Cycles: merging features from tomato branch.

=== BVH build time optimizations ===

* BVH building was multithreaded. Not all building is multithreaded, packing
  and the initial bounding/splitting is still single threaded, but recursive
  splitting is, which was the main bottleneck.

* Object splitting now uses binning rather than sorting of all elements, using
  code from the Embree raytracer from Intel.
  http://software.intel.com/en-us/articles/embree-photo-realistic-ray-tracing-kernels/

* Other small changes to avoid allocations, pack memory more tightly, avoid
  some unnecessary operations, ...

These optimizations do not work yet when Spatial Splits are enabled, for that
more work is needed. There's also other optimizations still needed, in
particular for the case of many low poly objects, the packing step and node
memory allocation.

BVH raytracing time should remain about the same, but BVH build time should be
significantly reduced, test here show speedup of about 5x to 10x on a dual core
and 5x to 25x on an 8-core machine, depending on the scene.

=== Threads ===

Centralized task scheduler for multithreading, which is basically the
CPU device threading code wrapped into something reusable.

Basic idea is that there is a single TaskScheduler that keeps a pool of threads,
one for each core. Other places in the code can then create a TaskPool that they
can drop Tasks in to be executed by the scheduler, and wait for them to complete
or cancel them early.

=== Normal ====

Added a Normal output to the texture coordinate node. This currently
gives the object space normal, which is the same under object animation.

In the future this might become a "generated" normal so it's also stable for
deforming objects, but for now it's already useful for non-deforming objects.

=== Render Layers ===

Per render layer Samples control, leaving it to 0 will use the common scene
setting.

Environment pass will now render environment even if film is set to transparent.

Exclude Layers" added. Scene layers (all object that influence the render,
directly or indirectly) are shared between all render layers. However sometimes
it's useful to leave out some object influence for a particular render layer.
That's what this option allows you to do.

=== Filter Glossy ===

When using a value higher than 0.0, this will blur glossy reflections after
blurry bounces, to reduce noise at the cost of accuracy. 1.0 is a good
starting value to tweak.

Some light paths have a low probability of being found while contributing much
light to the pixel. As a result these light paths will be found in some pixels
and not in others, causing fireflies. An example of such a difficult path might
be a small light that is causing a small specular highlight on a sharp glossy
material, which we are seeing through a rough glossy material. With path tracing
it is difficult to find the specular highlight, but if we increase the roughness
on the material the highlight gets bigger and softer, and so easier to find.

Often this blurring will be hardly noticeable, because we are seeing it through
a blurry material anyway, but there are also cases where this will lead to a
loss of detail in lighting.

1 files changed, 283 insertions, 353 deletions

diff --git a/intern/cycles/bvh/bvh_build.cpp b/intern/cycles/bvh/bvh_build.cpp
index 38674c2c561..c5b4f1d01ae 100644
--- a/intern/cycles/bvh/bvh_build.cpp
+++ b/intern/cycles/bvh/bvh_build.cpp
@@ -15,22 +15,36 @@
  * limitations under the License.
  */
 
+#include "bvh_binning.h"
 #include "bvh_build.h"
 #include "bvh_node.h"
 #include "bvh_params.h"
-#include "bvh_sort.h"
+#include "bvh_split.h"
 
 #include "mesh.h"
 #include "object.h"
 #include "scene.h"
 
-#include "util_algorithm.h"
+#include "util_debug.h"
 #include "util_foreach.h"
 #include "util_progress.h"
 #include "util_time.h"
 
 CCL_NAMESPACE_BEGIN
 
+/* BVH Build Task */
+
+class BVHBuildTask : public Task {
+public:
+	BVHBuildTask(InnerNode *node_, int child_, BVHObjectBinning& range_, int level_)
+	: node(node_), child(child_), level(level_), range(range_) {}
+
+	InnerNode *node;
+	int child;
+	int level;
+	BVHObjectBinning range;
+};
+
 /* Constructor / Destructor */
 
 BVHBuild::BVHBuild(const vector<Object*>& objects_,
@@ -41,10 +55,10 @@ BVHBuild::BVHBuild(const vector<Object*>& objects_,
   prim_object(prim_object_),
   params(params_),
   progress(progress_),
-  progress_start_time(0.0)
+  progress_start_time(0.0),
+  task_pool(function_bind(&BVHBuild::thread_build_node, this, _1, _2))
 {
 	spatial_min_overlap = 0.0f;
-	progress_num_duplicates = 0;
 }
 
