1 /*
2 Bullet Continuous Collision Detection and Physics Library
3 Copyright (c) 2003-2006 Erwin Coumans http://continuousphysics.com/Bullet/
4
5 This software is provided 'as-is', without any express or implied warranty.
6 In no event will the authors be held liable for any damages arising from the use of this software.
7 Permission is granted to anyone to use this software for any purpose,
8 including commercial applications, and to alter it and redistribute it freely,
9 subject to the following restrictions:
10
11 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
12 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
13 3. This notice may not be removed or altered from any source distribution.
14 */
15
16 #ifndef BT_QUANTIZED_BVH_H
17 #define BT_QUANTIZED_BVH_H
18
19 class btSerializer;
20
21 //#define DEBUG_CHECK_DEQUANTIZATION 1
22 #ifdef DEBUG_CHECK_DEQUANTIZATION
23 #ifdef __SPU__
24 #define printf spu_printf
25 #endif //__SPU__
26
27 #include <stdio.h>
28 #include <stdlib.h>
29 #endif //DEBUG_CHECK_DEQUANTIZATION
30
31 #include "LinearMath/btVector3.h"
32 #include "LinearMath/btAlignedAllocator.h"
33
34 #ifdef BT_USE_DOUBLE_PRECISION
35 #define btQuantizedBvhData btQuantizedBvhDoubleData
36 #define btOptimizedBvhNodeData btOptimizedBvhNodeDoubleData
37 #define btQuantizedBvhDataName "btQuantizedBvhDoubleData"
38 #else
39 #define btQuantizedBvhData btQuantizedBvhFloatData
40 #define btOptimizedBvhNodeData btOptimizedBvhNodeFloatData
41 #define btQuantizedBvhDataName "btQuantizedBvhFloatData"
42 #endif
43
44 //http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vclang/html/vclrf__m128.asp
45
46 //Note: currently we have 16 bytes per quantized node
47 #define MAX_SUBTREE_SIZE_IN_BYTES 2048
48
49 // 10 gives the potential for 1024 parts, with at most 2^21 (2097152) (minus one
50 // actually) triangles each (since the sign bit is reserved
51 #define MAX_NUM_PARTS_IN_BITS 10
52
53 ///btQuantizedBvhNode is a compressed aabb node, 16 bytes.
54 ///Node can be used for leafnode or internal node. Leafnodes can point to 32-bit triangle index (non-negative range).
ATTRIBUTE_ALIGNED16(struct)55 ATTRIBUTE_ALIGNED16(struct)
56 btQuantizedBvhNode
57 {
58 BT_DECLARE_ALIGNED_ALLOCATOR();
59
60 //12 bytes
61 unsigned short int m_quantizedAabbMin[3];
62 unsigned short int m_quantizedAabbMax[3];
63 //4 bytes
64 int m_escapeIndexOrTriangleIndex;
65
66 bool isLeafNode() const
67 {
68 //skipindex is negative (internal node), triangleindex >=0 (leafnode)
69 return (m_escapeIndexOrTriangleIndex >= 0);
70 }
71 int getEscapeIndex() const
72 {
73 btAssert(!isLeafNode());
74 return -m_escapeIndexOrTriangleIndex;
75 }
76 int getTriangleIndex() const
77 {
78 btAssert(isLeafNode());
79 unsigned int x = 0;
80 unsigned int y = (~(x & 0)) << (31 - MAX_NUM_PARTS_IN_BITS);
81 // Get only the lower bits where the triangle index is stored
82 return (m_escapeIndexOrTriangleIndex & ~(y));
83 }
84 int getPartId() const
85 {
86 btAssert(isLeafNode());
87 // Get only the highest bits where the part index is stored
88 return (m_escapeIndexOrTriangleIndex >> (31 - MAX_NUM_PARTS_IN_BITS));
89 }
90 };
91
92 /// btOptimizedBvhNode contains both internal and leaf node information.
93 /// Total node size is 44 bytes / node. You can use the compressed version of 16 bytes.
