| 1 | // Copyright 2009-2021 Intel Corporation |
| 2 | // SPDX-License-Identifier: Apache-2.0 |
| 3 | |
| 4 | #include "bvh_statistics.h" |
| 5 | #include "../../common/algorithms/parallel_reduce.h" |
| 6 | |
| 7 | namespace embree |
| 8 | { |
| 9 | template<int N> |
| 10 | BVHNStatistics<N>::BVHNStatistics (BVH* bvh) : bvh(bvh) |
| 11 | { |
| 12 | double A = max(0.0f,bvh->getLinearBounds().expectedHalfArea()); |
| 13 | stat = statistics(bvh->root,A,BBox1f(0.0f,1.0f)); |
| 14 | } |
| 15 | |
| 16 | template<int N> |
| 17 | std::string BVHNStatistics<N>::str() |
| 18 | { |
| 19 | std::ostringstream stream; |
| 20 | stream.setf(std::ios::fixed, std::ios::floatfield); |
| 21 | stream << " primitives = " << bvh->numPrimitives << ", vertices = " << bvh->numVertices << ", depth = " << stat.depth << std::endl; |
| 22 | size_t totalBytes = stat.bytes(bvh); |
| 23 | double totalSAH = stat.sah(bvh); |
| 24 | stream << " total : sah = " << std::setw(7) << std::setprecision(3) << totalSAH << " (100.00%), " ; |
| 25 | stream << "#bytes = " << std::setw(7) << std::setprecision(2) << totalBytes/1E6 << " MB (100.00%), " ; |
| 26 | stream << "#nodes = " << std::setw(7) << stat.size() << " (" << std::setw(6) << std::setprecision(2) << 100.0*stat.fillRate(bvh) << "% filled), " ; |
| 27 | stream << "#bytes/prim = " << std::setw(6) << std::setprecision(2) << double(totalBytes)/double(bvh->numPrimitives) << std::endl; |
| 28 | if (stat.statAABBNodes.numNodes ) stream << " getAABBNodes : " << stat.statAABBNodes.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 29 | if (stat.statOBBNodes.numNodes ) stream << " ungetAABBNodes : " << stat.statOBBNodes.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 30 | if (stat.statAABBNodesMB.numNodes ) stream << " getAABBNodesMB : " << stat.statAABBNodesMB.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 31 | if (stat.statAABBNodesMB4D.numNodes) stream << " getAABBNodesMB4D : " << stat.statAABBNodesMB4D.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 32 | if (stat.statOBBNodesMB.numNodes) stream << " ungetAABBNodesMB : " << stat.statOBBNodesMB.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 33 | if (stat.statQuantizedNodes.numNodes ) stream << " quantizedNodes : " << stat.statQuantizedNodes.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 34 | if (true) stream << " leaves : " << stat.statLeaf.toString(bvh,totalSAH,totalBytes) << std::endl; |
| 35 | if (true) stream << " histogram : " << stat.statLeaf.histToString() << std::endl; |
| 36 | return stream.str(); |
| 37 | } |
| 38 | |
| 39 | template<int N> |
| 40 | typename BVHNStatistics<N>::Statistics BVHNStatistics<N>::statistics(NodeRef node, const double A, const BBox1f t0t1) |
| 41 | { |
| 42 | Statistics s; |
| 43 | assert(t0t1.size() > 0.0f); |
| 44 | double dt = max(0.0f,t0t1.size()); |
| 45 | if (node.isAABBNode()) |
| 46 | { |
| 47 | AABBNode* n = node.getAABBNode(); |
| 48 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 49 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 50 | const double Ai = max(0.0f,halfArea(n->extend(i))); |
| 51 | Statistics s = statistics(n->child(i),Ai,t0t1); |
| 52 | s.statAABBNodes.numChildren++; |
| 53 | return s; |
| 54 | }, Statistics::add); |
| 55 | s.statAABBNodes.numNodes++; |
| 56 | s.statAABBNodes.nodeSAH += dt*A; |
| 57 | s.depth++; |
| 58 | } |
| 59 | else if (node.isOBBNode()) |
| 60 | { |
| 61 | OBBNode* n = node.ungetAABBNode(); |
| 62 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 63 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 64 | const double Ai = max(0.0f,halfArea(n->extent(i))); |
| 65 | Statistics s = statistics(n->child(i),Ai,t0t1); |
| 66 | s.statOBBNodes.numChildren++; |
| 67 | return s; |