 BVHBuild::~BVHBuild()
@@ -53,57 +67,63 @@ BVHBuild::~BVHBuild()
 
 /* Adding References */
 
-void BVHBuild::add_reference_mesh(NodeSpec& root, Mesh *mesh, int i)
+void BVHBuild::add_reference_mesh(BoundBox& root, BoundBox& center, Mesh *mesh, int i)
 {
 	for(uint j = 0; j < mesh->triangles.size(); j++) {
 		Mesh::Triangle t = mesh->triangles[j];
-		Reference ref;
+		BoundBox bounds = BoundBox::empty;
 
 		for(int k = 0; k < 3; k++) {
 			float3 pt = mesh->verts[t.v[k]];
-			ref.bounds.grow(pt);
+			bounds.grow(pt);
 		}
 
-		if(ref.bounds.valid()) {
-			ref.prim_index = j;
-			ref.prim_object = i;
-
-			references.push_back(ref);
-			root.bounds.grow(ref.bounds);
+		if(bounds.valid()) {
+			references.push_back(BVHReference(bounds, j, i));
+			root.grow(bounds);
+			center.grow(bounds.center2());
 		}
 	}
 }
 
-void BVHBuild::add_reference_object(NodeSpec& root, Object *ob, int i)
+void BVHBuild::add_reference_object(BoundBox& root, BoundBox& center, Object *ob, int i)
 {
-	Reference ref;
-
-	ref.prim_index = -1;
-	ref.prim_object = i;
-	ref.bounds = ob->bounds;
-
-	references.push_back(ref);
-	root.bounds.grow(ref.bounds);
+	references.push_back(BVHReference(ob->bounds, -1, i));
+	root.grow(ob->bounds);
+	center.grow(ob->bounds.center2());
 }
 
-void BVHBuild::add_references(NodeSpec& root)
+void BVHBuild::add_references(BVHRange& root)
 {
-	/* init root spec */
-	root.num = 0;
-	root.bounds = BoundBox();
+	/* reserve space for references */
+	size_t num_alloc_references = 0;
+
+	foreach(Object *ob, objects) {
+		if(params.top_level) {
+			if(ob->mesh->transform_applied)
+				num_alloc_references += ob->mesh->triangles.size();
+			else
+				num_alloc_references++;
+		}
+		else
+			num_alloc_references += ob->mesh->triangles.size();
+	}
+
+	references.reserve(num_alloc_references);
 
-	/* add objects */
+	/* add references from objects */
+	BoundBox bounds = BoundBox::empty, center = BoundBox::empty;
 	int i = 0;
 
 	foreach(Object *ob, objects) {
 		if(params.top_level) {
 			if(ob->mesh->transform_applied)
-				add_reference_mesh(root, ob->mesh, i);
+				add_reference_mesh(bounds, center, ob->mesh, i);
 			else
-				add_reference_object(root, ob, i);
+				add_reference_object(bounds, center, ob, i);
 		}
 		else
-			add_reference_mesh(root, ob->mesh, i);
+			add_reference_mesh(bounds, center, ob->mesh, i);
 
 		i++;
 
@@ -111,129 +131,213 @@ void BVHBuild::add_references(NodeSpec& root)
 	}
 
 	/* happens mostly on empty meshes */
-	if(!root.bounds.valid())
-		root.bounds.grow(make_float3(0.0f, 0.0f, 0.0f));
+	if(!bounds.valid())
+		bounds.grow(make_float3(0.0f, 0.0f, 0.0f));
 
-	root.num = references.size();
+	root = BVHRange(bounds, center, 0, references.size());
 }
 
 /* Build */
 
 BVHNode* BVHBuild::run()
 {
-	NodeSpec root;
+	BVHRange root;
 
 	/* add references */
 	add_references(root);
 
-	if(progress.get_cancel()) return NULL;
+	if(progress.get_cancel())
+		return NULL;
 
 	/* init spatial splits */
 	if(params.top_level) /* todo: get rid of this */
 		params.use_spatial_split = false;
 
