Cycles: Switch node address to absolute values in BVH tree

This seems to be straightforward way to support heterogeneous nodes
in the same tree.

There is some penalty related on 4gig limit of the address space now,
but here's are the thing:

Traversal code was already using ints to store final offset, so
there can't be regressions really.

This is a required commit to make it possible to encode both aligned
and unaligned nodes in the same array. Also, in the future we can use
this to get rid of __leaf_nodes array (which is a bit tricky to do since
trickery in pack_instances().

13 files changed, 82 insertions, 78 deletions

diff --git a/intern/cycles/bvh/bvh.cpp b/intern/cycles/bvh/bvh.cpp
index 5669bab857c..ab4dbe814ab 100644
--- a/intern/cycles/bvh/bvh.cpp
+++ b/intern/cycles/bvh/bvh.cpp
@@ -288,8 +288,8 @@ void BVH::pack_instances(size_t nodes_size, size_t leaf_nodes_size)
 
 		BVH *bvh = mesh->bvh;
 
-		int noffset = nodes_offset/nsize;
-		int noffset_leaf = nodes_leaf_offset/nsize_leaf;
+		int noffset = nodes_offset;
+		int noffset_leaf = nodes_leaf_offset;
 		int mesh_tri_offset = mesh->tri_offset;
 		int mesh_curve_offset = mesh->curve_offset;
 
@@ -421,7 +421,7 @@ void RegularBVH::pack_leaf(const BVHStackEntry& e, const LeafNode *leaf)
 		data[0].w = __uint_as_float(pack.prim_type[leaf->m_lo]);
 	}
 
-	memcpy(&pack.leaf_nodes[e.idx * BVH_NODE_LEAF_SIZE], data, sizeof(float4)*BVH_NODE_LEAF_SIZE);
+	memcpy(&pack.leaf_nodes[e.idx], data, sizeof(float4)*BVH_NODE_LEAF_SIZE);
 }
 
 void RegularBVH::pack_inner(const BVHStackEntry& e, const BVHStackEntry& e0, const BVHStackEntry& e1)
@@ -439,7 +439,7 @@ void RegularBVH::pack_node(int idx, const BoundBox& b0, const BoundBox& b1, int
 		make_int4(c0, c1, visibility0, visibility1)
 	};
 
-	memcpy(&pack.nodes[idx * BVH_NODE_SIZE], data, sizeof(int4)*BVH_NODE_SIZE);
+	memcpy(&pack.nodes[idx], data, sizeof(int4)*BVH_NODE_SIZE);
 }
 
 void RegularBVH::pack_nodes(const BVHNode *root)
@@ -465,10 +465,13 @@ void RegularBVH::pack_nodes(const BVHNode *root)
 
 	vector<BVHStackEntry> stack;
 	stack.reserve(BVHParams::MAX_DEPTH*2);
-	if(root->is_leaf())
+	if(root->is_leaf()) {
 		stack.push_back(BVHStackEntry(root, nextLeafNodeIdx++));
-	else
-		stack.push_back(BVHStackEntry(root, nextNodeIdx++));
+	}
+	else {
+		stack.push_back(BVHStackEntry(root, nextNodeIdx));
+		nextNodeIdx += BVH_NODE_SIZE;
+	}
 
 	while(stack.size()) {
 		BVHStackEntry e = stack.back();
@@ -481,10 +484,19 @@ void RegularBVH::pack_nodes(const BVHNode *root)
 		}
 		else {
 			/* innner node */
-			int idx0 = (e.node->get_child(0)->is_leaf())? (nextLeafNodeIdx++) : (nextNodeIdx++);
-			int idx1 = (e.node->get_child(1)->is_leaf())? (nextLeafNodeIdx++) : (nextNodeIdx++);
-			stack.push_back(BVHStackEntry(e.node->get_child(0), idx0));
-			stack.push_back(BVHStackEntry(e.node->get_child(1), idx1));
+			int idx[2];
+			for (int i = 0; i < 2; ++i) {
+				if (e.node->get_child(i)->is_leaf()) {
+					idx[i] = nextLeafNodeIdx++;
+				}
+				else {
+					idx[i] = nextNodeIdx;
+					nextNodeIdx += BVH_NODE_SIZE;
+				}
+			}
+
+			stack.push_back(BVHStackEntry(e.node->get_child(0), idx[0]));
+			stack.push_back(BVHStackEntry(e.node->get_child(1), idx[1]));
 
