/* prints the bvh4.triangle4v data structure */
void print_bvh4_triangle4v(BVH4::NodeRef node, size_t depth)
{
if (node.isNode())
{
BVH4::Node* n = node.node();
std::cout << "Node {" << std::endl;
for (size_t i=0; i<BVH4::N; i++)
{
for (size_t k=0; k<depth; k++) std::cout << " ";
std::cout << " bounds" << i << " = " << n->bounds(i) << std::endl;
}
for (size_t i=0; i<BVH4::N; i++)
{
if (n->child(i) == BVH4::emptyNode)
continue;
for (size_t k=0; k<depth; k++) std::cout << " ";
std::cout << " child" << i << " = ";
print_bvh4_triangle4v(n->child(i),depth+1);
}
for (size_t k=0; k<depth; k++) std::cout << " ";
std::cout << "}" << std::endl;
}
else
{
size_t num;
const Triangle4v* tri = (const Triangle4v*) node.leaf(num);
std::cout << "Leaf {" << std::endl;
for (size_t i=0; i<num; i++) {
for (size_t j=0; j<tri[i].size(); j++) {
for (size_t k=0; k<depth; k++) std::cout << " ";
std::cout << " Triangle { v0 = (" << tri[i].v0.x[j] << ", " << tri[i].v0.y[j] << ", " << tri[i].v0.z[j] << "), "
"v1 = (" << tri[i].v1.x[j] << ", " << tri[i].v1.y[j] << ", " << tri[i].v1.z[j] << "), "
"v2 = (" << tri[i].v2.x[j] << ", " << tri[i].v2.y[j] << ", " << tri[i].v2.z[j] << "), "
"geomID = " << tri[i].geomID(j) << ", primID = " << tri[i].primID(j) << " }" << std::endl;
}
}
for (size_t k=0; k<depth; k++) std::cout << " ";
std::cout << "}" << std::endl;
}
}
void BVH4BuilderTopLevel::open_sequential()
{
size_t N = max(2*refs.size(),size_t(MIN_OPEN_SIZE));
refs.reserve(N);
std::make_heap(refs.begin(),refs.end());
while (refs.size()+3 <= N)
{
std::pop_heap (refs.begin(),refs.end());
BVH4::NodeRef ref = refs.back().node;
if (ref.isLeaf()) break;
refs.pop_back();
BVH4::Node* node = ref.node();
for (size_t i=0; i<4; i++) {
if (node->child(i) == BVH4::emptyNode) continue;
refs.push_back(BuildRef(node->bounds(i),node->child(i)));
std::push_heap (refs.begin(),refs.end());
}
}
}
void BVH4BuilderTopLevel::task_open_parallel(size_t threadIndex, size_t threadCount, size_t taskIndex, size_t taskCount, TaskScheduler::Event* event)
{
size_t N = global_dest;
size_t M = refs1.size();
const size_t start0 = (threadIndex+0)*N/threadCount;
const size_t end0 = (threadIndex+1)*N/threadCount;
const size_t start1 = (threadIndex+0)*M/threadCount;
const size_t end1 = (threadIndex+1)*M/threadCount;
assert(end1-start1 >= end0-start0);
BuildRef* prefs1 = &refs1[0];
/* copy from refs buffer to refs1 buffer */
for (size_t i=start0, j=start1; i<end0; i++, j++)
refs1[j] = refs[i];
/* create max heap in our set of items */
size_t start = start1;
size_t end = start1+end0-start0;
std::make_heap(&prefs1[start],&prefs1[end]);
float max_volume = 0.0f;
while (true)
{
barrier.wait(threadIndex,threadCount);
if (threadIndex == 0) global_max_volume = 0.0f;
barrier.wait(threadIndex,threadCount);
/* parallel calculation of maximal volume */
max_volume = 0.0f;
if (end+3 <= end1)
for (size_t i=start; i<end; i++)
max_volume = max(max_volume,prefs1[i].lower.w);
atomic_max_f32(&global_max_volume,max_volume);
barrier.wait(threadIndex,threadCount);
max_volume = global_max_volume;
barrier.wait(threadIndex,threadCount);
/* if maximal volume is 0, all threads are finished */
if (max_volume == 0.0f) break;
/* open all nodes that are considered large in this iteration */
while (end+3 <= end1)
{
if (end-start == 0) break;
std::pop_heap(&prefs1[start],&prefs1[end]);
BVH4::NodeRef ref = prefs1[end-1].node;
float vol = prefs1[end-1].lower.w;
if (ref.isLeaf() || vol < 0.5f*max_volume) {
std::push_heap(&prefs1[start],&prefs1[end]);
break;
}
end--;
BVH4::Node* node = ref.node();
for (size_t i=0; i<4; i++) {
if (node->child(i) == BVH4::emptyNode) continue;
prefs1[end++] = BuildRef(node->bounds(i),node->child(i));
std::push_heap(&prefs1[start],&prefs1[end]);
}
}
}
if (threadIndex == 0) global_dest = 0;
barrier.wait(threadIndex,threadCount);
/* copy again back to refs array */
size_t dest = atomic_add(&global_dest,end-start);
for (size_t i=start, j=dest; i<end; i++, j++)
refs[j] = refs1[i];
}