ATTRIBUTE_ALIGNED16(struct)94 ATTRIBUTE_ALIGNED16(struct)
95 btOptimizedBvhNode
96 {
97 BT_DECLARE_ALIGNED_ALLOCATOR();
98
99 //32 bytes
100 btVector3 m_aabbMinOrg;
101 btVector3 m_aabbMaxOrg;
102
103 //4
104 int m_escapeIndex;
105
106 //8
107 //for child nodes
108 int m_subPart;
109 int m_triangleIndex;
110
111 //pad the size to 64 bytes
112 char m_padding[20];
113 };
114
115 ///btBvhSubtreeInfo provides info to gather a subtree of limited size
ATTRIBUTE_ALIGNED16(class)116 ATTRIBUTE_ALIGNED16(class)
117 btBvhSubtreeInfo
118 {
119 public:
120 BT_DECLARE_ALIGNED_ALLOCATOR();
121
122 //12 bytes
123 unsigned short int m_quantizedAabbMin[3];
124 unsigned short int m_quantizedAabbMax[3];
125 //4 bytes, points to the root of the subtree
126 int m_rootNodeIndex;
127 //4 bytes
128 int m_subtreeSize;
129 int m_padding[3];
130
131 btBvhSubtreeInfo()
132 {
133 //memset(&m_padding[0], 0, sizeof(m_padding));
134 }
135
136 void setAabbFromQuantizeNode(const btQuantizedBvhNode& quantizedNode)
137 {
138 m_quantizedAabbMin[0] = quantizedNode.m_quantizedAabbMin[0];
139 m_quantizedAabbMin[1] = quantizedNode.m_quantizedAabbMin[1];
140 m_quantizedAabbMin[2] = quantizedNode.m_quantizedAabbMin[2];
141 m_quantizedAabbMax[0] = quantizedNode.m_quantizedAabbMax[0];
142 m_quantizedAabbMax[1] = quantizedNode.m_quantizedAabbMax[1];
143 m_quantizedAabbMax[2] = quantizedNode.m_quantizedAabbMax[2];
144 }
145 };
146
147 class btNodeOverlapCallback
148 {
149 public:
~btNodeOverlapCallback()150 virtual ~btNodeOverlapCallback(){};
151
152 virtual void processNode(int subPart, int triangleIndex) = 0;
153 };
154
155 #include "LinearMath/btAlignedAllocator.h"
156 #include "LinearMath/btAlignedObjectArray.h"
157
158 ///for code readability:
159 typedef btAlignedObjectArray<btOptimizedBvhNode> NodeArray;
160 typedef btAlignedObjectArray<btQuantizedBvhNode> QuantizedNodeArray;
161 typedef btAlignedObjectArray<btBvhSubtreeInfo> BvhSubtreeInfoArray;
162
163 ///The btQuantizedBvh class stores an AABB tree that can be quickly traversed on CPU and Cell SPU.
164 ///It is used by the btBvhTriangleMeshShape as midphase.
165 ///It is recommended to use quantization for better performance and lower memory requirements.
ATTRIBUTE_ALIGNED16(class)166 ATTRIBUTE_ALIGNED16(class)
167 btQuantizedBvh
168 {
169 public:
170 enum btTraversalMode
171 {
172 TRAVERSAL_STACKLESS = 0,
173 TRAVERSAL_STACKLESS_CACHE_FRIENDLY,
174 TRAVERSAL_RECURSIVE
175 };
176
177 protected:
178 btVector3 m_bvhAabbMin;
179 btVector3 m_bvhAabbMax;
180 btVector3 m_bvhQuantization;
181
182 int m_bulletVersion; //for serialization versioning. It could also be used to detect endianess.
183
184 int m_curNodeIndex;
185 //quantization data
186 bool m_useQuantization;
187
188 NodeArray m_leafNodes;
189 NodeArray m_contiguousNodes;
190 QuantizedNodeArray m_quantizedLeafNodes;
191 QuantizedNodeArray m_quantizedContiguousNodes;
192
193 btTraversalMode m_traversalMode;
194 BvhSubtreeInfoArray m_SubtreeHeaders;
195
196 //This is only used for serialization so we don't have to add serialization directly to btAlignedObjectArray
197 mutable int m_subtreeHeaderCount;
198
199 ///two versions, one for quantized and normal nodes. This allows code-reuse while maintaining readability (no template/macro!)
