Cycles: fix some update issues with camera motion blur, and do some more work

for getting object motion blur ready.

19 files changed, 285 insertions, 875 deletions

diff --git a/intern/cycles/blender/blender_object.cpp b/intern/cycles/blender/blender_object.cpp
index 3d74c57288b..e10ffb3cf98 100644
--- a/intern/cycles/blender/blender_object.cpp
+++ b/intern/cycles/blender/blender_object.cpp
@@ -396,6 +396,8 @@ void BlenderSync::sync_motion(BL::SpaceView3D b_v3d, BL::Object b_override)
 	if(b_override)
 		b_cam = b_override;
 
+	Camera prevcam = *(scene->camera);
+	
 	/* go back and forth one frame */
 	int frame = b_scene.frame_current();
 
@@ -411,6 +413,10 @@ void BlenderSync::sync_motion(BL::SpaceView3D b_v3d, BL::Object b_override)
 	}
 
 	scene_frame_set(b_scene, frame);
+
+	/* tag camera for motion update */
+	if(scene->camera->motion_modified(prevcam))
+		scene->camera->tag_update();
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/blender/blender_sync.cpp b/intern/cycles/blender/blender_sync.cpp
index 00130f357dd..6c63872333d 100644
--- a/intern/cycles/blender/blender_sync.cpp
+++ b/intern/cycles/blender/blender_sync.cpp
@@ -17,6 +17,7 @@
  */
 
 #include "background.h"
+#include "camera.h"
 #include "film.h"
 #include "../render/filter.h"
 #include "graph.h"
@@ -179,6 +180,11 @@ void BlenderSync::sync_integrator()
 
 	integrator->sample_clamp = get_float(cscene, "sample_clamp");
 #ifdef __CAMERA_MOTION__
+	if(integrator->motion_blur != r.use_motion_blur()) {
+		scene->object_manager->tag_update(scene);
+		scene->camera->tag_update();
+	}
+
 	integrator->motion_blur = (!preview && r.use_motion_blur());
 #endif
 
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index df8a9b1d5b4..ad1ce1df295 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -31,13 +31,11 @@ set(SRC_HEADERS
 	kernel_globals.h
 	kernel_light.h
 	kernel_math.h
-	kernel_mbvh.h
 	kernel_montecarlo.h
 	kernel_object.h
 	kernel_passes.h
 	kernel_path.h
 	kernel_projection.h
-	kernel_qbvh.h
 	kernel_random.h
 	kernel_shader.h
 	kernel_textures.h
diff --git a/intern/cycles/kernel/kernel_bvh.h b/intern/cycles/kernel/kernel_bvh.h
index 90aec2e46b3..9d8ad6f3072 100644
--- a/intern/cycles/kernel/kernel_bvh.h
+++ b/intern/cycles/kernel/kernel_bvh.h
@@ -57,7 +57,7 @@ __device_inline float3 bvh_inverse_direction(float3 dir)
 
 __device_inline void bvh_instance_push(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, const float tmax)
 {
-	Transform tfm = object_fetch_transform(kg, object, ray->time, OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
 
 	*P = transform_point(&tfm, ray->P);
 
@@ -75,7 +75,7 @@ __device_inline void bvh_instance_push(KernelGlobals *kg, int object, const Ray
 __device_inline void bvh_instance_pop(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, const float tmax)
 {
 	if(*t != FLT_MAX) {
-		Transform tfm = object_fetch_transform(kg, object, ray->time, OBJECT_TRANSFORM);
+		Transform tfm = object_fetch_transform(kg, object, OBJECT_TRANSFORM);
 		*t *= len(transform_direction(&tfm, 1.0f/(*idir)));
 	}
 
@@ -83,6 +83,36 @@ __device_inline void bvh_instance_pop(KernelGlobals *kg, int object, const Ray *
 	*idir = bvh_inverse_direction(ray->D);
 }
 
+#ifdef __OBJECT_MOTION__
+__device_inline void bvh_instance_motion_push(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, Transform *tfm, const float tmax)
+{
+	Transform itfm;
+	*tfm = object_fetch_transform_motion(kg, object, ray->time, &itfm);
+
+	*P = transform_point(&itfm, ray->P);
+
+	float3 dir = transform_direction(&itfm, ray->D);
+
+	float len;
+	dir = normalize_len(dir, &len);
+
+	*idir = bvh_inverse_direction(dir);
+
+	if(*t != FLT_MAX)
+		*t *= len;
+}
+
+__device_inline void bvh_instance_motion_pop(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, Transform *tfm, const float tmax)
+{
+	if(*t != FLT_MAX) {
+		*t *= len(transform_direction(tfm, 1.0f/(*idir)));
+	}
+
+	*P = ray->P;
+	*idir = bvh_inverse_direction(ray->D);
+}
+#endif
+
 /* intersect two bounding boxes */
 __device_inline void bvh_node_intersect(KernelGlobals *kg,
 	bool *traverseChild0, bool *traverseChild1,
@@ -133,7 +163,7 @@ __device_inline void bvh_node_intersect(KernelGlobals *kg,
 
 /* Sven Woop's algorithm */
 __device_inline void bvh_triangle_intersect(KernelGlobals *kg, Intersection *isect,
-	float3 P, float3 idir, uint visibility, int object, int triAddr)
+	float3 P, float3 idir, uint visibility, int object, int triAddr, Transform *tfm)
 {
 	/* compute and check intersection t-value */
 	float4 v00 = kernel_tex_fetch(__tri_woop, triAddr*TRI_NODE_SIZE+0);
@@ -176,7 +206,7 @@ __device_inline void bvh_triangle_intersect(KernelGlobals *kg, Intersection *ise
 	}
 }
 
-__device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const uint visibility, Intersection *isect)
+__device_inline bool bvh_intersect(KernelGlobals *kg, const Ray *ray, const uint visibility, Intersection *isect)
 {
 	/* traversal stack in CUDA thread-local memory */
 	int traversalStack[BVH_STACK_SIZE];
@@ -255,7 +285,7 @@ __device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const ui
 					/* triangle intersection */
 					while(primAddr < primAddr2) {
 						/* intersect ray against triangle */
-						bvh_triangle_intersect(kg, isect, P, idir, visibility, object, primAddr);
+						bvh_triangle_intersect(kg, isect, P, idir, visibility, object, primAddr, NULL);
 
 						/* shadow ray early termination */
 						if(visibility == PATH_RAY_SHADOW_OPAQUE && isect->prim != ~0)
@@ -268,7 +298,6 @@ __device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const ui
 				else {
 					/* instance push */
 					object = kernel_tex_fetch(__prim_object, -primAddr-1);
-
 					bvh_instance_push(kg, object, ray, &P, &idir, &isect->t, tmax);
 
 					++stackPtr;
@@ -296,6 +325,133 @@ __device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const ui
 	return (isect->prim != ~0);
 }
 