-	spatial_min_overlap = root.bounds.area() * params.spatial_split_alpha;
+	spatial_min_overlap = root.bounds().safe_area() * params.spatial_split_alpha;
 	spatial_right_bounds.clear();
-	spatial_right_bounds.resize(max(root.num, (int)BVHParams::NUM_SPATIAL_BINS) - 1);
+	spatial_right_bounds.resize(max(root.size(), (int)BVHParams::NUM_SPATIAL_BINS) - 1);
 
 	/* init progress updates */
-	progress_num_duplicates = 0;
 	progress_start_time = time_dt();
+	progress_count = 0;
+	progress_total = references.size();
+	progress_original_total = progress_total;
+
+	prim_index.resize(references.size());
+	prim_object.resize(references.size());
 
 	/* build recursively */
-	return build_node(root, 0, 0.0f, 1.0f);
+	BVHNode *rootnode;
+
+	if(params.use_spatial_split) {
+		/* singlethreaded spatial split build */
+		rootnode = build_node(root, 0);
+	}
+	else {
+		/* multithreaded binning build */
+		BVHObjectBinning rootbin(root, &references[0]);
+		rootnode = build_node(rootbin, 0);
+		task_pool.wait();
+	}
+
+	/* delete if we cancelled */
+	if(rootnode) {
+		if(progress.get_cancel()) {
+			rootnode->deleteSubtree();
+			rootnode = NULL;
+		}
+		else if(!params.use_spatial_split) {
+			/*rotate(rootnode, 4, 5);*/
+			rootnode->update_visibility();
+		}
+	}
+
+	return rootnode;
 }
 
-void BVHBuild::progress_update(float progress_start, float progress_end)
+void BVHBuild::progress_update()
 {
 	if(time_dt() - progress_start_time < 0.25f)
 		return;
+	
+	double progress_start = (double)progress_count/(double)progress_total;
+	double duplicates = (double)(progress_total - progress_original_total)/(double)progress_total;
 
-	float duplicates = (float)progress_num_duplicates/(float)references.size();
 	string msg = string_printf("Building BVH %.0f%%, duplicates %.0f%%",
 		progress_start*100.0f, duplicates*100.0f);
 
 	progress.set_substatus(msg);
-	progress_start_time = time_dt();
+	progress_start_time = time_dt(); 
 }
 
-BVHNode* BVHBuild::build_node(const NodeSpec& spec, int level, float progress_start, float progress_end)
+void BVHBuild::thread_build_node(Task *task_, int thread_id)
 {
-	/* progress update */
-	progress_update(progress_start, progress_end);
-	if(progress.get_cancel()) return NULL;
+	if(progress.get_cancel())
+		return;
 
-	/* small enough or too deep => create leaf. */
-	if(spec.num <= params.min_leaf_size || level >= BVHParams::MAX_DEPTH)
-		return create_leaf_node(spec);
-
-	/* find split candidates. */
-	float area = spec.bounds.area();
-	float leafSAH = area * params.triangle_cost(spec.num);
-	float nodeSAH = area * params.node_cost(2);
-	ObjectSplit object = find_object_split(spec, nodeSAH);
-	SpatialSplit spatial;
-
-	if(params.use_spatial_split && level < BVHParams::MAX_SPATIAL_DEPTH) {
-		BoundBox overlap = object.left_bounds;
-		overlap.intersect(object.right_bounds);
-
-		if(overlap.area() >= spatial_min_overlap)
-			spatial = find_spatial_split(spec, nodeSAH);
-	}
+	/* build nodes */
+	BVHBuildTask *task = (BVHBuildTask*)task_;
+	BVHNode *node = build_node(task->range, task->level);
+
+	/* set child in inner node */
+	task->node->children[task->child] = node;
 
-	/* leaf SAH is the lowest => create leaf. */
-	float minSAH = min(min(leafSAH, object.sah), spatial.sah);
+	/* update progress */
+	if(task->range.size() < THREAD_TASK_SIZE) {
+		/*rotate(node, INT_MAX, 5);*/
 
-	if(minSAH == leafSAH && spec.num <= params.max_leaf_size)
-		return create_leaf_node(spec);
+		thread_scoped_lock lock(build_mutex);
 