200 ///this might be refactored into a virtual, it is usually not calculated at run-time
201 void setInternalNodeAabbMin(int nodeIndex, const btVector3& aabbMin)
202 {
203 if (m_useQuantization)
204 {
205 quantize(&m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[0], aabbMin, 0);
206 }
207 else
208 {
209 m_contiguousNodes[nodeIndex].m_aabbMinOrg = aabbMin;
210 }
211 }
212 void setInternalNodeAabbMax(int nodeIndex, const btVector3& aabbMax)
213 {
214 if (m_useQuantization)
215 {
216 quantize(&m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[0], aabbMax, 1);
217 }
218 else
219 {
220 m_contiguousNodes[nodeIndex].m_aabbMaxOrg = aabbMax;
221 }
222 }
223
224 btVector3 getAabbMin(int nodeIndex) const
225 {
226 if (m_useQuantization)
227 {
228 return unQuantize(&m_quantizedLeafNodes[nodeIndex].m_quantizedAabbMin[0]);
229 }
230 //non-quantized
231 return m_leafNodes[nodeIndex].m_aabbMinOrg;
232 }
233 btVector3 getAabbMax(int nodeIndex) const
234 {
235 if (m_useQuantization)
236 {
237 return unQuantize(&m_quantizedLeafNodes[nodeIndex].m_quantizedAabbMax[0]);
238 }
239 //non-quantized
240 return m_leafNodes[nodeIndex].m_aabbMaxOrg;
241 }
242
243 void setInternalNodeEscapeIndex(int nodeIndex, int escapeIndex)
244 {
245 if (m_useQuantization)
246 {
247 m_quantizedContiguousNodes[nodeIndex].m_escapeIndexOrTriangleIndex = -escapeIndex;
248 }
249 else
250 {
251 m_contiguousNodes[nodeIndex].m_escapeIndex = escapeIndex;
252 }
253 }
254
255 void mergeInternalNodeAabb(int nodeIndex, const btVector3& newAabbMin, const btVector3& newAabbMax)
256 {
257 if (m_useQuantization)
258 {
259 unsigned short int quantizedAabbMin[3];
260 unsigned short int quantizedAabbMax[3];
261 quantize(quantizedAabbMin, newAabbMin, 0);
262 quantize(quantizedAabbMax, newAabbMax, 1);
263 for (int i = 0; i < 3; i++)
264 {
265 if (m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[i] > quantizedAabbMin[i])
266 m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMin[i] = quantizedAabbMin[i];
267
268 if (m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[i] < quantizedAabbMax[i])
269 m_quantizedContiguousNodes[nodeIndex].m_quantizedAabbMax[i] = quantizedAabbMax[i];
270 }
271 }
272 else
273 {
274 //non-quantized
275 m_contiguousNodes[nodeIndex].m_aabbMinOrg.setMin(newAabbMin);
276 m_contiguousNodes[nodeIndex].m_aabbMaxOrg.setMax(newAabbMax);
277 }
278 }
279
280 void swapLeafNodes(int firstIndex, int secondIndex);
281
282 void assignInternalNodeFromLeafNode(int internalNode, int leafNodeIndex);
283
284 protected:
285 void buildTree(int startIndex, int endIndex);
286
287 int calcSplittingAxis(int startIndex, int endIndex);
288
289 int sortAndCalcSplittingIndex(int startIndex, int endIndex, int splitAxis);
290
291 void walkStacklessTree(btNodeOverlapCallback * nodeCallback, const btVector3& aabbMin, const btVector3& aabbMax) const;
292