+#ifdef __OBJECT_MOTION__
+__device_inline bool bvh_intersect_motion(KernelGlobals *kg, const Ray *ray, const uint visibility, Intersection *isect)
+{
+	/* traversal stack in CUDA thread-local memory */
+	int traversalStack[BVH_STACK_SIZE];
+	traversalStack[0] = ENTRYPOINT_SENTINEL;
+
+	/* traversal variables in registers */
+	int stackPtr = 0;
+	int nodeAddr = kernel_data.bvh.root;
+
+	/* ray parameters in registers */
+	const float tmax = ray->t;
+	float3 P = ray->P;
+	float3 idir = bvh_inverse_direction(ray->D);
+	int object = ~0;
+
+	Transform ob_tfm;
+
+	isect->t = tmax;
+	isect->object = ~0;
+	isect->prim = ~0;
+	isect->u = 0.0f;
+	isect->v = 0.0f;
+
+	/* traversal loop */
+	do {
+		do
+		{
+			/* traverse internal nodes */
+			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL)
+			{
+				bool traverseChild0, traverseChild1, closestChild1;
+				int nodeAddrChild1;
+
+				bvh_node_intersect(kg, &traverseChild0, &traverseChild1,
+					&closestChild1, &nodeAddr, &nodeAddrChild1,
+					P, idir, isect->t, visibility, nodeAddr);
+
+				if(traverseChild0 != traverseChild1) {
+					/* one child was intersected */
+					if(traverseChild1) {
+						nodeAddr = nodeAddrChild1;
+					}
+				}
+				else {
+					if(!traverseChild0) {
+						/* neither child was intersected */
+						nodeAddr = traversalStack[stackPtr];
+						--stackPtr;
+					}
+					else {
+						/* both children were intersected, push the farther one */
+						if(closestChild1) {
+							int tmp = nodeAddr;
+							nodeAddr = nodeAddrChild1;
+							nodeAddrChild1 = tmp;
+						}
+
+						++stackPtr;
+						traversalStack[stackPtr] = nodeAddrChild1;
+					}
+				}
+			}
+
+			/* if node is leaf, fetch triangle list */
+			if(nodeAddr < 0) {
+				float4 leaf = kernel_tex_fetch(__bvh_nodes, (-nodeAddr-1)*BVH_NODE_SIZE+(BVH_NODE_SIZE-1));
+				int primAddr = __float_as_int(leaf.x);
+
+				if(primAddr >= 0) {
+					int primAddr2 = __float_as_int(leaf.y);
+
+					/* pop */
+					nodeAddr = traversalStack[stackPtr];
+					--stackPtr;
+
+					/* triangle intersection */
+					while(primAddr < primAddr2) {
+						/* intersect ray against triangle */
+						bvh_triangle_intersect(kg, isect, P, idir, visibility, object, primAddr, &ob_tfm);
+
+						/* shadow ray early termination */
+						if(visibility == PATH_RAY_SHADOW_OPAQUE && isect->prim != ~0)
+							return true;
+
+						primAddr++;
+					}
+				}
+				else {
+					/* instance push */
+					object = kernel_tex_fetch(__prim_object, -primAddr-1);
+					bvh_instance_motion_push(kg, object, ray, &P, &idir, &isect->t, &ob_tfm, tmax);
+
+					++stackPtr;
+					traversalStack[stackPtr] = ENTRYPOINT_SENTINEL;
+
+					nodeAddr = kernel_tex_fetch(__object_node, object);
+				}
+			}
+		} while(nodeAddr != ENTRYPOINT_SENTINEL);
+
+		if(stackPtr >= 0) {
+			kernel_assert(object != ~0);
+
+			/* instance pop */
+			bvh_instance_motion_pop(kg, object, ray, &P, &idir, &isect->t, &ob_tfm, tmax);
+			object = ~0;
+			nodeAddr = traversalStack[stackPtr];
+			--stackPtr;
+		}
+	} while(nodeAddr != ENTRYPOINT_SENTINEL);
+
+	return (isect->prim != ~0);
+}
+#endif
+
+__device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const uint visibility, Intersection *isect)
+{
+#ifdef __OBJECT_MOTION__
+	if(kernel_data.bvh.have_motion)
+		return bvh_intersect_motion(kg, ray, visibility, isect);
+	else
+#endif
+		return bvh_intersect(kg, ray, visibility, isect);
+}
+
 __device_inline float3 ray_offset(float3 P, float3 Ng)
 {
 #ifdef __INTERSECTION_REFINE__
@@ -352,7 +508,7 @@ __device_inline float3 bvh_triangle_refine(KernelGlobals *kg, ShaderData *sd, co
 #ifdef __OBJECT_MOTION__
 		Transform tfm = sd->ob_itfm;
 #else
-		Transform tfm = object_fetch_transform(kg, isect->object, ray->time, OBJECT_INVERSE_TRANSFORM);
+		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
 #endif
 
 		P = transform_point(&tfm, P);
@@ -373,7 +529,7 @@ __device_inline float3 bvh_triangle_refine(KernelGlobals *kg, ShaderData *sd, co
 #ifdef __OBJECT_MOTION__
 		Transform tfm = sd->ob_tfm;
 #else
-		Transform tfm = object_fetch_transform(kg, isect->object, ray->time, OBJECT_TRANSFORM);
+		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
 #endif
 
 		P = transform_point(&tfm, P);
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 1084415d0cf..4bb17c0bd5a 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -301,8 +301,13 @@ __device void triangle_light_sample(KernelGlobals *kg, int prim, int object,
 #ifdef __INSTANCING__
 	/* instance transform */
 	if(ls->object >= 0) {
-		Transform tfm = object_fetch_transform(kg, ls->object, time, OBJECT_TRANSFORM);
-		Transform itfm = object_fetch_transform(kg, ls->object, time, OBJECT_INVERSE_TRANSFORM);
+#ifdef __OBJECT_MOTION__
+		Transform itfm;
+		Transform tfm = object_fetch_transform_motion(kg, ls->object, time, &itfm);
+#else
+		Transform tfm = object_fetch_transform(kg, ls->object, OBJECT_TRANSFORM);
+		Transform itfm = object_fetch_transform(kg, ls->object, OBJECT_INVERSE_TRANSFORM);
+#endif
 
