void build_sah(vector_t<PrimRef>& prims, isa::PrimInfo& pinfo)
{
size_t N = pinfo.size();
/* fast allocator that supports thread local operation */
FastAllocator allocator;
for (size_t i=0; i<2; i++)
{
std::cout << "iteration " << i << ": building BVH over " << N << " primitives, " << std::flush;
double t0 = getSeconds();
allocator.reset();
Node* root;
isa::BVHBuilderBinnedSAH::build<Node*>(
root,
/* thread local allocator for fast allocations */
[&] () -> FastAllocator::ThreadLocal* {
return allocator.threadLocal();
},
/* lambda function that creates BVH nodes */
[&](const isa::BVHBuilderBinnedSAH::BuildRecord& current, isa::BVHBuilderBinnedSAH::BuildRecord* children, const size_t N, FastAllocator::ThreadLocal* alloc) -> int
{
assert(N <= 2);
InnerNode* node = new (alloc->malloc(sizeof(InnerNode))) InnerNode;
for (size_t i=0; i<N; i++) {
node->bounds[i] = children[i].pinfo.geomBounds;
children[i].parent = (size_t*) &node->children[i];
}
*current.parent = (size_t) node;
return 0;
},
/* lambda function that creates BVH leaves */
[&](const isa::BVHBuilderBinnedSAH::BuildRecord& current, FastAllocator::ThreadLocal* alloc) -> int
{
assert(current.prims.size() == 1);
Node* node = new (alloc->malloc(sizeof(LeafNode))) LeafNode(prims[current.prims.begin()].ID(),prims[current.prims.begin()].bounds());
*current.parent = (size_t) node;
return 0;
},
/* progress monitor function */
[&] (size_t dn) {
// throw an exception here to cancel the build operation
},
prims.data(),pinfo,2,1024,1,1,1,1.0f,1.0f);
double t1 = getSeconds();
std::cout << 1000.0f*(t1-t0) << "ms, " << 1E-6*double(N)/(t1-t0) << " Mprims/s, sah = " << root->sah() << " [DONE]" << std::endl;
}
}
bool run ()
{
numFailed.store(0);
size_t numThreads = getNumberOfLogicalThreads();
barrier.init(numThreads+1);
/* create threads */
std::vector<thread_t> threads;
for (size_t i=0; i<numThreads; i++)
threads.push_back(createThread((thread_func)thread_alloc,this));
/* run test */
for (size_t i=0; i<1000; i++)
{
alloc.reset();
barrier.wait();
barrier.wait();
}
/* destroy threads */
for (size_t i=0; i<numThreads; i++)
join(threads[i]);
return numFailed == 0;
}
inline char* AllocateBlock( size_t s )
{
# if !defined (NO_FAST_ALLOCATOR)
char * const p = (char*)gFastAllocator.Allocate( s );
if ( p != 0 )
return p;
# endif
return (char*)EH_CSTD::malloc(s);
}
void _STLP_CALL operator delete(void* s)
# endif
{
if ( s != 0 )
{
if ( !using_alloc_set )
{
alloc_count--;
if ( gTestController.TrackingEnabled() && gTestController.LeakDetectionEnabled() )
{
using_alloc_set = true;
allocation_set::iterator p = alloc_set().find( (char*)s );
EH_ASSERT( p != alloc_set().end() );
alloc_set().erase( p );
using_alloc_set = false;
}
}
# if ! defined (NO_FAST_ALLOCATOR)
if ( !gFastAllocator.Free( s ) )
# endif
EH_CSTD::free(s);
}
}
inline void* MonomLex::operator new(std::size_t)
{
return Allocator.Allocate();
}
inline void MonomLex::operator delete(void* ptr)
{
Allocator.Free(ptr);
}
void build_morton(vector_t<PrimRef>& prims, isa::PrimInfo& pinfo)
{
size_t N = pinfo.size();
/* array for morton builder */
vector_t<isa::MortonID32Bit> morton_src(N);
vector_t<isa::MortonID32Bit> morton_tmp(N);
for (size_t i=0; i<N; i++)
morton_src[i].index = i;
/* fast allocator that supports thread local operation */
FastAllocator allocator;
for (size_t i=0; i<2; i++)
{
std::cout << "iteration " << i << ": building BVH over " << N << " primitives, " << std::flush;
double t0 = getSeconds();
allocator.reset();
std::pair<Node*,BBox3fa> node_bounds = isa::bvh_builder_morton<Node*>(
/* thread local allocator for fast allocations */
[&] () -> FastAllocator::ThreadLocal* {
return allocator.threadLocal();
},
BBox3fa(empty),
/* lambda function that allocates BVH nodes */
[&] ( isa::MortonBuildRecord<Node*>& current, isa::MortonBuildRecord<Node*>* children, size_t N, FastAllocator::ThreadLocal* alloc ) -> InnerNode*
{
assert(N <= 2);
InnerNode* node = new (alloc->malloc(sizeof(InnerNode))) InnerNode;
*current.parent = node;
for (size_t i=0; i<N; i++)
children[i].parent = &node->children[i];
return node;
},
/* lambda function that sets bounds */
[&] (InnerNode* node, const BBox3fa* bounds, size_t N) -> BBox3fa
{
BBox3fa res = empty;
for (size_t i=0; i<N; i++) {
const BBox3fa b = bounds[i];
res.extend(b);
node->bounds[i] = b;
}
return res;
},
/* lambda function that creates BVH leaves */
[&]( isa::MortonBuildRecord<Node*>& current, FastAllocator::ThreadLocal* alloc, BBox3fa& box_o) -> Node*
{
assert(current.size() == 1);
const size_t id = morton_src[current.begin].index;
const BBox3fa bounds = prims[id].bounds(); // FIXME: dont use morton_src, should be input
Node* node = new (alloc->malloc(sizeof(LeafNode))) LeafNode(id,bounds);
*current.parent = node;
box_o = bounds;
return node;
},
/* lambda that calculates the bounds for some primitive */
[&] (const isa::MortonID32Bit& morton) -> BBox3fa {
return prims[morton.index].bounds();
},
/* progress monitor function */
[&] (size_t dn) {
// throw an exception here to cancel the build operation
},
morton_src.data(),morton_tmp.data(),prims.size(),2,1024,1,1);
Node* root = node_bounds.first;
double t1 = getSeconds();
std::cout << 1000.0f*(t1-t0) << "ms, " << 1E-6*double(N)/(t1-t0) << " Mprims/s, sah = " << root->sah() << " [DONE]" << std::endl;
}
}