 			pack_inner(e, stack[stack.size()-2], stack[stack.size()-1]);
 		}
@@ -506,7 +518,7 @@ void RegularBVH::refit_nodes()
 void RegularBVH::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 {
 	if(leaf) {
-		int4 *data = &pack.leaf_nodes[idx*BVH_NODE_LEAF_SIZE];
+		int4 *data = &pack.leaf_nodes[idx];
 		int c0 = data[0].x;
 		int c1 = data[0].y;
 		/* refit leaf node */
@@ -580,12 +592,12 @@ void RegularBVH::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility
 		leaf_data[0].y = __int_as_float(c1);
 		leaf_data[0].z = __uint_as_float(visibility);
 		leaf_data[0].w = __uint_as_float(data[0].w);
-		memcpy(&pack.leaf_nodes[idx * BVH_NODE_LEAF_SIZE],
+		memcpy(&pack.leaf_nodes[idx],
 		       leaf_data,
 		       sizeof(float4)*BVH_NODE_LEAF_SIZE);
 	}
 	else {
-		int4 *data = &pack.nodes[idx*BVH_NODE_SIZE];
+		int4 *data = &pack.nodes[idx];
 		int c0 = data[3].x;
 		int c1 = data[3].y;
 		/* refit inner node, set bbox from children */
@@ -630,7 +642,7 @@ void QBVH::pack_leaf(const BVHStackEntry& e, const LeafNode *leaf)
 		data[0].w = __uint_as_float(pack.prim_type[leaf->m_lo]);
 	}
 
-	memcpy(&pack.leaf_nodes[e.idx * BVH_QNODE_LEAF_SIZE], data, sizeof(float4)*BVH_QNODE_LEAF_SIZE);
+	memcpy(&pack.leaf_nodes[e.idx], data, sizeof(float4)*BVH_QNODE_LEAF_SIZE);
 }
 
 void QBVH::pack_inner(const BVHStackEntry& e, const BVHStackEntry *en, int num)
@@ -668,7 +680,7 @@ void QBVH::pack_inner(const BVHStackEntry& e, const BVHStackEntry *en, int num)
 		data[7][i] = __int_as_float(0);
 	}
 
-	memcpy(&pack.nodes[e.idx * BVH_QNODE_SIZE], data, sizeof(float4)*BVH_QNODE_SIZE);
+	memcpy(&pack.nodes[e.idx], data, sizeof(float4)*BVH_QNODE_SIZE);
 }
 
 /* Quad SIMD Nodes */
@@ -701,7 +713,8 @@ void QBVH::pack_nodes(const BVHNode *root)
 		stack.push_back(BVHStackEntry(root, nextLeafNodeIdx++));
 	}
 	else {
-		stack.push_back(BVHStackEntry(root, nextNodeIdx++));
+		stack.push_back(BVHStackEntry(root, nextNodeIdx));
+		nextNodeIdx += BVH_QNODE_SIZE;
 	}
 
 	while(stack.size()) {
@@ -746,7 +759,8 @@ void QBVH::pack_nodes(const BVHNode *root)
 					idx = nextLeafNodeIdx++;
 				}
 				else {
-					idx = nextNodeIdx++;
+					idx = nextNodeIdx;
+					nextNodeIdx += BVH_QNODE_SIZE;
 				}
 				stack.push_back(BVHStackEntry(nodes[i], idx));
 			}
@@ -772,7 +786,7 @@ void QBVH::refit_nodes()
 void QBVH::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 {
 	if(leaf) {
-		int4 *data = &pack.leaf_nodes[idx*BVH_QNODE_LEAF_SIZE];
+		int4 *data = &pack.leaf_nodes[idx];
 		int4 c = data[0];
 		/* Refit leaf node. */
 		for(int prim = c.x; prim < c.y; prim++) {
@@ -854,12 +868,12 @@ void QBVH::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 		leaf_data[0].y = __int_as_float(c.y);
 		leaf_data[0].z = __uint_as_float(visibility);
 		leaf_data[0].w = __uint_as_float(c.w);
-		memcpy(&pack.leaf_nodes[idx * BVH_QNODE_LEAF_SIZE],
+		memcpy(&pack.leaf_nodes[idx],
 		       leaf_data,
 		       sizeof(float4)*BVH_QNODE_LEAF_SIZE);
 	}
 	else {
-		int4 *data = &pack.nodes[idx*BVH_QNODE_SIZE];
+		int4 *data = &pack.nodes[idx];
 		int4 c = data[7];
 		/* Refit inner node, set bbox from children. */
 		BoundBox child_bbox[4] = {BoundBox::empty,
@@ -895,7 +909,7 @@ void QBVH::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 			inner_data[6][i] = bb_max.z;
 			inner_data[7][i] = __int_as_float(c[i]);
 		}
-		memcpy(&pack.nodes[idx * BVH_QNODE_SIZE],
+		memcpy(&pack.nodes[idx],
 		       inner_data,
 		       sizeof(float4)*BVH_QNODE_SIZE);
 	}
diff --git a/intern/cycles/kernel/geom/geom.h b/intern/cycles/kernel/geom/geom.h
index 26e0af99bbc..33e91d1ee44 100644
--- a/intern/cycles/kernel/geom/geom.h
+++ b/intern/cycles/kernel/geom/geom.h
@@ -21,10 +21,6 @@
 /* 64 object BVH + 64 mesh BVH + 64 object node splitting */
 #define BVH_STACK_SIZE 192
 #define BVH_QSTACK_SIZE 384
-#define BVH_NODE_SIZE 4
-#define BVH_NODE_LEAF_SIZE 1
-#define BVH_QNODE_SIZE 8
-#define BVH_QNODE_LEAF_SIZE 1
 #define TRI_NODE_SIZE 3
 