 		ls->P = transform_point(&tfm, ls->P);
 		ls->Ng = normalize(transform_direction_transposed(&itfm, ls->Ng));
diff --git a/intern/cycles/kernel/kernel_mbvh.h b/intern/cycles/kernel/kernel_mbvh.h
deleted file mode 100644
index ccbd3d069b4..00000000000
--- a/intern/cycles/kernel/kernel_mbvh.h
+++ /dev/null
@@ -1,394 +0,0 @@
-/*
- * Copyright 2011, Blender Foundation.
- *
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
- */
-
-CCL_NAMESPACE_BEGIN
-
-#define MBVH_OBJECT_SENTINEL 0x76543210
-#define MBVH_NODE_SIZE 8
-#define MBVH_STACK_SIZE 1024
-#define MBVH_RAY_STACK_SIZE 10000
-
-typedef struct MBVHTask {
-	int node;
-	int index;
-	int num;
-	int object;
-} MBVHTask;
-
-typedef struct MVBHRay {
-	float3 P;
-	float u;
-	float3 idir;
-	float v;
-	float t;
-	int index;
-	int object;
-
-	float3 origP;
-	float3 origD;
-	float tmax;
-} MBVHRay;
-
-__device float3 mbvh_inverse_direction(float3 dir)
-{
-	// Avoid divide by zero (ooeps = exp2f(-80.0f))
-	float ooeps = 0.00000000000000000000000082718061255302767487140869206996285356581211090087890625f;
-	float3 idir;
-
-	idir.x = 1.0f / (fabsf(dir.x) > ooeps ? dir.x : copysignf(ooeps, dir.x));
-	idir.y = 1.0f / (fabsf(dir.y) > ooeps ? dir.y : copysignf(ooeps, dir.y));
-	idir.z = 1.0f / (fabsf(dir.z) > ooeps ? dir.z : copysignf(ooeps, dir.z));
-
-	return idir;
-}
-
-__device void mbvh_instance_push(KernelGlobals *kg, int object, MBVHRay *ray)
-{
-	Transform tfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
-
-	ray->P = transform_point(&tfm, ray->origP);
-
-	float3 dir = ray->origD;
-
-	if(ray->t != ray->tmax) dir *= ray->t;
-
-	dir = transform_direction(&tfm, dir);
-	ray->idir = mbvh_inverse_direction(normalize(dir));
-
-	if(ray->t != ray->tmax) ray->t = len(dir);
-}
-
-__device void mbvh_instance_pop(KernelGlobals *kg, int object, MBVHRay *ray)
-{
-	Transform tfm = object_fetch_transform(kg, object, OBJECT_TRANSFORM);
-
-	if(ray->t != ray->tmax)
-		ray->t = len(transform_direction(&tfm, (1.0f/(ray->idir)) * (ray->t)));
-
-	ray->P = ray->origP;
-	ray->idir = mbvh_inverse_direction(ray->origD);
-}
-
-/* Sven Woop's algorithm */
-__device void mbvh_triangle_intersect(KernelGlobals *kg, MBVHRay *ray, int object, int triAddr)
-{
-	float3 P = ray->P;
-	float3 idir = ray->idir;
-
-	/* compute and check intersection t-value */
-	float4 v00 = kernel_tex_fetch(__tri_woop, triAddr*MBVH_NODE_SIZE+0);
-	float4 v11 = kernel_tex_fetch(__tri_woop, triAddr*MBVH_NODE_SIZE+1);
-	float3 dir = 1.0f/idir;
-
-	float Oz = v00.w - P.x*v00.x - P.y*v00.y - P.z*v00.z;
-	float invDz = 1.0f/(dir.x*v00.x + dir.y*v00.y + dir.z*v00.z);
-	float t = Oz * invDz;
-
-	if(t > 0.0f && t < ray->t) {
-		/* compute and check barycentric u */
-		float Ox = v11.w + P.x*v11.x + P.y*v11.y + P.z*v11.z;
-		float Dx = dir.x*v11.x + dir.y*v11.y + dir.z*v11.z;
-		float u = Ox + t*Dx;
-
-		if(u >= 0.0f) {
-			/* compute and check barycentric v */
-			float4 v22 = kernel_tex_fetch(__tri_woop, triAddr*MBVH_NODE_SIZE+2);
-			float Oy = v22.w + P.x*v22.x + P.y*v22.y + P.z*v22.z;
-			float Dy = dir.x*v22.x + dir.y*v22.y + dir.z*v22.z;
-			float v = Oy + t*Dy;
-
-			if(v >= 0.0f && u + v <= 1.0f) {
-				/* record intersection */
-				ray->index = triAddr;
-				ray->object = object;
-				ray->u = u;
-				ray->v = v;
-				ray->t = t;
-			}
-		}
-	}
-}
-
-__device void mbvh_node_intersect(KernelGlobals *kg, __m128 *traverseChild,
-	__m128 *tHit, float3 P, float3 idir, float t, int nodeAddr)
-{
-	/* X axis */
-	const __m128 bminx = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+0);
-	const __m128 t0x = _mm_mul_ps(_mm_sub_ps(bminx, _mm_set_ps1(P.x)), _mm_set_ps1(idir.x));
-	const __m128 bmaxx = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+1);
-	const __m128 t1x = _mm_mul_ps(_mm_sub_ps(bmaxx, _mm_set_ps1(P.x)), _mm_set_ps1(idir.x));
-
-	__m128 tmin = _mm_max_ps(_mm_min_ps(t0x, t1x), _mm_setzero_ps());
-	__m128 tmax = _mm_min_ps(_mm_max_ps(t0x, t1x), _mm_set_ps1(t));
-
-	/* Y axis */
-	const __m128 bminy = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+2);
-	const __m128 t0y = _mm_mul_ps(_mm_sub_ps(bminy, _mm_set_ps1(P.y)), _mm_set_ps1(idir.y));
-	const __m128 bmaxy = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+3);
-	const __m128 t1y = _mm_mul_ps(_mm_sub_ps(bmaxy, _mm_set_ps1(P.y)), _mm_set_ps1(idir.y));
-
-	tmin = _mm_max_ps(_mm_min_ps(t0y, t1y), tmin);
-	tmax = _mm_min_ps(_mm_max_ps(t0y, t1y), tmax);
-
-	/* Z axis */
-	const __m128 bminz = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+4);
-	const __m128 t0z = _mm_mul_ps(_mm_sub_ps(bminz, _mm_set_ps1(P.z)), _mm_set_ps1(idir.z));
-	const __m128 bmaxz = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*MBVH_NODE_SIZE+5);
-	const __m128 t1z = _mm_mul_ps(_mm_sub_ps(bmaxz, _mm_set_ps1(P.z)), _mm_set_ps1(idir.z));
-
-	tmin = _mm_max_ps(_mm_min_ps(t0z, t1z), tmin);
-	tmax = _mm_min_ps(_mm_max_ps(t0z, t1z), tmax);
-
-	/* compare and get mask */
-	*traverseChild = _mm_cmple_ps(tmin, tmax);
-
-	/* get distance XXX probably wrong */
-	*tHit = tmin;
-}
-
-static void mbvh_sort_by_length(int id[4], float len[4])
-{
-	for(int i = 1; i < 4; i++) {
-		int j = i - 1;
-
-		while(j >= 0 && len[j] > len[j+1]) {
-			swap(len[j], len[j+1]);
-			swap(id[j], id[j+1]);
-			j--;
-		}
-	}
-}
-
-__device void scene_intersect(KernelGlobals *kg, MBVHRay *rays, int numrays)
-{
-	/* traversal stacks */
-	MBVHTask task_stack[MBVH_STACK_SIZE];
-	int active_ray_stacks[4][MBVH_RAY_STACK_SIZE];
-	int num_task, num_active[4] = {0, 0, 0, 0};
-	__m128i one_mm = _mm_set1_epi32(1);
-
-	/* push root node task on stack */
-	task_stack[0].node = kernel_data.bvh.root;
-	task_stack[0].index = 0;
-	task_stack[0].num = numrays;
-	task_stack[0].object = ~0;
-	num_task = 1;
-
-	/* push all rays in first SIMD lane */
-	for(int i = 0; i < numrays; i++)
-		active_ray_stacks[0][i] = i;
-	num_active[0] = numrays;
