Optimize octree generation and traversal

include/nori/octree.h

@@ -17,7 +17,7 @@ public:
     explicit Octree(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> triangles) : bbox(bbox), mesh(mesh), triangles(triangles) {
         children = std::vector<Octree*>(8, nullptr);
     }
-    static Octree *build(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> triangles, int recursionDepth = 0);
+    static Octree *build(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> *triangles, int recursionDepth = 0);
 
     uint32_t getIntersectingTriangle(Ray3f &ray, Intersection &its, bool shadowRay);
     BoundingBox3f getBoundingBox() const { return *bbox; }

src/accel.cpp

@@ -30,9 +30,9 @@ void Accel::addMesh(Mesh *mesh) {
 }
 
 void Accel::build() {
-    auto triangles = std::vector<uint32_t>(m_mesh->getTriangleCount());
+    auto triangles = new std::vector<uint32_t>(m_mesh->getTriangleCount());
     for (uint32_t idx = 0; idx < m_mesh->getTriangleCount(); ++idx) {
-        triangles[idx] = idx;
+        (*triangles)[idx] = idx;
     }
 
     cout << "Building octree" << endl;
@@ -40,7 +40,7 @@ void Accel::build() {
     m_octree = Octree::build(&m_bbox, m_mesh, triangles);
     auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::high_resolution_clock::now() - start);
 
-    std::cout << "Octree build took " << duration.count()/1000.0f << " s" << std::endl;
+    std::cout << "Octree build took " << duration.count() << "ms" << std::endl;
 }
 
 bool Accel::rayIntersect(const Ray3f &ray_, Intersection &its, bool shadowRay) const {

src/octree.cpp

@@ -5,12 +5,9 @@
 NORI_NAMESPACE_BEGIN
 
 
-Octree *Octree::build(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> triangles, int recursionDepth) {
+Octree *Octree::build(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> *triangles, int recursionDepth) {
-    if (triangles.empty()) {
+    if (triangles->size() <= 10 || recursionDepth > 10) {
-        return nullptr;
+        return new Octree(bbox, mesh, *triangles);
-    }
-    if (triangles.size() <= 20 || recursionDepth > 10 || bbox->getVolume() < 0.001f) {
-        return new Octree(bbox, mesh, triangles);
     }
 
     auto m_F = mesh->getIndices();
@@ -33,31 +30,43 @@ Octree *Octree::build(BoundingBox3f *bbox, Mesh *mesh, std::vector<uint32_t> tri
         new BoundingBox3f(bMin + Vector3f(w2, h2, d2), bMin + Vector3f(w,  h,   d))
     };
 
-    std::vector<uint32_t> newTriangles[8];
+    std::vector<uint32_t> *newTriangles[8];
     BoundingBox3f triangleBbox;
     for(int i = 0; i < 8; i++) {
-        newTriangles[i].reserve(triangles.size() / 8);
+        newTriangles[i] = new std::vector<uint32_t>();
-        for (uint32_t idx : triangles) {
+        newTriangles[i]->reserve(triangles->size() / 8);
+        for (uint32_t idx : *triangles) {
             auto a = m_V.col(m_F(0, idx));
             auto b = m_V.col(m_F(1, idx));
             auto c = m_V.col(m_F(2, idx));
             triangleBbox.min = a.cwiseMin(b).cwiseMin(c);
             triangleBbox.max = a.cwiseMax(b).cwiseMax(c);
             if (newBboxes[i]->overlaps(triangleBbox)) {
-                newTriangles[i].push_back(idx);
+                newTriangles[i]->push_back(idx);
             }
         }
-        newTriangles[i].shrink_to_fit();
+        newTriangles[i]->shrink_to_fit();
     }
 
     if (recursionDepth < 2) {
         tbb::parallel_for(tbb::blocked_range<size_t>(0, 8), [&](const tbb::blocked_range<size_t> &range) {
             for (size_t i = range.begin(); i != range.end(); ++i) {
+                if (newTriangles[i]->empty()) {
+                    node->children[i] = nullptr;
+                    continue;
+                }
                 node->children[i] = build(newBboxes[i], mesh, newTriangles[i], recursionDepth + 1);
             }
         });
     } else {
         for (int i = 0; i < 8; i++) {
+            if (newTriangles[i]->empty()) {
+                node->children[i] = nullptr;
+                continue;
+            }
+            if (newTriangles[i]->size() > 10) {
+                node->children[i] = build(newBboxes[i], mesh, newTriangles[i], recursionDepth + 1);
+            }
             node->children[i] = build(newBboxes[i], mesh, newTriangles[i], recursionDepth + 1);
         }
     }
@@ -72,15 +81,15 @@ uint32_t Octree::getIntersectingTriangle(Ray3f &ray, Intersection &its, bool sha
     uint32_t triangle;
 
     if (triangles.empty()) {
-        std::sort(children.begin(), children.end(), [ray](Octree *a, Octree *b) {
+        // std::sort(children.begin(), children.end(), [ray](Octree *a, Octree *b) {
-            if (a == nullptr) {
+        //     if (a == nullptr) {
-                return false;
+        //         return false;
-            }
+        //     }
-            if (b == nullptr) {
+        //     if (b == nullptr) {
-                return true;
+        //         return true;
-            }
+        //     }
-            return a->bbox->distanceTo(ray.o) < b->bbox->distanceTo(ray.o);
+        //     return a->bbox->distanceTo(ray.o) < b->bbox->distanceTo(ray.o);
-        });
+        // });
         for (auto child : children) {
             if (child == nullptr || !child->bbox->rayIntersect(ray)) {
                 continue;
@@ -90,8 +99,7 @@ uint32_t Octree::getIntersectingTriangle(Ray3f &ray, Intersection &its, bool sha
                 return triangle;
             }
         }
-
+        return -1;
-        return closestTriangle;
     }
 
     for (auto triangle : triangles) {