 #include "geom_attribute.h"
diff --git a/intern/cycles/kernel/geom/geom_bvh_shadow.h b/intern/cycles/kernel/geom/geom_bvh_shadow.h
index 4991eb5045a..60cc50e8bfd 100644
--- a/intern/cycles/kernel/geom/geom_bvh_shadow.h
+++ b/intern/cycles/kernel/geom/geom_bvh_shadow.h
@@ -102,10 +102,10 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				float t = isect_t;
 
 				/* fetch node data */
-				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
-				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
-				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
+				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
 
 				/* intersect ray against child nodes */
 				float c0lox = (node0.x - P.x) * idir.x;
@@ -140,7 +140,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
 
 				/* fetch node data */
-				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
+				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr;
 				const float4 cnodes = ((float4*)bvh_nodes)[3];
 
 				/* intersect ray against child nodes */
@@ -200,7 +200,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 
 			/* if node is leaf, fetch triangle list */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_NODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)
diff --git a/intern/cycles/kernel/geom/geom_bvh_subsurface.h b/intern/cycles/kernel/geom/geom_bvh_subsurface.h
index a5243f079f2..e43eedad7bc 100644
--- a/intern/cycles/kernel/geom/geom_bvh_subsurface.h
+++ b/intern/cycles/kernel/geom/geom_bvh_subsurface.h
@@ -113,10 +113,10 @@ ccl_device void BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				float t = isect_t;
 
 				/* fetch node data */
-				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
-				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
-				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
+				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
 
 				/* intersect ray against child nodes */
 				float c0lox = (node0.x - P.x) * idir.x;
@@ -145,7 +145,7 @@ ccl_device void BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
 
 				/* fetch node data */
-				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
+				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr;
 				const float4 cnodes = ((float4*)bvh_nodes)[3];
 
 				/* intersect ray against child nodes */
@@ -199,7 +199,7 @@ ccl_device void BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 
 			/* if node is leaf, fetch triangle list */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_NODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 				const int primAddr2 = __float_as_int(leaf.y);
diff --git a/intern/cycles/kernel/geom/geom_bvh_traversal.h b/intern/cycles/kernel/geom/geom_bvh_traversal.h
index 89e6ba6162d..5b9c7b46f82 100644
--- a/intern/cycles/kernel/geom/geom_bvh_traversal.h
+++ b/intern/cycles/kernel/geom/geom_bvh_traversal.h
@@ -109,10 +109,10 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				float t = isect->t;
 
 				/* fetch node data */
-				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
-				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
-				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
+				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
 
 				/* intersect ray against child nodes */
 				float c0lox = (node0.x - P.x) * idir.x;
@@ -162,7 +162,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
 
 				/* fetch node data */
-				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
+				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr;
 				const float4 cnodes = ((float4*)bvh_nodes)[3];
 
 				/* intersect ray against child nodes */
@@ -243,7 +243,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 
 			/* if node is leaf, fetch triangle list */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_NODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)
diff --git a/intern/cycles/kernel/geom/geom_bvh_volume.h b/intern/cycles/kernel/geom/geom_bvh_volume.h
index 195c4043cd7..36033ecf459 100644
--- a/intern/cycles/kernel/geom/geom_bvh_volume.h
+++ b/intern/cycles/kernel/geom/geom_bvh_volume.h
@@ -99,10 +99,10 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				float t = isect->t;
 
 				/* fetch node data */
-				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
-				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
-				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
+				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
 
 				/* intersect ray against child nodes */
 				float c0lox = (node0.x - P.x) * idir.x;
@@ -131,7 +131,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
 
