3ff(f0),an0,f1*bn0);
297 dn0= madd(Vec3ff(f0),adn0,f1*bdn0);
298 p1 = madd(Vec3ff(f0),ap1,f1*bp1);
299 t1 = madd(Vec3ff(f0),at1,f1*bt1);
300 n1 = madd(Vec3ff(f0),an1,f1*bn1);
301 dn1= madd(Vec3ff(f0),adn1,f1*bdn1);
302 }
303
304 template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
305 __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const size_t itime) const
306 {
307 Vec3ff v0,t0,v1,t1; Vec3fa n0,dn0,n1,dn1;
308 unsigned int vertexID = curve(primID);
309 gather_hermite(v0,t0,n0,dn0,v1,t1,n1,dn1,vertexID,itime);
310
311 SourceCurve3ff ccurve(v0,t0,v1,t1);
312 SourceCurve3fa ncurve(n0,dn0,n1,dn1);
313 ccurve = enlargeRadiusToMinWidth(context,this,ray_org,ccurve);
314 return TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(ccurve,ncurve);
315 }
316
317 template<typename SourceCurve3ff, typename SourceCurve3fa, typename TensorLinearCubicBezierSurface3fa>
318 __forceinline TensorLinearCubicBezierSurface3fa getNormalOrientedHermiteCurve(IntersectContext* context, const Vec3fa& ray_org, const unsigned int primID, const float time) const
319 {
320 float ftime;
321 const size_t itime = timeSegment(time, ftime);
322 const TensorLinearCubicBezierSurface3fa curve0 = getNormalOrientedHermiteCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,itime+0);
323 const TensorLinearCubicBezierSurface3fa curve1 = getNormalOrientedHermiteCurve<SourceCurve3ff, SourceCurve3fa, TensorLinearCubicBezierSurface3fa>(context, ray_org, primID,itime+1);
324 return clerp(curve0,curve1,ftime);
325 }
326
327 private:
328 void resizeBuffers(unsigned int numSteps);
329
330 public:
331 BufferView<unsigned int> curves; //!< array of curve indices
332 BufferView<Vec3ff> vertices0; //!< fast access to first vertex buffer
333 BufferView<Vec3fa> normals0; //!< fast access to first normal buffer
334 BufferView<Vec3ff> tangents0; //!< fast access to first tangent buffer
335 BufferView<Vec3fa> dnormals0; //!< fast access to first normal derivative buffer
336 vector<BufferView<Vec3ff>> vertices; //!< vertex array for each timestep
337 vector<BufferView<Vec3fa>> normals; //!< normal array for each timestep
338 vector<BufferView<Vec3ff>> tangents; //!< tangent array for each timestep
339 vector<BufferView<Vec3fa>> dnormals; //!< normal derivative array for each timestep
340 BufferView<char> flags; //!< start, end flag per segment
341 vector<BufferView<char>> vertexAttribs; //!< user buffers
342 int tessellationRate; //!< tessellation rate for flat curve
343 float maxRadiusScale = 1.0; //!< maximal min-width scaling of curve radii
344 };
345
346 namespace isa
347 {
348
349 template<template<typename Ty> class Curve>
350 struct CurveGeometryInterface : public CurveGeometry
351 {
352 typedef Curve<Vec3ff> Curve3ff;
353 typedef Curve<Vec3fa> Curve3fa;
354
355 CurveGeometryInterface (Device* device, Geometry::GType gtype)
356 : CurveGeometry(device,gtype) {}
357
358 __forceinline const Curve3ff getCurveScaledRadius(size_t i, size_t itime = 0) const
359 {
360 const unsigned int index = curve(i);
361 Vec3ff v0 = vertex(index+0,itime);
362 Vec3ff v1 = vertex(index+1,itime);
363 Vec3ff v2 = vertex(index+2,itime);
364 Vec3ff v3 = vertex(index+3,itime);
365 v0.w *= maxRadiusScale;
366 v1.w *= maxRadiusScale;
367 v2.w *= maxRadiusScale;
368 v3.w *= maxRadiusScale;
369 return Curve3ff (v0,v1,v2,v3);
370 }
371