293 void walkStacklessQuantizedTreeAgainstRay(btNodeOverlapCallback * nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin, const btVector3& aabbMax, int startNodeIndex, int endNodeIndex) const;
294 void walkStacklessQuantizedTree(btNodeOverlapCallback * nodeCallback, unsigned short int* quantizedQueryAabbMin, unsigned short int* quantizedQueryAabbMax, int startNodeIndex, int endNodeIndex) const;
295 void walkStacklessTreeAgainstRay(btNodeOverlapCallback * nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin, const btVector3& aabbMax, int startNodeIndex, int endNodeIndex) const;
296
297 ///tree traversal designed for small-memory processors like PS3 SPU
298 void walkStacklessQuantizedTreeCacheFriendly(btNodeOverlapCallback * nodeCallback, unsigned short int* quantizedQueryAabbMin, unsigned short int* quantizedQueryAabbMax) const;
299
300 ///use the 16-byte stackless 'skipindex' node tree to do a recursive traversal
301 void walkRecursiveQuantizedTreeAgainstQueryAabb(const btQuantizedBvhNode* currentNode, btNodeOverlapCallback* nodeCallback, unsigned short int* quantizedQueryAabbMin, unsigned short int* quantizedQueryAabbMax) const;
302
303 ///use the 16-byte stackless 'skipindex' node tree to do a recursive traversal
304 void walkRecursiveQuantizedTreeAgainstQuantizedTree(const btQuantizedBvhNode* treeNodeA, const btQuantizedBvhNode* treeNodeB, btNodeOverlapCallback* nodeCallback) const;
305
306 void updateSubtreeHeaders(int leftChildNodexIndex, int rightChildNodexIndex);
307
308 public:
309 BT_DECLARE_ALIGNED_ALLOCATOR();
310
311 btQuantizedBvh();
312
313 virtual ~btQuantizedBvh();
314
315 ///***************************************** expert/internal use only *************************
316 void setQuantizationValues(const btVector3& bvhAabbMin, const btVector3& bvhAabbMax, btScalar quantizationMargin = btScalar(1.0));
317 QuantizedNodeArray& getLeafNodeArray() { return m_quantizedLeafNodes; }
318 ///buildInternal is expert use only: assumes that setQuantizationValues and LeafNodeArray are initialized
319 void buildInternal();
320 ///***************************************** expert/internal use only *************************
321
322 void reportAabbOverlappingNodex(btNodeOverlapCallback * nodeCallback, const btVector3& aabbMin, const btVector3& aabbMax) const;
323 void reportRayOverlappingNodex(btNodeOverlapCallback * nodeCallback, const btVector3& raySource, const btVector3& rayTarget) const;
324 void reportBoxCastOverlappingNodex(btNodeOverlapCallback * nodeCallback, const btVector3& raySource, const btVector3& rayTarget, const btVector3& aabbMin, const btVector3& aabbMax) const;
325
326 SIMD_FORCE_INLINE void quantize(unsigned short* out, const btVector3& point, int isMax) const
327 {
328 btAssert(m_useQuantization);
329
330 btAssert(point.getX() <= m_bvhAabbMax.getX());
331 btAssert(point.getY() <= m_bvhAabbMax.getY());
332 btAssert(point.getZ() <= m_bvhAabbMax.getZ());
333
334 btAssert(point.getX() >= m_bvhAabbMin.getX());
335 btAssert(point.getY() >= m_bvhAabbMin.getY());