-	/* perform split. */
-	NodeSpec left, right;
+		progress_count += task->range.size();
+		progress_update();
+	}
+}
+
+/* multithreaded binning builder */
+BVHNode* BVHBuild::build_node(const BVHObjectBinning& range, int level)
+{
+	size_t size = range.size();
+	float leafSAH = params.sah_triangle_cost * range.leafSAH;
+	float splitSAH = params.sah_node_cost * range.bounds().half_area() + params.sah_triangle_cost * range.splitSAH;
 
-	if(params.use_spatial_split && minSAH == spatial.sah)
-		do_spatial_split(left, right, spec, spatial);
-	if(!left.num || !right.num)
-		do_object_split(left, right, spec, object);
+	/* make leaf node when threshold reached or SAH tells us */
+	if(params.small_enough_for_leaf(size, level) || (size <= params.max_leaf_size && leafSAH < splitSAH))
+		return create_leaf_node(range);
+
+	/* perform split */
+	BVHObjectBinning left, right;
+	range.split(&references[0], left, right);
 
 	/* create inner node. */
-	progress_num_duplicates += left.num + right.num - spec.num;
+	InnerNode *inner;
 
-	float progress_mid = lerp(progress_start, progress_end, (float)right.num / (float)(left.num + right.num));
+	if(range.size() < THREAD_TASK_SIZE) {
+		/* local build */
+		BVHNode *leftnode = build_node(left, level + 1);
+		BVHNode *rightnode = build_node(right, level + 1);
 
-	BVHNode* rightNode = build_node(right, level + 1, progress_start, progress_mid);
-	if(progress.get_cancel()) {
-		if(rightNode) rightNode->deleteSubtree();
-		return NULL;
+		inner = new InnerNode(range.bounds(), leftnode, rightnode);
 	}
+	else {
+		/* threaded build */
+		inner = new InnerNode(range.bounds());
+
+		task_pool.push(new BVHBuildTask(inner, 0, left, level + 1), true);
+		task_pool.push(new BVHBuildTask(inner, 1, right, level + 1), true);
+	}
+
+	return inner;
+}
 
-	BVHNode* leftNode = build_node(left, level + 1, progress_mid, progress_end);
-	if(progress.get_cancel()) {
-		if(leftNode) leftNode->deleteSubtree();
+/* single threaded spatial split builder */
+BVHNode* BVHBuild::build_node(const BVHRange& range, int level)
+{
+	/* progress update */
+	progress_update();
+	if(progress.get_cancel())
 		return NULL;
+
+	/* small enough or too deep => create leaf. */
+	if(params.small_enough_for_leaf(range.size(), level)) {
+		progress_count += range.size();
+		return create_leaf_node(range);
+	}
+
+	/* splitting test */
+	BVHMixedSplit split(this, range, level);
+
+	if(split.no_split) {
+		progress_count += range.size();
+		return create_leaf_node(range);
 	}
+	
+	/* do split */
+	BVHRange left, right;
+	split.split(this, left, right, range);
+
+	progress_total += left.size() + right.size() - range.size();
+	size_t total = progress_total;
+
+	/* leaft node */
+	BVHNode *leftnode = build_node(left, level + 1);
+
+	/* right node (modify start for splits) */
+	right.set_start(right.start() + progress_total - total);
+	BVHNode *rightnode = build_node(right, level + 1);
 
-	return new InnerNode(spec.bounds, leftNode, rightNode);
+	/* inner node */
+	return new InnerNode(range.bounds(), leftnode, rightnode);
 }
 
-BVHNode *BVHBuild::create_object_leaf_nodes(const Reference *ref, int num)
+/* Create Nodes */
+
+BVHNode *BVHBuild::create_object_leaf_nodes(const BVHReference *ref, int start, int num)
 {
 	if(num == 0) {
-		BoundBox bounds;
+		BoundBox bounds = BoundBox::empty;
 		return new LeafNode(bounds, 0, 0, 0);
 	}
 	else if(num == 1) {
-		prim_index.push_back(ref[0].prim_index);
-		prim_object.push_back(ref[0].prim_object);
-		uint visibility = objects[ref[0].prim_object]->visibility;
-		return new LeafNode(ref[0].bounds, visibility, prim_index.size()-1, prim_index.size());
+		if(start == prim_index.size()) {
+			assert(params.use_spatial_split);
+
+			prim_index.push_back(ref->prim_index());
+			prim_object.push_back(ref->prim_object());
+		}
+		else {
+			prim_index[start] = ref->prim_index();
+			prim_object[start] = ref->prim_object();
+		}
+
+		uint visibility = objects[ref->prim_object()]->visibility;
+		return new LeafNode(ref->bounds(), visibility, start, start+1);
 	}
 	else {
 		int mid = num/2;
-		BVHNode *leaf0 = create_object_leaf_nodes(ref, mid); 
-		BVHNode *leaf1 = create_object_leaf_nodes(ref+mid, num-mid); 
+		BVHNode *leaf0 = create_object_leaf_nodes(ref, start, mid); 
+		BVHNode *leaf1 = create_object_leaf_nodes(ref+mid, start+mid, num-mid); 
 