372 __forceinline const Curve3ff getCurveScaledRadius(const LinearSpace3fa& space, size_t i, size_t itime = 0) const
373 {
374 const unsigned int index = curve(i);
375 const Vec3ff v0 = vertex(index+0,itime);
376 const Vec3ff v1 = vertex(index+1,itime);
377 const Vec3ff v2 = vertex(index+2,itime);
378 const Vec3ff v3 = vertex(index+3,itime);
379 const Vec3ff w0(xfmPoint(space,(Vec3fa)v0), maxRadiusScale*v0.w);
380 const Vec3ff w1(xfmPoint(space,(Vec3fa)v1), maxRadiusScale*v1.w);
381 const Vec3ff w2(xfmPoint(space,(Vec3fa)v2), maxRadiusScale*v2.w);
382 const Vec3ff w3(xfmPoint(space,(Vec3fa)v3), maxRadiusScale*v3.w);
383 return Curve3ff(w0,w1,w2,w3);
384 }
385
386 __forceinline const Curve3ff getCurveScaledRadius(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t i, size_t itime = 0) const
387 {
388 const float r_scale = r_scale0*scale;
389 const unsigned int index = curve(i);
390 const Vec3ff v0 = vertex(index+0,itime);
391 const Vec3ff v1 = vertex(index+1,itime);
392 const Vec3ff v2 = vertex(index+2,itime);
393 const Vec3ff v3 = vertex(index+3,itime);
394 const Vec3ff w0(xfmPoint(space,((Vec3fa)v0-ofs)*Vec3fa(scale)), maxRadiusScale*v0.w*r_scale);
395 const Vec3ff w1(xfmPoint(space,((Vec3fa)v1-ofs)*Vec3fa(scale)), maxRadiusScale*v1.w*r_scale);
396 const Vec3ff w2(xfmPoint(space,((Vec3fa)v2-ofs)*Vec3fa(scale)), maxRadiusScale*v2.w*r_scale);
397 const Vec3ff w3(xfmPoint(space,((Vec3fa)v3-ofs)*Vec3fa(scale)), maxRadiusScale*v3.w*r_scale);
398 return Curve3ff(w0,w1,w2,w3);
399 }
400
401 __forceinline const Curve3fa getNormalCurve(size_t i, size_t itime = 0) const
402 {
403 const unsigned int index = curve(i);
404 const Vec3fa n0 = normal(index+0,itime);
405 const Vec3fa n1 = normal(index+1,itime);
406 const Vec3fa n2 = normal(index+2,itime);
407 const Vec3fa n3 = normal(index+3,itime);
408 return Curve3fa (n0,n1,n2,n3);
409 }
410
411 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(size_t i, size_t itime = 0) const
412 {
413 const Curve3ff center = getCurveScaledRadius(i,itime);
414 const Curve3fa normal = getNormalCurve(i,itime);
415 const TensorLinearCubicBezierSurface3fa ocurve = TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(center,normal);
416 return ocurve;
417 }
418
419 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(const LinearSpace3fa& space, size_t i, size_t itime = 0) const {
420 return getOrientedCurveScaledRadius(i,itime).xfm(space);
421 }
422
423 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(const Vec3fa& ofs, const float scale, const LinearSpace3fa& space, size_t i, size_t itime = 0) const {
424 return getOrientedCurveScaledRadius(i,itime).xfm(space,ofs,scale);
425 }
426
427 /*! check if the i'th primitive is valid at the itime'th time step */
428 __forceinline bool valid(Geometry::GType ctype, size_t i, const range<size_t>& itime_range) const
429 {
430 const unsigned int index = curve(i);
431 if (index+3 >= numVertices()) return false;
432
433 for (size_t itime = itime_range.begin(); itime <= itime_range.end(); itime++)
434 {
435 const float r0 = radius(index+0,itime);
436 const float r1 = radius(index+1,itime);
437 const float r2 = radius(index+2,itime);
438 const float r3 = radius(index+3,itime);
439 if (!isvalid(r0) || !isvalid(r1) || !isvalid(r2) || !isvalid(r3))
440 return false;
441
442 const Vec3fa v0 = vertex(index+0,itime);