-	
-	while(num_task >= 1) {
-		/* pop task */
-		MBVHTask task = task_stack[--num_task];
-
-		if(task.node == MBVH_OBJECT_SENTINEL) {
-			/* instance pop */
-
-			/* pop rays from stack */
-			num_active[task.index] -= task.num;
-			int ray_offset = num_active[task.index];
-
-			/* transform rays */
-			for(int i = 0; i < task.num; i++) {
-				MBVHRay *ray = &rays[active_ray_stacks[task.index][ray_offset + i]];
-				mbvh_instance_pop(kg, task.object, ray);
-			}
-		}
-		else if(task.node >= 0) {
-			/* inner node? */
-
-			/* pop rays from stack*/
-			num_active[task.index] -= task.num;
-			int ray_offset = num_active[task.index];
-
-			/* initialze simd values */
-			__m128i num_active_mm = _mm_load_si128((__m128i*)num_active);
-			__m128 len_mm = _mm_set_ps1(0.0f);
-
-			for(int i = 0; i < task.num; i++) {
-				int rayid = active_ray_stacks[task.index][ray_offset + i];
-				MVBHRay *ray = rays + rayid;
-
-				/* intersect 4 QBVH node children */
-				__m128 result;
-				__m128 thit;
-
-				mbvh_node_intersect(kg, &result, &thit, ray->P, ray->idir, ray->t, task.node);
-
-				/* update length for sorting */
-				len_mm = _mm_add_ps(len_mm, _mm_and_ps(thit, result));
-
-				/* push rays on stack */
-				for(int j = 0; j < 4; j++)
-					active_ray_stacks[j][num_active[j]] = rayid;
-
-				/* update num active */
-				__m128i resulti = _mm_and_si128(*((__m128i*)&result), one_mm);
-				num_active_mm = _mm_add_epi32(resulti, num_active_mm);
-				_mm_store_si128((__m128i*)num_active, num_active_mm);
-			}
-
-			if(num_active[0] || num_active[1] || num_active[2] || num_active[3]) {
-				/* load child node addresses */
-				float4 cnodes = kernel_tex_fetch(__bvh_nodes, task.node);
-				int child[4] = {
-					__float_as_int(cnodes.x),
-					__float_as_int(cnodes.y),
-					__float_as_int(cnodes.z),
-					__float_as_int(cnodes.w)};
-
-				/* sort nodes by average intersection distance */
-				int ids[4] = {0, 1, 2, 3};
-				float len[4];
-
-				_mm_store_ps(len, len_mm);
-				mbvh_sort_by_length(ids, len);
-
-				/* push new tasks on stack */
-				for(int j = 0; j < 4; j++) {
-					if(num_active[j]) {
-						int id = ids[j];
-
-						task_stack[num_task].node = child[id];
-						task_stack[num_task].index = id;
-						task_stack[num_task].num = num_active[id];
-						task_stack[num_task].object = task.object;
-						num_task++;
-					}
-				}
-			}
-		}
-		else {
-			/* fetch leaf node data */
-			float4 leaf = kernel_tex_fetch(__bvh_nodes, (-task.node-1)*MBVH_NODE_SIZE+(MBVH_NODE_SIZE-2));
-			int triAddr = __float_as_int(leaf.x);
-			int triAddr2 = __float_as_int(leaf.y);
-
-			/* pop rays from stack*/
-			num_active[task.index] -= task.num;
-			int ray_offset = num_active[task.index];
-
-			/* triangles */
-			if(triAddr >= 0) {
-				int i, numq = (task.num >> 2) << 2;
-
-				/* SIMD ray leaf intersection */
-				for(i = 0; i < numq; i += 4) {
-					MBVHRay *ray4[4] = {
-						&rays[active_ray_stacks[task.index][ray_offset + i + 0]],
-						&rays[active_ray_stacks[task.index][ray_offset + i + 1]],
-						&rays[active_ray_stacks[task.index][ray_offset + i + 2]],
-						&rays[active_ray_stacks[task.index][ray_offset + i + 3]]};
-
-					/* load SoA */
-
-					while(triAddr < triAddr2) {
-						mbvh_triangle_intersect(ray4[0], task.object, task.node);
-						mbvh_triangle_intersect(ray4[1], task.object, task.node);
-						mbvh_triangle_intersect(ray4[2], task.object, task.node);
-						mbvh_triangle_intersect(ray4[3], task.object, task.node);
-						triAddr++;
-
-						/* some shadow ray optim could be done by setting t=0 */
-					}
-
-					/* store AoS */
-				}
-
-				/* mono ray leaf intersection */
-				for(; i < task.num; i++) {
-					MBVHRay *ray = &rays[active_ray_stacks[task.index][ray_offset + i]];
-
-					while(triAddr < triAddr2) {
-						mbvh_triangle_intersect(kg, ray, task.object, task.node);
-						triAddr++;
-					}
-				}
-			}
-			else {
-				/* instance push */
-				int object = -triAddr-1;
-				int node = triAddr;
-
-				/* push instance pop task */
-				task_stack[num_task].node = MBVH_OBJECT_SENTINEL;
-				task_stack[num_task].index = task.index;
-				task_stack[num_task].num = task.num;
-				task_stack[num_task].object = object;
-				num_task++;
-
-				num_active[task.index] += task.num;
-
-				/* push node task */
-				task_stack[num_task].node = node;
-				task_stack[num_task].index = task.index;
-				task_stack[num_task].num = task.num;
-				task_stack[num_task].object = object;
-				num_task++;
-
-				for(int i = 0; i < task.num; i++) {
-					int rayid = active_ray_stacks[task.index][ray_offset + i];
-
-					/* push on stack for last task */
-					active_ray_stacks[task.index][num_active[task.index]] = rayid;
-					num_active[task.index]++;
-
-					/* transform ray */
-					MBVHRay *ray = &rays[rayid];
-					mbvh_instance_push(kg, object, ray);
-				}
-			}
-		}
-	}
-}
-
-__device void mbvh_set_ray(MBVHRay *rays, int i, Ray *ray, float tmax)
-{
-	MBVHRay *mray = &rays[i];
-
-	/* ray parameters in registers */
-	mray->P = ray->P;
-	mray->idir = mbvh_inverse_direction(ray->D);
-	mray->t = tmax;
-}
-
-__device bool mbvh_get_intersection(MVBHRay *rays, int i, Intersection *isect, float tmax)
-{
-	MBVHRay *mray = &rays[i];
-
-	if(mray->t == tmax)
-		return false;
-	
-	isect->t = mray->t;
-	isect->u = mray->u;
-	isect->v = mray->v;
-	isect->index = mray->index;
-	isect->object = mray->object;
-
-	return true;
-}
-
-__device bool mbvh_get_shadow(MBVHRay *rays, int i, float tmax)
-{
-	return (rays[i].t == tmax);
-}
-
-CCL_NAMESPACE_END
-
diff --git a/intern/cycles/kernel/kernel_object.h b/intern/cycles/kernel/kernel_object.h
index ad43120146a..d8ea2cf9926 100644
--- a/intern/cycles/kernel/kernel_object.h
+++ b/intern/cycles/kernel/kernel_object.h
@@ -27,53 +27,67 @@ enum ObjectTransform {
 	OBJECT_DUPLI = 16
 };
 