-		BoundBox bounds;
+		BoundBox bounds = BoundBox::empty;
 		bounds.grow(leaf0->m_bounds);
 		bounds.grow(leaf1->m_bounds);
 
@@ -241,310 +345,136 @@ BVHNode *BVHBuild::create_object_leaf_nodes(const Reference *ref, int num)
 	}
 }
 
-BVHNode* BVHBuild::create_leaf_node(const NodeSpec& spec)
+BVHNode* BVHBuild::create_leaf_node(const BVHRange& range)
 {
 	vector<int>& p_index = prim_index;
 	vector<int>& p_object = prim_object;
-	BoundBox bounds;
-	int num = 0;
+	BoundBox bounds = BoundBox::empty;
+	int num = 0, ob_num = 0;
 	uint visibility = 0;
 
-	for(int i = 0; i < spec.num; i++) {
-		if(references.back().prim_index != -1) {
-			p_index.push_back(references.back().prim_index);
-			p_object.push_back(references.back().prim_object);
-			bounds.grow(references.back().bounds);
-			visibility |= objects[references.back().prim_object]->visibility;
-			references.pop_back();
+	for(int i = 0; i < range.size(); i++) {
+		BVHReference& ref = references[range.start() + i];
+
+		if(ref.prim_index() != -1) {
+			if(range.start() + num == prim_index.size()) {
+				assert(params.use_spatial_split);
+
+				p_index.push_back(ref.prim_index());
+				p_object.push_back(ref.prim_object());
+			}
+			else {
+				p_index[range.start() + num] = ref.prim_index();
+				p_object[range.start() + num] = ref.prim_object();
+			}
+
+			bounds.grow(ref.bounds());
+			visibility |= objects[ref.prim_object()]->visibility;
 			num++;
 		}
+		else {
+			if(ob_num < i)
+				references[range.start() + ob_num] = ref;
+			ob_num++;
+		}
 	}
 
 	BVHNode *leaf = NULL;
 	
 	if(num > 0) {
-		leaf = new LeafNode(bounds, visibility, p_index.size() - num, p_index.size());
+		leaf = new LeafNode(bounds, visibility, range.start(), range.start() + num);
 
-		if(num == spec.num)
+		if(num == range.size())
 			return leaf;
 	}
 
 	/* while there may be multiple triangles in a leaf, for object primitives
-	 * we want them to be the only one, so we  */
-	int ob_num = spec.num - num;
-	const Reference *ref = (ob_num)? &references.back() - (ob_num - 1): NULL;
-	BVHNode *oleaf = create_object_leaf_nodes(ref, ob_num);
-	for(int i = 0; i < ob_num; i++)
-		references.pop_back();
+	 * we want there to be the only one, so we keep splitting */
+	const BVHReference *ref = (ob_num)? &references[range.start()]: NULL;
+	BVHNode *oleaf = create_object_leaf_nodes(ref, range.start() + num, ob_num);
 	
 	if(leaf)
-		return new InnerNode(spec.bounds, leaf, oleaf);
+		return new InnerNode(range.bounds(), leaf, oleaf);
 	else
 		return oleaf;
 }
 