336 btAssert(point.getZ() >= m_bvhAabbMin.getZ());
337
338 btVector3 v = (point - m_bvhAabbMin) * m_bvhQuantization;
339 ///Make sure rounding is done in a way that unQuantize(quantizeWithClamp(...)) is conservative
340 ///end-points always set the first bit, so that they are sorted properly (so that neighbouring AABBs overlap properly)
341 ///@todo: double-check this
342 if (isMax)
343 {
344 out[0] = (unsigned short)(((unsigned short)(v.getX() + btScalar(1.)) | 1));
345 out[1] = (unsigned short)(((unsigned short)(v.getY() + btScalar(1.)) | 1));
346 out[2] = (unsigned short)(((unsigned short)(v.getZ() + btScalar(1.)) | 1));
347 }
348 else
349 {
350 out[0] = (unsigned short)(((unsigned short)(v.getX()) & 0xfffe));
351 out[1] = (unsigned short)(((unsigned short)(v.getY()) & 0xfffe));
352 out[2] = (unsigned short)(((unsigned short)(v.getZ()) & 0xfffe));
353 }
354
355 #ifdef DEBUG_CHECK_DEQUANTIZATION
356 btVector3 newPoint = unQuantize(out);
357 if (isMax)
358 {
359 if (newPoint.getX() < point.getX())
360 {
361 printf("unconservative X, diffX = %f, oldX=%f,newX=%f\n", newPoint.getX() - point.getX(), newPoint.getX(), point.getX());
362 }
363 if (newPoint.getY() < point.getY())
364 {
365 printf("unconservative Y, diffY = %f, oldY=%f,newY=%f\n", newPoint.getY() - point.getY(), newPoint.getY(), point.getY());
366 }
367 if (newPoint.getZ() < point.getZ())
368 {
369 printf("unconservative Z, diffZ = %f, oldZ=%f,newZ=%f\n", newPoint.getZ() - point.getZ(), newPoint.getZ(), point.getZ());
370 }
371 }
372 else
373 {
374 if (newPoint.getX() > point.getX())
375 {
376 printf("unconservative X, diffX = %f, oldX=%f,newX=%f\n", newPoint.getX() - point.getX(), newPoint.getX(), point.getX());
377 }
378 if (newPoint.getY() > point.getY())
379 {
380 printf("unconservative Y, diffY = %f, oldY=%f,newY=%f\n", newPoint.getY() - point.getY(), newPoint.getY(), point.getY());
381 }
382 if (newPoint.getZ() > point.getZ())
383 {
384 printf("unconservative Z, diffZ = %f, oldZ=%f,newZ=%f\n", newPoint.getZ() - point.getZ(), newPoint.getZ(), point.getZ());
385 }
386 }
387 #endif //DEBUG_CHECK_DEQUANTIZATION
388 }
389
390 SIMD_FORCE_INLINE void quantizeWithClamp(unsigned short* out, const btVector3& point2, int isMax) const
391 {
392 btAssert(m_useQuantization);
393
394 btVector3 clampedPoint(point2);
395 clampedPoint.setMax(m_bvhAabbMin);
396 clampedPoint.setMin(m_bvhAabbMax);
397
398 quantize(out, clampedPoint, isMax);
399 }
400
401 SIMD_FORCE_INLINE btVector3 unQuantize(const unsigned short* vecIn) const
402 {
403 btVector3 vecOut;
404 vecOut.setValue(
405 (btScalar)(vecIn[0]) / (m_bvhQuantization.getX()),
406 (btScalar)(vecIn[1]) / (m_bvhQuantization.getY()),
407 (btScalar)(vecIn[2]) / (m_bvhQuantization.getZ()));
408 vecOut += m_bvhAabbMin;
409 return vecOut;
410 }
411
412 ///setTraversalMode let's you choose between stackless, recursive or stackless cache friendly tree traversal. Note this is only implemented for quantized trees.