443 const Vec3fa v1 = vertex(index+1,itime);
444 const Vec3fa v2 = vertex(index+2,itime);
445 const Vec3fa v3 = vertex(index+3,itime);
446 if (!isvalid(v0) || !isvalid(v1) || !isvalid(v2) || !isvalid(v3))
447 return false;
448
449 if (ctype == Geometry::GTY_SUBTYPE_ORIENTED_CURVE)
450 {
451 const Vec3fa n0 = normal(index+0,itime);
452 const Vec3fa n1 = normal(index+1,itime);
453 if (!isvalid(n0) || !isvalid(n1))
454 return false;
455
456 const BBox3fa b = getOrientedCurveScaledRadius(i,itime).accurateBounds();
457 if (!isvalid(b))
458 return false;
459 }
460 }
461
462 return true;
463 }
464
465 template<int N>
466 void interpolate_impl(const RTCInterpolateArguments* const args)
467 {
468 unsigned int primID = args->primID;
469 float u = args->u;
470 RTCBufferType bufferType = args->bufferType;
471 unsigned int bufferSlot = args->bufferSlot;
472 float* P = args->P;
473 float* dPdu = args->dPdu;
474 float* ddPdudu = args->ddPdudu;
475 unsigned int valueCount = args->valueCount;
476
477 /* calculate base pointer and stride */
478 assert((bufferType == RTC_BUFFER_TYPE_VERTEX && bufferSlot < numTimeSteps) ||
479 (bufferType == RTC_BUFFER_TYPE_VERTEX_ATTRIBUTE && bufferSlot <= vertexAttribs.size()));
480 const char* src = nullptr;
481 size_t stride = 0;
482 if (bufferType == RTC_BUFFER_TYPE_VERTEX_ATTRIBUTE) {
483 src = vertexAttribs[bufferSlot].getPtr();
484 stride = vertexAttribs[bufferSlot].getStride();
485 } else {
486 src = vertices[bufferSlot].getPtr();
487 stride = vertices[bufferSlot].getStride();
488 }
489
490 for (unsigned int i=0; i<valueCount; i+=N)
491 {
492 size_t ofs = i*sizeof(float);
493 const size_t index = curves[primID];
494 const vbool<N> valid = vint<N>((int)i)+vint<N>(step) < vint<N>((int)valueCount);
495 const vfloat<N> p0 = mem<vfloat<N>>::loadu(valid,(float*)&src[(index+0)*stride+ofs]);
496 const vfloat<N> p1 = mem<vfloat<N>>::loadu(valid,(float*)&src[(index+1)*stride+ofs]);
497 const vfloat<N> p2 = mem<vfloat<N>>::loadu(valid,(float*)&src[(index+2)*stride+ofs]);
498 const vfloat<N> p3 = mem<vfloat<N>>::loadu(valid,(float*)&src[(index+3)*stride+ofs]);
499
500 const Curve<vfloat<N>> curve(p0,p1,p2,p3);
501 if (P ) mem<vfloat<N>>::storeu(valid,P+i, curve.eval(u));
502 if (dPdu ) mem<vfloat<N>>::storeu(valid,dPdu+i, curve.eval_du(u));
503 if (ddPdudu) mem<vfloat<N>>::storeu(valid,ddPdudu+i,curve.eval_dudu(u));
504 }
505 }
506
507 void interpolate(const RTCInterpolateArguments* const args) {
508 interpolate_impl<4>(args);
509 }
510 };
511
512 template<template<typename Ty> class Curve>
513 struct HermiteCurveGeometryInterface : public CurveGeometry
514 {
515 typedef Curve<Vec3ff> HermiteCurve3ff;
516 typedef Curve<Vec3fa> HermiteCurve3fa;
517
518 HermiteCurveGeometryInterface (Device* device, Geometry::GType gtype)
519 : CurveGeometry(device,gtype) {}
520
521 __forceinline const HermiteCurve3ff getCurveScaledRadius(size_t i, size_t itime = 0) const
522 {
523 const unsigned int index = curve(i);
524 Vec3ff v0 = vertex(index+0,itime);
525 Vec3ff v1 = vertex(index+1,itime);
526 Vec3ff t0 = tangent(index+0,itime);
527 Vec3ff t1 = tangent(index+1,itime);
528 v0.w *= maxRadiusScale;
529 v1.w *= maxRadiusScale;
530 t0.w *= maxRadiusScale;
531 t1.w *= maxRadiusScale;
532 return HermiteCurve3ff (v0,t0,v1,t1);
533 }
534