| 68 | }, Statistics::add); |
| 69 | s.statOBBNodes.numNodes++; |
| 70 | s.statOBBNodes.nodeSAH += dt*A; |
| 71 | s.depth++; |
| 72 | } |
| 73 | else if (node.isAABBNodeMB()) |
| 74 | { |
| 75 | AABBNodeMB* n = node.getAABBNodeMB(); |
| 76 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 77 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 78 | const double Ai = max(0.0f,n->expectedHalfArea(i,t0t1)); |
| 79 | Statistics s = statistics(n->child(i),Ai,t0t1); |
| 80 | s.statAABBNodesMB.numChildren++; |
| 81 | return s; |
| 82 | }, Statistics::add); |
| 83 | s.statAABBNodesMB.numNodes++; |
| 84 | s.statAABBNodesMB.nodeSAH += dt*A; |
| 85 | s.depth++; |
| 86 | } |
| 87 | else if (node.isAABBNodeMB4D()) |
| 88 | { |
| 89 | AABBNodeMB4D* n = node.getAABBNodeMB4D(); |
| 90 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 91 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 92 | const BBox1f t0t1i = intersect(t0t1,n->timeRange(i)); |
| 93 | assert(!t0t1i.empty()); |
| 94 | const double Ai = n->AABBNodeMB::expectedHalfArea(i,t0t1i); |
| 95 | Statistics s = statistics(n->child(i),Ai,t0t1i); |
| 96 | s.statAABBNodesMB4D.numChildren++; |
| 97 | return s; |
| 98 | }, Statistics::add); |
| 99 | s.statAABBNodesMB4D.numNodes++; |
| 100 | s.statAABBNodesMB4D.nodeSAH += dt*A; |
| 101 | s.depth++; |
| 102 | } |
| 103 | else if (node.isOBBNodeMB()) |
| 104 | { |
| 105 | OBBNodeMB* n = node.ungetAABBNodeMB(); |
| 106 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 107 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 108 | const double Ai = max(0.0f,halfArea(n->extent0(i))); |
| 109 | Statistics s = statistics(n->child(i),Ai,t0t1); |
| 110 | s.statOBBNodesMB.numChildren++; |
| 111 | return s; |
| 112 | }, Statistics::add); |
| 113 | s.statOBBNodesMB.numNodes++; |
| 114 | s.statOBBNodesMB.nodeSAH += dt*A; |
| 115 | s.depth++; |
| 116 | } |
| 117 | else if (node.isQuantizedNode()) |
| 118 | { |
| 119 | QuantizedNode* n = node.quantizedNode(); |
| 120 | s = s + parallel_reduce(0,N,Statistics(),[&] ( const int i ) { |
| 121 | if (n->child(i) == BVH::emptyNode) return Statistics(); |
| 122 | const double Ai = max(0.0f,halfArea(n->extent(i))); |
| 123 | Statistics s = statistics(n->child(i),Ai,t0t1); |
| 124 | s.statQuantizedNodes.numChildren++; |
| 125 | return s; |
| 126 | }, Statistics::add); |
| 127 | s.statQuantizedNodes.numNodes++; |
| 128 | s.statQuantizedNodes.nodeSAH += dt*A; |
| 129 | s.depth++; |
| 130 | } |
| 131 | else if (node.isLeaf()) |
| 132 | { |
| 133 | size_t num; const char* tri = node.leaf(num); |
| 134 | if (num) |
| 135 | { |
| 136 | for (size_t i=0; i<num; i++) |
| 137 | { |
| 138 | const size_t bytes = bvh->primTy->getBytes(tri); |
| 139 | s.statLeaf.numPrimsActive += bvh->primTy->sizeActive(tri); |
| 140 | s.statLeaf.numPrimsTotal += bvh->primTy->sizeTotal(tri); |
| 141 | s.statLeaf.numBytes += bytes; |
| 142 | tri+=bytes; |
| 143 | } |
| 144 | s.statLeaf.numLeaves++; |
| 145 | s.statLeaf.numPrimBlocks += num; |
| 146 | s.statLeaf.leafSAH += dt*A*num; |
| 147 | if (num-1 < Statistics::LeafStat::NHIST) { |
| 148 | s.statLeaf.numPrimBlocksHistogram[num-1]++; |
| 149 | } |
| 150 | } |
| 151 | } |
| 152 | else { |
| 153 | // -- GODOT start -- |
| 154 | // throw std::runtime_error("not supported node type in bvh_statistics"); |
| 155 | abort(); |
| 156 | // -- GODOT end -- |
| 157 | } |
| 158 | return s; |
| 159 | } |
| 160 | |
| 161 | #if defined(__AVX__) |
| 162 | template class BVHNStatistics<8>; |
| 163 | #endif |
| 164 | |
| 165 | #if !defined(__AVX__) || (!defined(EMBREE_TARGET_SSE2) && !defined(EMBREE_TARGET_SSE42)) || defined(__aarch64__) |
| 166 | template class BVHNStatistics<4>; |
| 167 | #endif |
| 168 | } |
| 169 | |