-__device_inline Transform object_fetch_transform(KernelGlobals *kg, int object, float time, enum ObjectTransform type)
+__device_inline Transform object_fetch_transform(KernelGlobals *kg, int object, enum ObjectTransform type)
 {
+	int offset = object*OBJECT_SIZE + (int)type;
+
 	Transform tfm;
+	tfm.x = kernel_tex_fetch(__objects, offset + 0);
+	tfm.y = kernel_tex_fetch(__objects, offset + 1);
+	tfm.z = kernel_tex_fetch(__objects, offset + 2);
+	tfm.w = make_float4(0.0f, 0.0f, 0.0f, 1.0f);
+
+	return tfm;
+}
 
 #ifdef __OBJECT_MOTION__
+__device_inline Transform object_fetch_transform_motion(KernelGlobals *kg, int object, float time, Transform *itfm)
+{
+	Transform tfm;
+
+	int object_flag = kernel_tex_fetch(__object_flag, object);
+
 	/* if we do motion blur */
-	if(sd->flag & SD_OBJECT_MOTION) {
+	if(object_flag & SD_OBJECT_MOTION) {
 		/* fetch motion transforms */
 		MotionTransform motion;
 
-		motion.pre.x = have_motion;
+		int offset = object*OBJECT_SIZE + (int)OBJECT_TRANSFORM_MOTION_PRE;
+
+		motion.pre.x = kernel_tex_fetch(__objects, offset + 0);
 		motion.pre.y = kernel_tex_fetch(__objects, offset + 1);
 		motion.pre.z = kernel_tex_fetch(__objects, offset + 2);
 		motion.pre.w = kernel_tex_fetch(__objects, offset + 3);
 
+
 		motion.post.x = kernel_tex_fetch(__objects, offset + 4);
 		motion.post.y = kernel_tex_fetch(__objects, offset + 5);
 		motion.post.z = kernel_tex_fetch(__objects, offset + 6);
 		motion.post.w = kernel_tex_fetch(__objects, offset + 7);
 
-		/* interpolate (todo: do only once per object) */
 		transform_motion_interpolate(&tfm, &motion, time);
 
 		/* invert */
-		if(type == OBJECT_INVERSE_TRANSFORM)
-			tfm = transform_quick_inverse(tfm);
-
-		return tfm;
+		if(itfm)
+			*itfm = transform_quick_inverse(tfm);
 	}
-#endif
-
-	int offset = object*OBJECT_SIZE + (int)type;
+	else {
+		tfm = object_fetch_transform(kg, object, OBJECT_TRANSFORM);
 
-	tfm.x = kernel_tex_fetch(__objects, offset + 0);
-	tfm.y = kernel_tex_fetch(__objects, offset + 1);
-	tfm.z = kernel_tex_fetch(__objects, offset + 2);
-	tfm.w = make_float4(0.0f, 0.0f, 0.0f, 1.0f);
+		if(itfm)
+			*itfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
+	}
 