 				/* fetch node data */
-				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
+				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr;
 				const float4 cnodes = ((float4*)bvh_nodes)[3];
 
 				/* intersect ray against child nodes */
@@ -186,7 +186,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 
 			/* if node is leaf, fetch triangle list */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_NODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)
diff --git a/intern/cycles/kernel/geom/geom_bvh_volume_all.h b/intern/cycles/kernel/geom/geom_bvh_volume_all.h
index c76d6155065..f9536148933 100644
--- a/intern/cycles/kernel/geom/geom_bvh_volume_all.h
+++ b/intern/cycles/kernel/geom/geom_bvh_volume_all.h
@@ -103,10 +103,10 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				float t = isect_array->t;
 
 				/* fetch node data */
-				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+0);
-				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+1);
-				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+2);
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_NODE_SIZE+3);
+				float4 node0 = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
+				float4 node1 = kernel_tex_fetch(__bvh_nodes, nodeAddr+1);
+				float4 node2 = kernel_tex_fetch(__bvh_nodes, nodeAddr+2);
+				float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+3);
 
 				/* intersect ray against child nodes */
 				float c0lox = (node0.x - P.x) * idir.x;
@@ -135,7 +135,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				/* Intersect two child bounding boxes, SSE3 version adapted from Embree */
 
 				/* fetch node data */
-				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr*BVH_NODE_SIZE;
+				const ssef *bvh_nodes = (ssef*)kg->__bvh_nodes.data + nodeAddr;
 				const float4 cnodes = ((float4*)bvh_nodes)[3];
 
 				/* intersect ray against child nodes */
@@ -190,7 +190,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 
 			/* if node is leaf, fetch triangle list */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_NODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)
diff --git a/intern/cycles/kernel/geom/geom_qbvh.h b/intern/cycles/kernel/geom/geom_qbvh.h
index 118bceb4bec..30ed851d861 100644
--- a/intern/cycles/kernel/geom/geom_qbvh.h
+++ b/intern/cycles/kernel/geom/geom_qbvh.h
@@ -69,7 +69,7 @@ ccl_device_inline int qbvh_node_intersect(KernelGlobals *__restrict kg,
                                           const int nodeAddr,
                                           ssef *__restrict dist)
 {
-	const int offset = nodeAddr*BVH_QNODE_SIZE + 1;
+	const int offset = nodeAddr + 1;
 #ifdef __KERNEL_AVX2__
 	const ssef tnear_x = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset+near_x), idir.x, org_idir.x);
 	const ssef tnear_y = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset+near_y), idir.y, org_idir.y);
@@ -120,7 +120,7 @@ ccl_device_inline int qbvh_node_intersect_robust(KernelGlobals *__restrict kg,
                                                  const float difl,
                                                  ssef *__restrict dist)
 {
-	const int offset = nodeAddr*BVH_QNODE_SIZE + 1;
+	const int offset = nodeAddr + 1;
 #ifdef __KERNEL_AVX2__
 	const ssef tnear_x = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset+near_x), idir.x, P_idir.x);
 	const ssef tnear_y = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset+near_y), idir.y, P_idir.y);
diff --git a/intern/cycles/kernel/geom/geom_qbvh_shadow.h b/intern/cycles/kernel/geom/geom_qbvh_shadow.h
index 01babba0215..97a0ceb0687 100644
--- a/intern/cycles/kernel/geom/geom_qbvh_shadow.h
+++ b/intern/cycles/kernel/geom/geom_qbvh_shadow.h
@@ -97,7 +97,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 		do {
 			/* Traverse internal nodes. */
 			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
-				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_QNODE_SIZE+0);
+				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
 
 #ifdef __VISIBILITY_FLAG__
 				if((__float_as_uint(inodes.x) & PATH_RAY_SHADOW) == 0) {
@@ -124,8 +124,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 				                                        &dist);
 
 				if(traverseChild != 0) {
-					float4 cnodes = kernel_tex_fetch(__bvh_nodes,
-					                                 nodeAddr*BVH_QNODE_SIZE+7);
+					float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+7);
 
 					/* One child is hit, continue with that child. */
 					int r = __bscf(traverseChild);
@@ -218,7 +217,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 
 			/* If node is leaf, fetch triangle list. */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 #ifdef __VISIBILITY_FLAG__
 				if((__float_as_uint(leaf.z) & PATH_RAY_SHADOW) == 0) {
 					/* Pop. */
diff --git a/intern/cycles/kernel/geom/geom_qbvh_subsurface.h b/intern/cycles/kernel/geom/geom_qbvh_subsurface.h
index bc717a0747e..5d76ac4b1f1 100644
--- a/intern/cycles/kernel/geom/geom_qbvh_subsurface.h
+++ b/intern/cycles/kernel/geom/geom_qbvh_subsurface.h
@@ -123,8 +123,7 @@ ccl_device void BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 				                                        &dist);
 
 				if(traverseChild != 0) {
-					float4 cnodes = kernel_tex_fetch(__bvh_nodes,
-					                                 nodeAddr*BVH_QNODE_SIZE+7);
+					float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+7);
 