535 __forceinline const HermiteCurve3ff getCurveScaledRadius(const LinearSpace3fa& space, size_t i, size_t itime = 0) const
536 {
537 const unsigned int index = curve(i);
538 const Vec3ff v0 = vertex(index+0,itime);
539 const Vec3ff v1 = vertex(index+1,itime);
540 const Vec3ff t0 = tangent(index+0,itime);
541 const Vec3ff t1 = tangent(index+1,itime);
542 const Vec3ff V0(xfmPoint(space,(Vec3fa)v0),maxRadiusScale*v0.w);
543 const Vec3ff V1(xfmPoint(space,(Vec3fa)v1),maxRadiusScale*v1.w);
544 const Vec3ff T0(xfmVector(space,(Vec3fa)t0),maxRadiusScale*t0.w);
545 const Vec3ff T1(xfmVector(space,(Vec3fa)t1),maxRadiusScale*t1.w);
546 return HermiteCurve3ff(V0,T0,V1,T1);
547 }
548
549 __forceinline const HermiteCurve3ff getCurveScaledRadius(const Vec3fa& ofs, const float scale, const float r_scale0, const LinearSpace3fa& space, size_t i, size_t itime = 0) const
550 {
551 const float r_scale = r_scale0*scale;
552 const unsigned int index = curve(i);
553 const Vec3ff v0 = vertex(index+0,itime);
554 const Vec3ff v1 = vertex(index+1,itime);
555 const Vec3ff t0 = tangent(index+0,itime);
556 const Vec3ff t1 = tangent(index+1,itime);
557 const Vec3ff V0(xfmPoint(space,(v0-ofs)*Vec3fa(scale)), maxRadiusScale*v0.w*r_scale);
558 const Vec3ff V1(xfmPoint(space,(v1-ofs)*Vec3fa(scale)), maxRadiusScale*v1.w*r_scale);
559 const Vec3ff T0(xfmVector(space,t0*Vec3fa(scale)), maxRadiusScale*t0.w*r_scale);
560 const Vec3ff T1(xfmVector(space,t1*Vec3fa(scale)), maxRadiusScale*t1.w*r_scale);
561 return HermiteCurve3ff(V0,T0,V1,T1);
562 }
563
564 __forceinline const HermiteCurve3fa getNormalCurve(size_t i, size_t itime = 0) const
565 {
566 const unsigned int index = curve(i);
567 const Vec3fa n0 = normal(index+0,itime);
568 const Vec3fa n1 = normal(index+1,itime);
569 const Vec3fa dn0 = dnormal(index+0,itime);
570 const Vec3fa dn1 = dnormal(index+1,itime);
571 return HermiteCurve3fa (n0,dn0,n1,dn1);
572 }
573
574 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(size_t i, size_t itime = 0) const
575 {
576 const HermiteCurve3ff center = getCurveScaledRadius(i,itime);
577 const HermiteCurve3fa normal = getNormalCurve(i,itime);
578 const TensorLinearCubicBezierSurface3fa ocurve = TensorLinearCubicBezierSurface3fa::fromCenterAndNormalCurve(center,normal);
579 return ocurve;
580 }
581
582 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(const LinearSpace3fa& space, size_t i, size_t itime = 0) const {
583 return getOrientedCurveScaledRadius(i,itime).xfm(space);
584 }
585
586 __forceinline const TensorLinearCubicBezierSurface3fa getOrientedCurveScaledRadius(const Vec3fa& ofs, const float scale, const LinearSpace3fa& space, size_t i, size_t itime = 0) const {
587 return getOrientedCurveScaledRadius(i,itime).xfm(space,ofs,scale);
588 }
589
590 /*! check if the i'th primitive is valid at the itime'th time step */
591 __forceinline bool valid(Geometry::GType ctype, size_t i, const range<size_t>& itime_range) const
592 {
593 const unsigned int index = curve(i);
594 if (index+1 >= numVertices()) return false;
595
596 for (size_t itime = itime_range.begin(); itime <= itime_range.end(); itime++)
597 {
598 const Vec3ff v0 = vertex(index+0,itime);
599 const Vec3ff v1 = vertex(index+1,itime);
600 if (!isvalid4(v0) || !isvalid4(v1))
601 return false;
602
603 const Vec3ff t0 = tangent(index+0,itime);
604 const Vec3ff t1 = tangent(index+1,itime);