 	return tfm;
 }
+#endif
 
 __device_inline void object_position_transform(KernelGlobals *kg, ShaderData *sd, float3 *P)
 {
 #ifdef __OBJECT_MOTION__
 	*P = transform_point(&sd->ob_tfm, *P);
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	*P = transform_point(&tfm, *P);
 #endif
 }
@@ -83,7 +97,7 @@ __device_inline void object_inverse_position_transform(KernelGlobals *kg, Shader
 #ifdef __OBJECT_MOTION__
 	*P = transform_point(&sd->ob_itfm, *P);
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	*P = transform_point(&tfm, *P);
 #endif
 }
@@ -93,7 +107,7 @@ __device_inline void object_inverse_normal_transform(KernelGlobals *kg, ShaderDa
 #ifdef __OBJECT_MOTION__
 	*N = normalize(transform_direction_transposed(&sd->ob_tfm, *N));
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	*N = normalize(transform_direction_transposed(&tfm, *N));
 #endif
 }
@@ -103,7 +117,7 @@ __device_inline void object_normal_transform(KernelGlobals *kg, ShaderData *sd,
 #ifdef __OBJECT_MOTION__
 	*N = normalize(transform_direction_transposed(&sd->ob_itfm, *N));
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	*N = normalize(transform_direction_transposed(&tfm, *N));
 #endif
 }
@@ -113,7 +127,7 @@ __device_inline void object_dir_transform(KernelGlobals *kg, ShaderData *sd, flo
 #ifdef __OBJECT_MOTION__
 	*D = transform_direction(&sd->ob_tfm, *D);
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, 0.0f, OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	*D = transform_direction(&tfm, *D);
 #endif
 }
@@ -123,7 +137,7 @@ __device_inline float3 object_location(KernelGlobals *kg, ShaderData *sd)
 #ifdef __OBJECT_MOTION__
 	return make_float3(sd->ob_tfm.x.w, sd->ob_tfm.y.w, sd->ob_tfm.z.w);
 #else
-	Transform tfm = object_fetch_transform(kg, sd->object, 0.0f, OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	return make_float3(tfm.x.w, tfm.y.w, tfm.z.w);
 #endif
 }
diff --git a/intern/cycles/kernel/kernel_qbvh.h b/intern/cycles/kernel/kernel_qbvh.h
deleted file mode 100644
index 525b616921d..00000000000
--- a/intern/cycles/kernel/kernel_qbvh.h
+++ /dev/null
@@ -1,413 +0,0 @@
-/*
- * Adapted from code Copyright 2009-2010 NVIDIA Corporation
- * Modifications Copyright 2011, Blender Foundation.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-CCL_NAMESPACE_BEGIN
-
-/*
- * "Persistent while-while kernel" used in:
- *
- * "Understanding the Efficiency of Ray Traversal on GPUs",
- * Timo Aila and Samuli Laine,
- * Proc. High-Performance Graphics 2009
- */
-
-/* bottom-most stack entry, indicating the end of traversal */
-
-#define ENTRYPOINT_SENTINEL 0x76543210
-/* 64 object BVH + 64 mesh BVH + 64 object node splitting */
-#define QBVH_STACK_SIZE 192
-#define QBVH_NODE_SIZE 8
-#define TRI_NODE_SIZE 3
-
-__device_inline float3 qbvh_inverse_direction(float3 dir)
-{
-	// Avoid divide by zero (ooeps = exp2f(-80.0f))
-	float ooeps = 0.00000000000000000000000082718061255302767487140869206996285356581211090087890625f;
-	float3 idir;
-
-	idir.x = 1.0f/((fabsf(dir.x) > ooeps)? dir.x: copysignf(ooeps, dir.x));
-	idir.y = 1.0f/((fabsf(dir.y) > ooeps)? dir.y: copysignf(ooeps, dir.y));
-	idir.z = 1.0f/((fabsf(dir.z) > ooeps)? dir.z: copysignf(ooeps, dir.z));
-
-	return idir;
-}
-
-__device_inline void qbvh_instance_push(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, const float tmax)
-{
-	Transform tfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
-
-	*P = transform_point(&tfm, ray->P);
-
-	float3 dir = transform_direction(&tfm, ray->D);
-
-	float len;
-	dir = normalize_len(dir, &len);
-
-	*idir = qbvh_inverse_direction(dir);
-
-	if(*t != FLT_MAX)
-		*t *= len;
-}
-
-__device_inline void qbvh_instance_pop(KernelGlobals *kg, int object, const Ray *ray, float3 *P, float3 *idir, float *t, const float tmax)
-{
-	Transform tfm = object_fetch_transform(kg, object, OBJECT_TRANSFORM);
-
-	if(*t != FLT_MAX)
-		*t *= len(transform_direction(&tfm, 1.0f/(*idir)));
-
-	*P = ray->P;
-	*idir = qbvh_inverse_direction(ray->D);
-}
-
-#ifdef __KERNEL_CPU__
-
-__device_inline void qbvh_node_intersect(KernelGlobals *kg, int *traverseChild,
-	int nodeAddrChild[4], float3 P, float3 idir, float t, int nodeAddr)
-{
-	/* X axis */
-	const __m128 bminx = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+0);
-	const __m128 t0x = _mm_mul_ps(_mm_sub_ps(bminx, _mm_set_ps1(P.x)), _mm_set_ps1(idir.x));
-	const __m128 bmaxx = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+1);
-	const __m128 t1x = _mm_mul_ps(_mm_sub_ps(bmaxx, _mm_set_ps1(P.x)), _mm_set_ps1(idir.x));
-
-	__m128 tmin = _mm_max_ps(_mm_min_ps(t0x, t1x), _mm_setzero_ps());
-	__m128 tmax = _mm_min_ps(_mm_max_ps(t0x, t1x), _mm_set_ps1(t));
-
-	/* Y axis */
-	const __m128 bminy = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+2);
-	const __m128 t0y = _mm_mul_ps(_mm_sub_ps(bminy, _mm_set_ps1(P.y)), _mm_set_ps1(idir.y));
-	const __m128 bmaxy = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+3);
-	const __m128 t1y = _mm_mul_ps(_mm_sub_ps(bmaxy, _mm_set_ps1(P.y)), _mm_set_ps1(idir.y));
-
-	tmin = _mm_max_ps(_mm_min_ps(t0y, t1y), tmin);
-	tmax = _mm_min_ps(_mm_max_ps(t0y, t1y), tmax);
-
-	/* Z axis */
-	const __m128 bminz = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+4);
-	const __m128 t0z = _mm_mul_ps(_mm_sub_ps(bminz, _mm_set_ps1(P.z)), _mm_set_ps1(idir.z));
-	const __m128 bmaxz = kernel_tex_fetch_m128(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+5);
-	const __m128 t1z = _mm_mul_ps(_mm_sub_ps(bmaxz, _mm_set_ps1(P.z)), _mm_set_ps1(idir.z));
-
-	tmin = _mm_max_ps(_mm_min_ps(t0z, t1z), tmin);
-	tmax = _mm_min_ps(_mm_max_ps(t0z, t1z), tmax);
-
-	/* compare and get mask */
-	*traverseChild = _mm_movemask_ps(_mm_cmple_ps(tmin, tmax));
-
-	/* get node addresses */
-	float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+6);
-
-	nodeAddrChild[0] = __float_as_int(cnodes.x);
-	nodeAddrChild[1] = __float_as_int(cnodes.y);
-	nodeAddrChild[2] = __float_as_int(cnodes.z);
-	nodeAddrChild[3] = __float_as_int(cnodes.w);
-}
-
-#else
-
-__device_inline bool qbvh_bb_intersect(float3 bmin, float3 bmax, float3 P, float3 idir, float t)
-{
-	float t0x = (bmin.x - P.x)*idir.x;
-	float t1x = (bmax.x - P.x)*idir.x;
-	float t0y = (bmin.y - P.y)*idir.y;
-	float t1y = (bmax.y - P.y)*idir.y;
-	float t0z = (bmin.z - P.z)*idir.z;
-	float t1z = (bmax.z - P.z)*idir.z;
-
-	float minx = min(t0x, t1x);
-	float maxx = max(t0x, t1x);
-	float miny = min(t0y, t1y);
-	float maxy = max(t0y, t1y);
-	float minz = min(t0z, t1z);
-	float maxz = max(t0z, t1z);
-
-	float tmin = max4(0.0f, minx, miny, minz);
-	float tmax = min4(t, maxx, maxy, maxz);
-
-	return (tmin <= tmax);
-}
-
-/* intersect four bounding boxes */
-__device_inline void qbvh_node_intersect(KernelGlobals *kg, int *traverseChild,
-	int nodeAddrChild[4], float3 P, float3 idir, float t, int nodeAddr)
-{
-	/* fetch node data */
-	float4 minx = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+0);
-	float4 miny = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+2);
-	float4 minz = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+4);
-	float4 maxx = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+1);
-	float4 maxy = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+3);
-	float4 maxz = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+5);