 					/* One child is hit, continue with that child. */
 					int r = __bscf(traverseChild);
@@ -217,7 +216,7 @@ ccl_device void BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 
 			/* If node is leaf, fetch triangle list. */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 				int primAddr2 = __float_as_int(leaf.y);
diff --git a/intern/cycles/kernel/geom/geom_qbvh_traversal.h b/intern/cycles/kernel/geom/geom_qbvh_traversal.h
index 098881a6d7e..1588ae3605c 100644
--- a/intern/cycles/kernel/geom/geom_qbvh_traversal.h
+++ b/intern/cycles/kernel/geom/geom_qbvh_traversal.h
@@ -106,8 +106,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 		do {
 			/* Traverse internal nodes. */
 			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
-				float4 inodes = kernel_tex_fetch(__bvh_nodes, 
-				                                 nodeAddr*BVH_QNODE_SIZE+0);
+				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
 
 				if(UNLIKELY(nodeDist > isect->t)
 #ifdef __VISIBILITY_FLAG__
@@ -167,8 +166,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 				}
 
 				if(traverseChild != 0) {
-					float4 cnodes = kernel_tex_fetch(__bvh_nodes,
-					                                 nodeAddr*BVH_QNODE_SIZE+7);
+					float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+7);
 
 					/* One child is hit, continue with that child. */
 					int r = __bscf(traverseChild);
@@ -266,7 +264,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 
 			/* If node is leaf, fetch triangle list. */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 
 #ifdef __VISIBILITY_FLAG__
 				if(UNLIKELY((nodeDist > isect->t) ||
diff --git a/intern/cycles/kernel/geom/geom_qbvh_volume.h b/intern/cycles/kernel/geom/geom_qbvh_volume.h
index 2aae90380c1..d66c6e2b1e5 100644
--- a/intern/cycles/kernel/geom/geom_qbvh_volume.h
+++ b/intern/cycles/kernel/geom/geom_qbvh_volume.h
@@ -94,7 +94,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			/* Traverse internal nodes. */
 			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
 #ifdef __VISIBILITY_FLAG__
-				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_QNODE_SIZE+0);
+				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
 				if((__float_as_uint(inodes.x) & visibility) == 0) {
 					/* Pop. */
 					nodeAddr = traversalStack[stackPtr].addr;
@@ -119,8 +119,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 				                                        &dist);
 
 				if(traverseChild != 0) {
-					float4 cnodes = kernel_tex_fetch(__bvh_nodes,
-					                                 nodeAddr*BVH_QNODE_SIZE+7);
+					float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+7);
 
 					/* One child is hit, continue with that child. */
 					int r = __bscf(traverseChild);
@@ -213,7 +212,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 
 			/* If node is leaf, fetch triangle list. */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)
diff --git a/intern/cycles/kernel/geom/geom_qbvh_volume_all.h b/intern/cycles/kernel/geom/geom_qbvh_volume_all.h
index e6b668eb758..89950f17c64 100644
--- a/intern/cycles/kernel/geom/geom_qbvh_volume_all.h
+++ b/intern/cycles/kernel/geom/geom_qbvh_volume_all.h
@@ -98,7 +98,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			/* Traverse internal nodes. */
 			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
 #ifdef __VISIBILITY_FLAG__
-				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_QNODE_SIZE+0);
+				float4 inodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+0);
 				if((__float_as_uint(inodes.x) & visibility) == 0) {
 					/* Pop. */
 					nodeAddr = traversalStack[stackPtr].addr;
@@ -123,8 +123,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 				                                        &dist);
 
 				if(traverseChild != 0) {
-					float4 cnodes = kernel_tex_fetch(__bvh_nodes,
-					                                 nodeAddr*BVH_QNODE_SIZE+7);
+					float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr+7);
 
 					/* One child is hit, continue with that child. */
 					int r = __bscf(traverseChild);
@@ -217,7 +216,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 
 			/* If node is leaf, fetch triangle list. */
 			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1)*BVH_QNODE_LEAF_SIZE);
+				float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-nodeAddr-1));
 				int primAddr = __float_as_int(leaf.x);
 
 #if BVH_FEATURE(BVH_INSTANCING)