-/* Object Split */
+/* Tree Rotations */
 
-BVHBuild::ObjectSplit BVHBuild::find_object_split(const NodeSpec& spec, float nodeSAH)
+void BVHBuild::rotate(BVHNode *node, int max_depth, int iterations)
 {
-	ObjectSplit split;
-	const Reference *ref_ptr = &references[references.size() - spec.num];
-
-	for(int dim = 0; dim < 3; dim++) {
-		/* sort references */
-		bvh_reference_sort(references.size() - spec.num, references.size(), &references[0], dim);
-
-		/* sweep right to left and determine bounds. */
-		BoundBox right_bounds;
-
-		for(int i = spec.num - 1; i > 0; i--) {
-			right_bounds.grow(ref_ptr[i].bounds);
-			spatial_right_bounds[i - 1] = right_bounds;
-		}
-
-		/* sweep left to right and select lowest SAH. */
-		BoundBox left_bounds;
-
-		for(int i = 1; i < spec.num; i++) {
-			left_bounds.grow(ref_ptr[i - 1].bounds);
-			right_bounds = spatial_right_bounds[i - 1];
-
-			float sah = nodeSAH +
-				left_bounds.area() * params.triangle_cost(i) +
-				right_bounds.area() * params.triangle_cost(spec.num - i);
-
-			if(sah < split.sah) {
-				split.sah = sah;
-				split.dim = dim;
-				split.num_left = i;
-				split.left_bounds = left_bounds;
-				split.right_bounds = right_bounds;
-			}
-		}
-	}
-
-	return split;
+	/* in tested scenes, this resulted in slightly slower raytracing, so disabled
+	 * it for now. could be implementation bug, or depend on the scene */
+	if(node)
+		for(int i = 0; i < iterations; i++)
+			rotate(node, max_depth);
 }
 
-void BVHBuild::do_object_split(NodeSpec& left, NodeSpec& right, const NodeSpec& spec, const ObjectSplit& split)
+void BVHBuild::rotate(BVHNode *node, int max_depth)
 {
-	/* sort references according to split */
-	int start = references.size() - spec.num;
-	int end = references.size(); /* todo: is this right? */
-
-	bvh_reference_sort(start, end, &references[0], split.dim);
-
-	/* split node specs */
-	left.num = split.num_left;
-	left.bounds = split.left_bounds;
-	right.num = spec.num - split.num_left;
-	right.bounds = split.right_bounds;
-}
-
-/* Spatial Split */
-
-BVHBuild::SpatialSplit BVHBuild::find_spatial_split(const NodeSpec& spec, float nodeSAH)
-{
-	/* initialize bins. */
-	float3 origin = spec.bounds.min;
-	float3 binSize = (spec.bounds.max - origin) * (1.0f / (float)BVHParams::NUM_SPATIAL_BINS);
-	float3 invBinSize = 1.0f / binSize;
-
-	for(int dim = 0; dim < 3; dim++) {
-		for(int i = 0; i < BVHParams::NUM_SPATIAL_BINS; i++) {
-			SpatialBin& bin = spatial_bins[dim][i];
-
-			bin.bounds = BoundBox();
-			bin.enter = 0;
-			bin.exit = 0;
-		}
-	}
-
-	/* chop references into bins. */
-	for(unsigned int refIdx = references.size() - spec.num; refIdx < references.size(); refIdx++) {
-		const Reference& ref = references[refIdx];
-		float3 firstBinf = (ref.bounds.min - origin) * invBinSize;
-		float3 lastBinf = (ref.bounds.max - origin) * invBinSize;
-		int3 firstBin = make_int3((int)firstBinf.x, (int)firstBinf.y, (int)firstBinf.z);
-		int3 lastBin = make_int3((int)lastBinf.x, (int)lastBinf.y, (int)lastBinf.z);
+	/* nothing to rotate if we reached a leaf node. */
+	if(node->is_leaf() || max_depth < 0)
+		return;
+	
+	InnerNode *parent = (InnerNode*)node;
 
-		firstBin = clamp(firstBin, 0, BVHParams::NUM_SPATIAL_BINS - 1);
-		lastBin = clamp(lastBin, firstBin, BVHParams::NUM_SPATIAL_BINS - 1);
+	/* rotate all children first */
+	for(size_t c = 0; c < 2; c++)
+		rotate(parent->children[c], max_depth-1);
 
-		for(int dim = 0; dim < 3; dim++) {
-			Reference currRef = ref;
+	/* compute current area of all children */
+	BoundBox bounds0 = parent->children[0]->m_bounds;
+	BoundBox bounds1 = parent->children[1]->m_bounds;
 
-			for(int i = firstBin[dim]; i < lastBin[dim]; i++) {
-				Reference leftRef, rightRef;
+	float area0 = bounds0.half_area();
+	float area1 = bounds1.half_area();
+	float4 child_area = make_float4(area0, area1, 0.0f, 0.0f);
 