-
-	/* intersect bounding boxes */
-	bool traverseChild0 = qbvh_bb_intersect(make_float3(minx.x, miny.x, minz.x), make_float3(maxx.x, maxy.x, maxz.x), P, idir, t);
-	bool traverseChild1 = qbvh_bb_intersect(make_float3(minx.y, miny.y, minz.y), make_float3(maxx.y, maxy.y, maxz.y), P, idir, t);
-	bool traverseChild2 = qbvh_bb_intersect(make_float3(minx.z, miny.z, minz.z), make_float3(maxx.z, maxy.z, maxz.z), P, idir, t);
-	bool traverseChild3 = qbvh_bb_intersect(make_float3(minx.w, miny.w, minz.w), make_float3(maxx.w, maxy.w, maxz.w), P, idir, t);
-
-	*traverseChild = 0;
-	if(traverseChild0) *traverseChild |= 1;
-	if(traverseChild1) *traverseChild |= 2;
-	if(traverseChild2) *traverseChild |= 4;
-	if(traverseChild3) *traverseChild |= 8;
-
-	/* get node addresses */
-	float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*QBVH_NODE_SIZE+6);
-
-	nodeAddrChild[0] = __float_as_int(cnodes.x);
-	nodeAddrChild[1] = __float_as_int(cnodes.y);
-	nodeAddrChild[2] = __float_as_int(cnodes.z);
-	nodeAddrChild[3] = __float_as_int(cnodes.w);
-}
-
-#endif
-
-/* Sven Woop's algorithm */
-__device_inline void qbvh_triangle_intersect(KernelGlobals *kg, Intersection *isect, float3 P, float3 idir, int object, int triAddr)
-{
-	/* compute and check intersection t-value */
-	float4 v00 = kernel_tex_fetch(__tri_woop, triAddr*TRI_NODE_SIZE+0);
-	float4 v11 = kernel_tex_fetch(__tri_woop, triAddr*TRI_NODE_SIZE+1);
-	float3 dir = 1.0f/idir;
-
-	float Oz = v00.w - P.x*v00.x - P.y*v00.y - P.z*v00.z;
-	float invDz = 1.0f/(dir.x*v00.x + dir.y*v00.y + dir.z*v00.z);
-	float t = Oz * invDz;
-
-	if(t > 0.0f && t < isect->t) {
-		/* compute and check barycentric u */
-		float Ox = v11.w + P.x*v11.x + P.y*v11.y + P.z*v11.z;
-		float Dx = dir.x*v11.x + dir.y*v11.y + dir.z*v11.z;
-		float u = Ox + t*Dx;
-
-		if(u >= 0.0f) {
-			/* compute and check barycentric v */
-			float4 v22 = kernel_tex_fetch(__tri_woop, triAddr*TRI_NODE_SIZE+2);
-			float Oy = v22.w + P.x*v22.x + P.y*v22.y + P.z*v22.z;
-			float Dy = dir.x*v22.x + dir.y*v22.y + dir.z*v22.z;
-			float v = Oy + t*Dy;
-
-			if(v >= 0.0f && u + v <= 1.0f) {
-				/* record intersection */
-				isect->prim = triAddr;
-				isect->object = object;
-				isect->u = u;
-				isect->v = v;
-				isect->t = t;
-			}
-		}
-	}
-}
-
-__device_inline bool scene_intersect(KernelGlobals *kg, const Ray *ray, const bool isshadowray, Intersection *isect)
-{
-	/* traversal stack in CUDA thread-local memory */
-	int traversalStack[QBVH_STACK_SIZE];
-	traversalStack[0] = ENTRYPOINT_SENTINEL;
-
-	/* traversal variables in registers */
-	int stackPtr = 0;
-	int nodeAddr = kernel_data.bvh.root;
-
-	/* ray parameters in registers */
-	const float tmax = ray->t;
-	float3 P = ray->P;
-	float3 idir = qbvh_inverse_direction(ray->D);
-	int object = ~0;
-
-	isect->t = tmax;
-	isect->object = ~0;
-	isect->prim = ~0;
-	isect->u = 0.0f;
-	isect->v = 0.0f;
-
-	/* traversal loop */
-	do {
-		do
-		{
-			/* traverse internal nodes */
-			while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL)
-			{
-				int traverseChild, nodeAddrChild[4];
-
-				qbvh_node_intersect(kg, &traverseChild, nodeAddrChild,
-					P, idir, isect->t, nodeAddr);
-
-				if(traverseChild & 1) {
-					++stackPtr;
-					traversalStack[stackPtr] = nodeAddrChild[0];
-				}
-
-				if(traverseChild & 2) {
-					++stackPtr;
-					traversalStack[stackPtr] = nodeAddrChild[1];
-				}
-				if(traverseChild & 4) {
-					++stackPtr;
-					traversalStack[stackPtr] = nodeAddrChild[2];
-				}
-
-				if(traverseChild & 8) {
-					++stackPtr;
-					traversalStack[stackPtr] = nodeAddrChild[3];
-				}
-
-				nodeAddr = traversalStack[stackPtr];
-				--stackPtr;
-			}
-
-			/* if node is leaf, fetch triangle list */
-			if(nodeAddr < 0) {
-				float4 leaf = kernel_tex_fetch(__bvh_nodes, (-nodeAddr-1)*QBVH_NODE_SIZE+(QBVH_NODE_SIZE-2));
-				int primAddr = __float_as_int(leaf.x);
-
-#ifdef __INSTANCING__
-				if(primAddr >= 0) {
-#endif
-					int primAddr2 = __float_as_int(leaf.y);
-
-					/* pop */
-					nodeAddr = traversalStack[stackPtr];
-					--stackPtr;
-
-					/* triangle intersection */
-					while(primAddr < primAddr2) {
-						/* intersect ray against triangle */
-						qbvh_triangle_intersect(kg, isect, P, idir, object, primAddr);
-
-						/* shadow ray early termination */
-						if(isshadowray && isect->prim != ~0)
-							return true;
-
-						primAddr++;
-					}
-#ifdef __INSTANCING__
-				}
-				else {
-					/* instance push */
-					object = kernel_tex_fetch(__prim_object, -primAddr-1);
-
-					qbvh_instance_push(kg, object, ray, &P, &idir, &isect->t, tmax);
-
-					++stackPtr;
-					traversalStack[stackPtr] = ENTRYPOINT_SENTINEL;
-
-					nodeAddr = kernel_tex_fetch(__object_node, object);
-				}
-#endif
-			}
-		} while(nodeAddr != ENTRYPOINT_SENTINEL);
-
-#ifdef __INSTANCING__
-		if(stackPtr >= 0) {
-			kernel_assert(object != ~0);
-
-			/* instance pop */
-			qbvh_instance_pop(kg, object, ray, &P, &idir, &isect->t, tmax);
-			object = ~0;
-			nodeAddr = traversalStack[stackPtr];
-			--stackPtr;
-		}
-#endif
-	} while(nodeAddr != ENTRYPOINT_SENTINEL);
-
-	return (isect->prim != ~0);
-}
-
-__device_inline float3 ray_offset(float3 P, float3 Ng)
-{
-#ifdef __INTERSECTION_REFINE__
-	const float epsilon_f = 1e-5f;
-	const int epsilon_i = 32;
-
-	float3 res;
-
-	/* x component */
-	if(fabsf(P.x) < epsilon_f) {
-		res.x = P.x + Ng.x*epsilon_f;
-	}
-	else {
-		uint ix = __float_as_uint(P.x);
-		ix += ((ix ^ __float_as_uint(Ng.x)) >> 31)? -epsilon_i: epsilon_i;
-		res.x = __uint_as_float(ix);
-	}
-
-	/* y component */
-	if(fabsf(P.y) < epsilon_f) {
-		res.y = P.y + Ng.y*epsilon_f;
-	}
-	else {
-		uint iy = __float_as_uint(P.y);
-		iy += ((iy ^ __float_as_uint(Ng.y)) >> 31)? -epsilon_i: epsilon_i;
-		res.y = __uint_as_float(iy);
-	}
-
-	/* z component */
-	if(fabsf(P.z) < epsilon_f) {
-		res.z = P.z + Ng.z*epsilon_f;
-	}
-	else {
-		uint iz = __float_as_uint(P.z);
-		iz += ((iz ^ __float_as_uint(Ng.z)) >> 31)? -epsilon_i: epsilon_i;
-		res.z = __uint_as_float(iz);
-	}
-
-	return res;
-#else
-	const float epsilon_f = 1e-4f;
-	return P + epsilon_f*Ng;
-#endif
-}
-
-__device_inline float3 bvh_triangle_refine(KernelGlobals *kg, const Intersection *isect, const Ray *ray)
-{
-	float3 P = ray->P;
-	float3 D = ray->D;
-	float t = isect->t;
-
-#ifdef __INTERSECTION_REFINE__
-	if(isect->object != ~0) {
-		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
-
-		P = transform_point(&tfm, P);
-		D = transform_direction(&tfm, D*t);
-		D = normalize_len(D, &t);
-	}
-
-	P = P + D*t;
-
-	float4 v00 = kernel_tex_fetch(__tri_woop, isect->prim*TRI_NODE_SIZE+0);
-	float Oz = v00.w - P.x*v00.x - P.y*v00.y - P.z*v00.z;
-	float invDz = 1.0f/(D.x*v00.x + D.y*v00.y + D.z*v00.z);
-	float rt = Oz * invDz;
-
-	P = P + D*rt;
-
-	if(isect->object != ~0) {
-		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
-		P = transform_point(&tfm, P);
-	}
-
-	return P;
-#else
-	return P + D*t;
-#endif
-}
-
-CCL_NAMESPACE_END
-
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index 1ed5e3d352c..2711012edef 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -67,10 +67,18 @@ __device_inline void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
 	sd->v = isect->v;
 #endif
 