413 void setTraversalMode(btTraversalMode traversalMode)
414 {
415 m_traversalMode = traversalMode;
416 }
417
418 SIMD_FORCE_INLINE QuantizedNodeArray& getQuantizedNodeArray()
419 {
420 return m_quantizedContiguousNodes;
421 }
422
423 SIMD_FORCE_INLINE BvhSubtreeInfoArray& getSubtreeInfoArray()
424 {
425 return m_SubtreeHeaders;
426 }
427
428 ////////////////////////////////////////////////////////////////////
429
430 /////Calculate space needed to store BVH for serialization
431 unsigned calculateSerializeBufferSize() const;
432
433 /// Data buffer MUST be 16 byte aligned
434 virtual bool serialize(void* o_alignedDataBuffer, unsigned i_dataBufferSize, bool i_swapEndian) const;
435
436 ///deSerializeInPlace loads and initializes a BVH from a buffer in memory 'in place'
437 static btQuantizedBvh* deSerializeInPlace(void* i_alignedDataBuffer, unsigned int i_dataBufferSize, bool i_swapEndian);
438
439 static unsigned int getAlignmentSerializationPadding();
440 //////////////////////////////////////////////////////////////////////
441
442 virtual int calculateSerializeBufferSizeNew() const;
443
444 ///fills the dataBuffer and returns the struct name (and 0 on failure)
445 virtual const char* serialize(void* dataBuffer, btSerializer* serializer) const;
446
447 virtual void deSerializeFloat(struct btQuantizedBvhFloatData & quantizedBvhFloatData);
448
449 virtual void deSerializeDouble(struct btQuantizedBvhDoubleData & quantizedBvhDoubleData);
450
451 ////////////////////////////////////////////////////////////////////
452
453 SIMD_FORCE_INLINE bool isQuantized()
454 {
455 return m_useQuantization;
456 }
457
458 private:
459 // Special "copy" constructor that allows for in-place deserialization
460 // Prevents btVector3's default constructor from being called, but doesn't inialize much else
461 // ownsMemory should most likely be false if deserializing, and if you are not, don't call this (it also changes the function signature, which we need)
462 btQuantizedBvh(btQuantizedBvh & other, bool ownsMemory);
463 };
464
465 // clang-format off
466 // parser needs * with the name
467 struct btBvhSubtreeInfoData
468 {
469 int m_rootNodeIndex;
470 int m_subtreeSize;
471 unsigned short m_quantizedAabbMin[3];
472 unsigned short m_quantizedAabbMax[3];
473 };
474
475 struct btOptimizedBvhNodeFloatData
476 {
477 btVector3FloatData m_aabbMinOrg;
478 btVector3FloatData m_aabbMaxOrg;
479 int m_escapeIndex;
480 int m_subPart;
481 int m_triangleIndex;
482 char m_pad[4];
483 };
484
485 struct btOptimizedBvhNodeDoubleData
486 {
487 btVector3DoubleData m_aabbMinOrg;
488 btVector3DoubleData m_aabbMaxOrg;
489 int m_escapeIndex;
490 int m_subPart;
491 int m_triangleIndex;
492 char m_pad[4];
493 };
494
495
496 struct btQuantizedBvhNodeData
497 {
498 unsigned short m_quantizedAabbMin[3];
499 unsigned short m_quantizedAabbMax[3];
500 int m_escapeIndexOrTriangleIndex;
501 };
502
503 struct btQuantizedBvhFloatData
504 {
505 btVector3FloatData m_bvhAabbMin;
506 btVector3FloatData m_bvhAabbMax;
507 btVector3FloatData m_bvhQuantization;
508 int m_curNodeIndex;
509 int m_useQuantization;
510 int m_numContiguousLeafNodes;
511 int m_numQuantizedContiguousNodes;
512 btOptimizedBvhNodeFloatData *m_contiguousNodesPtr;
513 btQuantizedBvhNodeData *m_quantizedContiguousNodesPtr;
514 btBvhSubtreeInfoData *m_subTreeInfoPtr;
515 int m_traversalMode;
516 int m_numSubtreeHeaders;
517
518 };
519
520 struct btQuantizedBvhDoubleData
521 {
522 btVector3DoubleData m_bvhAabbMin;
523 btVector3DoubleData m_bvhAabbMax;
524 btVector3DoubleData m_bvhQuantization;
525 int m_curNodeIndex;
526 int m_useQuantization;
527 int m_numContiguousLeafNodes;
528 int m_numQuantizedContiguousNodes;
529 btOptimizedBvhNodeDoubleData *m_contiguousNodesPtr;
530 btQuantizedBvhNodeData *m_quantizedContiguousNodesPtr;
531
532 int m_traversalMode;
533 int m_numSubtreeHeaders;
534 btBvhSubtreeInfoData *m_subTreeInfoPtr;
535 };
536 // clang-format on
537
calculateSerializeBufferSizeNew()538 SIMD_FORCE_INLINE int btQuantizedBvh::calculateSerializeBufferSizeNew() const
539 {
540 return sizeof(btQuantizedBvhData);
541 }
542
543 #endif //BT_QUANTIZED_BVH_H
544