-				split_reference(leftRef, rightRef, currRef, dim, origin[dim] + binSize[dim] * (float)(i + 1));
-				spatial_bins[dim][i].bounds.grow(leftRef.bounds);
-				currRef = rightRef;
-			}
+	/* find best rotation. we pick a target child of a first child, and swap
+	 * this with an other child. we perform the best such swap. */
+	float best_cost = FLT_MAX;
+	int best_child = -1, bets_target = -1, best_other = -1;
 
-			spatial_bins[dim][lastBin[dim]].bounds.grow(currRef.bounds);
-			spatial_bins[dim][firstBin[dim]].enter++;
-			spatial_bins[dim][lastBin[dim]].exit++;
-		}
-	}
+	for(size_t c = 0; c < 2; c++) {
+		/* ignore leaf nodes as we cannot descent into */
+		if(parent->children[c]->is_leaf())
+			continue;
 
-	/* select best split plane. */
-	SpatialSplit split;
+		InnerNode *child = (InnerNode*)parent->children[c];
+		BoundBox& other = (c == 0)? bounds1: bounds0;
 
-	for(int dim = 0; dim < 3; dim++) {
-		/* sweep right to left and determine bounds. */
-		BoundBox right_bounds;
+		/* transpose child bounds */
+		BoundBox target0 = child->children[0]->m_bounds;
+		BoundBox target1 = child->children[1]->m_bounds;
 
-		for(int i = BVHParams::NUM_SPATIAL_BINS - 1; i > 0; i--) {
-			right_bounds.grow(spatial_bins[dim][i].bounds);
-			spatial_right_bounds[i - 1] = right_bounds;
-		}
+		/* compute cost for both possible swaps */
+		float cost0 = merge(other, target1).half_area() - child_area[c];
+		float cost1 = merge(target0, other).half_area() - child_area[c];
 
-		/* sweep left to right and select lowest SAH. */
-		BoundBox left_bounds;
-		int leftNum = 0;
-		int rightNum = spec.num;
+		if(min(cost0,cost1) < best_cost) {
+			best_child = (int)c;
+			best_other = (int)(1-c);
 
-		for(int i = 1; i < BVHParams::NUM_SPATIAL_BINS; i++) {
-			left_bounds.grow(spatial_bins[dim][i - 1].bounds);
-			leftNum += spatial_bins[dim][i - 1].enter;
-			rightNum -= spatial_bins[dim][i - 1].exit;
-
-			float sah = nodeSAH +
-				left_bounds.area() * params.triangle_cost(leftNum) +
-				spatial_right_bounds[i - 1].area() * params.triangle_cost(rightNum);
-
-			if(sah < split.sah) {
-				split.sah = sah;
-				split.dim = dim;
-				split.pos = origin[dim] + binSize[dim] * (float)i;
+			if(cost0 < cost1) {
+				best_cost = cost0;
+				bets_target = 0;
+			}
+			else {
+				best_cost = cost0;
+				bets_target = 1;
 			}
 		}
 	}
 