605 if (!isvalid4(t0) || !isvalid4(t1))
606 return false;
607
608 if (ctype == Geometry::GTY_SUBTYPE_ORIENTED_CURVE)
609 {
610 const Vec3fa n0 = normal(index+0,itime);
611 const Vec3fa n1 = normal(index+1,itime);
612 if (!isvalid(n0) || !isvalid(n1))
613 return false;
614
615 const Vec3fa dn0 = dnormal(index+0,itime);
616 const Vec3fa dn1 = dnormal(index+1,itime);
617 if (!isvalid(dn0) || !isvalid(dn1))
618 return false;
619
620 const BBox3fa b = getOrientedCurveScaledRadius(i,itime).accurateBounds();
621 if (!isvalid(b))
622 return false;
623 }
624 }
625
626 return true;
627 }
628
629 template<int N>
630 void interpolate_impl(const RTCInterpolateArguments* const args)
631 {
632 unsigned int primID = args->primID;
633 float u = args->u;
634 RTCBufferType bufferType = args->bufferType;
635 unsigned int bufferSlot = args->bufferSlot;
636 float* P = args->P;
637 float* dPdu = args->dPdu;
638 float* ddPdudu = args->ddPdudu;
639 unsigned int valueCount = args->valueCount;
640
641 /* we interpolate vertex attributes linearly for hermite basis */
642 if (bufferType == RTC_BUFFER_TYPE_VERTEX_ATTRIBUTE)
643 {
644 assert(bufferSlot <= vertexAttribs.size());
645 const char* vsrc = vertexAttribs[bufferSlot].getPtr();
646 const size_t vstride = vertexAttribs[bufferSlot].getStride();
647
648 for (unsigned int i=0; i<valueCount; i+=N)
649 {
650 const size_t ofs = i*sizeof(float);
651 const size_t index = curves[primID];
652 const vbool<N> valid = vint<N>((int)i)+vint<N>(step) < vint<N>((int)valueCount);
653 const vfloat<N> p0 = mem<vfloat<N>>::loadu(valid,(float*)&vsrc[(index+0)*vstride+ofs]);
654 const vfloat<N> p1 = mem<vfloat<N>>::loadu(valid,(float*)&vsrc[(index+1)*vstride+ofs]);
655
656 if (P ) mem<vfloat<N>>::storeu(valid,P+i, madd(1.0f-u,p0,u*p1));
657 if (dPdu ) mem<vfloat<N>>::storeu(valid,dPdu+i, p1-p0);
658 if (ddPdudu) mem<vfloat<N>>::storeu(valid,ddPdudu+i,vfloat<N>(zero));
659 }
660 }
661
662 /* interpolation for vertex buffers */
663 else
664 {
665 assert(bufferSlot < numTimeSteps);
666 const char* vsrc = vertices[bufferSlot].getPtr();
667 const char* tsrc = tangents[bufferSlot].getPtr();
668 const size_t vstride = vertices[bufferSlot].getStride();
669 const size_t tstride = vertices[bufferSlot].getStride();
670
671 for (unsigned int i=0; i<valueCount; i+=N)
672 {
673 const size_t ofs = i*sizeof(float);
674 const size_t index = curves[primID];
675 const vbool<N> valid = vint<N>((int)i)+vint<N>(step) < vint<N>((int)valueCount);
676 const vfloat<N> p0 = mem<vfloat<N>>::loadu(valid,(float*)&vsrc[(index+0)*vstride+ofs]);
677 const vfloat<N> p1 = mem<vfloat<N>>::loadu(valid,(float*)&vsrc[(index+1)*vstride+ofs]);
678 const vfloat<N> t0 = mem<vfloat<N>>::loadu(valid,(float*)&tsrc[(index+0)*tstride+ofs]);
679 const vfloat<N> t1 = mem<vfloat<N>>::loadu(valid,(float*)&tsrc[(index+1)*tstride+ofs]);
680
681 const HermiteCurveT<vfloat<N>> curve(p0,t0,p1,t1);
682 if (P ) mem<vfloat<N>>::storeu(valid,P+i, curve.eval(u));
683 if (dPdu ) mem<vfloat<N>>::storeu(valid,dPdu+i, curve.eval_du(u));
684 if (ddPdudu) mem<vfloat<N>>::storeu(valid,ddPdudu+i,curve.eval_dudu(u));
685 }
686 }
687 }
688
689 void interpolate(const RTCInterpolateArguments* const args) {
690 interpolate_impl<4>(args);
691 }
692 };
693 }
694
695 DECLARE_ISA_FUNCTION(CurveGeometry*, createCurves, Device* COMMA Geometry::GType);
696}
697