-	return split;
-}
-
-void BVHBuild::do_spatial_split(NodeSpec& left, NodeSpec& right, const NodeSpec& spec, const SpatialSplit& split)
-{
-	/* Categorize references and compute bounds.
-	 *
-	 * Left-hand side:			[left_start, left_end[
-	 * Uncategorized/split:		[left_end, right_start[
-	 * Right-hand side:			[right_start, refs.size()[ */
-
-	vector<Reference>& refs = references;
-	int left_start = refs.size() - spec.num;
-	int left_end = left_start;
-	int right_start = refs.size();
-
-	left.bounds = right.bounds = BoundBox();
-
-	for(int i = left_end; i < right_start; i++) {
-		if(refs[i].bounds.max[split.dim] <= split.pos) {
-			/* entirely on the left-hand side */
-			left.bounds.grow(refs[i].bounds);
-			swap(refs[i], refs[left_end++]);
-		}
-		else if(refs[i].bounds.min[split.dim] >= split.pos) {
-			/* entirely on the right-hand side */
-			right.bounds.grow(refs[i].bounds);
-			swap(refs[i--], refs[--right_start]);
-		}
-	}
-
-	/* duplicate or unsplit references intersecting both sides. */
-	while(left_end < right_start) {
-		/* split reference. */
-		Reference lref, rref;
-
-		split_reference(lref, rref, refs[left_end], split.dim, split.pos);
-
-		/* compute SAH for duplicate/unsplit candidates. */
-		BoundBox lub = left.bounds;		// Unsplit to left:		new left-hand bounds.
-		BoundBox rub = right.bounds;	// Unsplit to right:	new right-hand bounds.
-		BoundBox ldb = left.bounds;		// Duplicate:			new left-hand bounds.
-		BoundBox rdb = right.bounds;	// Duplicate:			new right-hand bounds.
-
-		lub.grow(refs[left_end].bounds);
-		rub.grow(refs[left_end].bounds);
-		ldb.grow(lref.bounds);
-		rdb.grow(rref.bounds);
-
-		float lac = params.triangle_cost(left_end - left_start);
-		float rac = params.triangle_cost(refs.size() - right_start);
-		float lbc = params.triangle_cost(left_end - left_start + 1);
-		float rbc = params.triangle_cost(refs.size() - right_start + 1);
-
-		float unsplitLeftSAH = lub.area() * lbc + right.bounds.area() * rac;
-		float unsplitRightSAH = left.bounds.area() * lac + rub.area() * rbc;
-		float duplicateSAH = ldb.area() * lbc + rdb.area() * rbc;
-		float minSAH = min(min(unsplitLeftSAH, unsplitRightSAH), duplicateSAH);
-
-		if(minSAH == unsplitLeftSAH) {
-			/* unsplit to left */
-			left.bounds = lub;
-			left_end++;
-		}
-		else if(minSAH == unsplitRightSAH) {
-			/* unsplit to right */
-			right.bounds = rub;
-			swap(refs[left_end], refs[--right_start]);
-		}
-		else {
-			/* duplicate */
-			left.bounds = ldb;
-			right.bounds = rdb;
-			refs[left_end++] = lref;
-			refs.push_back(rref);
-		}
-	}
-
-	left.num = left_end - left_start;
-	right.num = refs.size() - right_start;
-}
+	/* if we did not find a swap that improves the SAH then do nothing */
+	if(best_cost >= 0)
+		return;
 
-void BVHBuild::split_reference(Reference& left, Reference& right, const Reference& ref, int dim, float pos)
-{
-	/* initialize references. */
-	left.prim_index = right.prim_index = ref.prim_index;
-	left.prim_object = right.prim_object = ref.prim_object;
-	left.bounds = right.bounds = BoundBox();
-
-	/* loop over vertices/edges. */
-	Object *ob = objects[ref.prim_object];
-	const Mesh *mesh = ob->mesh;
-	const int *inds = mesh->triangles[ref.prim_index].v;
-	const float3 *verts = &mesh->verts[0];
-	const float3* v1 = &verts[inds[2]];
-
-	for(int i = 0; i < 3; i++) {
-		const float3* v0 = v1;
-		int vindex = inds[i];
-		v1 = &verts[vindex];
-		float v0p = (*v0)[dim];
-		float v1p = (*v1)[dim];
-
-		/* insert vertex to the boxes it belongs to. */
-		if(v0p <= pos)
-			left.bounds.grow(*v0);
-
-		if(v0p >= pos)
-			right.bounds.grow(*v0);
-
-		/* edge intersects the plane => insert intersection to both boxes. */
-		if((v0p < pos && v1p > pos) || (v0p > pos && v1p < pos)) {
-			float3 t = lerp(*v0, *v1, clamp((pos - v0p) / (v1p - v0p), 0.0f, 1.0f));
-			left.bounds.grow(t);
-			right.bounds.grow(t);
-		}
-	}
+	/* perform the best found tree rotation */
+	InnerNode *child = (InnerNode*)parent->children[best_child];
 
-	/* intersect with original bounds. */
-	left.bounds.max[dim] = pos;
-	right.bounds.min[dim] = pos;
-	left.bounds.intersect(ref.bounds);
-	right.bounds.intersect(ref.bounds);
+	swap(parent->children[best_other], child->children[bets_target]);
+	child->m_bounds = merge(child->children[0]->m_bounds, child->children[1]->m_bounds);
 }
 
 CCL_NAMESPACE_END