+	sd->flag = kernel_tex_fetch(__shader_flag, (shader & SHADER_MASK)*2);
+	sd->flag |= kernel_tex_fetch(__object_flag, sd->object);
+
 	/* matrices and time */
 #ifdef __OBJECT_MOTION__
-	sd->ob_tfm = object_fetch_transform(kg, sd->object, ray->time, OBJECT_TRANSFORM);
-	sd->ob_itfm = object_fetch_transform(kg, sd->object, ray->time, OBJECT_INVERSE_TRANSFORM);
+	if(sd->flag & SD_OBJECT_MOTION) {
+		sd->ob_tfm = object_fetch_transform_motion(kg, sd->object, time, &sd->ob_itfm);
+	}
+	else {
+		sd->ob_tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
+		sd->ob_itfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
+	}
 
 	sd->time = ray->time;
 #endif
@@ -87,9 +95,6 @@ __device_inline void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
 	if(sd->shader & SHADER_SMOOTH_NORMAL)
 		sd->N = triangle_smooth_normal(kg, sd->prim, sd->u, sd->v);
 
-	sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*2);
-	sd->flag |= kernel_tex_fetch(__object_flag, sd->object);
-
 #ifdef __DPDU__
 	/* dPdu/dPdv */
 	triangle_dPdudv(kg, &sd->dPdu, &sd->dPdv, sd->prim);
@@ -173,11 +178,20 @@ __device void shader_setup_from_sample(KernelGlobals *kg, ShaderData *sd,
 	}
 #endif
 
+	sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*2);
+	if(sd->object != -1)
+		sd->flag |= kernel_tex_fetch(__object_flag, sd->object);
+
 #ifdef __OBJECT_MOTION__
-	sd->time = time;
+	if(sd->flag & SD_OBJECT_MOTION) {
+		sd->ob_tfm = object_fetch_transform_motion(kg, sd->object, time, &sd->ob_itfm);
+	}
+	else {
+		sd->ob_tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
+		sd->ob_itfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
+	}
 
-	sd->ob_tfm = object_fetch_transform(kg, sd->object, time, OBJECT_TRANSFORM);
-	sd->ob_itfm = object_fetch_transform(kg, sd->object, time, OBJECT_INVERSE_TRANSFORM);
+	sd->time = time;
 #endif
 
 	/* smooth normal */
@@ -190,10 +204,6 @@ __device void shader_setup_from_sample(KernelGlobals *kg, ShaderData *sd,
 #endif
 	}
 
-	sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*2);
-	if(sd->object != -1)
-		sd->flag |= kernel_tex_fetch(__object_flag, sd->object);
-
 #ifdef __DPDU__
 	/* dPdu/dPdv */
 	if(sd->prim == ~0) {
diff --git a/intern/cycles/kernel/kernel_triangle.h b/intern/cycles/kernel/kernel_triangle.h
index f57c59a45eb..43cfa330724 100644
--- a/intern/cycles/kernel/kernel_triangle.h
+++ b/intern/cycles/kernel/kernel_triangle.h
@@ -201,10 +201,10 @@ __device float4 triangle_motion_vector(KernelGlobals *kg, ShaderData *sd)
 	 * transformation was set match the world/object space of motion_pre/post */
 	Transform tfm;
 	
-	tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_TRANSFORM_MOTION_PRE);
+	tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM_MOTION_PRE);
 	motion_pre = transform_point(&tfm, motion_pre);
 
-	tfm = object_fetch_transform(kg, sd->object, TIME_INVALID, OBJECT_TRANSFORM_MOTION_POST);
+	tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM_MOTION_POST);
 	motion_post = transform_point(&tfm, motion_post);
 
 	float3 P;
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index be49aa54e47..2acea04838a 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -113,7 +113,6 @@ CCL_NAMESPACE_BEGIN
 #endif
 
 //#define __SOBOL_FULL_SCREEN__
-//#define __QBVH__
 
 /* Shader Evaluation */
 
@@ -428,13 +427,6 @@ typedef struct ShaderData {
 	/* length of the ray being shaded */
 	float ray_length;
 
-#ifdef __OBJECT_MOTION__
-	/* object <-> world space transformations, cached to avoid
-	 * re-interpolating them constantly for shading */
-	Transform ob_tfm;
-	Transform ob_itfm;
-#endif
-
 #ifdef __RAY_DIFFERENTIALS__
 	/* differential of P. these are orthogonal to Ng, not N */
 	differential3 dP;
@@ -453,6 +445,13 @@ typedef struct ShaderData {
 	float3 T;
 #endif
 
+#ifdef __OBJECT_MOTION__
+	/* object <-> world space transformations, cached to avoid
+	 * re-interpolating them constantly for shading */
+	Transform ob_tfm;
+	Transform ob_itfm;
+#endif
+
 #ifdef __MULTI_CLOSURE__
 	/* Closure data, we store a fixed array of closures */
 	ShaderClosure closure[MAX_CLOSURE];
@@ -632,7 +631,8 @@ typedef struct KernelBVH {
 	/* root node */
 	int root;
 	int attributes_map_stride;
-	int pad1, pad2;
+	int have_motion;
+	int pad2;
 } KernelBVH;
 
 typedef struct KernelData {
diff --git a/intern/cycles/render/camera.cpp b/intern/cycles/render/camera.cpp
index 441f17d90e9..727b9801d95 100644
--- a/intern/cycles/render/camera.cpp
+++ b/intern/cycles/render/camera.cpp
@@ -19,6 +19,8 @@
 #include "camera.h"
 #include "scene.h"
 
+#include "device.h"
+
 #include "util_vector.h"
 
 CCL_NAMESPACE_BEGIN
@@ -141,7 +143,7 @@ void Camera::update()
 
 void Camera::device_update(Device *device, DeviceScene *dscene, Scene *scene)
 {
-	Scene::MotionType need_motion = scene->need_motion();
+	Scene::MotionType need_motion = scene->need_motion(device->info.advanced_shading);
 
 	update();
 
@@ -274,13 +276,17 @@ bool Camera::modified(const Camera& cam)
 		(border_bottom == cam.border_bottom) &&
 		(border_top == cam.border_top) &&
 		(matrix == cam.matrix) &&
-		(motion == cam.motion) &&
-		(use_motion == cam.use_motion) &&
 		(panorama_type == cam.panorama_type) &&
 		(fisheye_fov == cam.fisheye_fov) &&
 		(fisheye_lens == cam.fisheye_lens));
 }
 
+bool Camera::motion_modified(const Camera& cam)
+{
+	return !((motion == cam.motion) &&
+		(use_motion == cam.use_motion));
+}
+
 void Camera::tag_update()
 {
 	need_update = true;
diff --git a/intern/cycles/render/camera.h b/intern/cycles/render/camera.h
index 82852bde5e0..1407c86e7c2 100644
--- a/intern/cycles/render/camera.h
+++ b/intern/cycles/render/camera.h
@@ -103,6 +103,7 @@ public:
 	void device_free(Device *device, DeviceScene *dscene);
 
 	bool modified(const Camera& cam);
+	bool motion_modified(const Camera& cam);
 	void tag_update();
 };
 
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 014b78dec2b..3f2fe4ab093 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -723,7 +723,8 @@ void MeshManager::device_update(Device *device, DeviceScene *dscene, Scene *scen
 		shader->need_update_attributes = false;
 
 #ifdef __OBJECT_MOTION__
-	bool motion_blur = scene->need_motion() == Scene::MOTION_BLUR;
+	Scene::MotionType need_motion = scene->need_motion(device->info.advanced_shading);
+	bool motion_blur = need_motion == Scene::MOTION_BLUR;
 #else
 	bool motion_blur = false;
 #endif
diff --git a/intern/cycles/render/object.cpp b/intern/cycles/render/object.cpp
index 4a72dcc52f7..f5d78c080c8 100644
--- a/intern/cycles/render/object.cpp
+++ b/intern/cycles/render/object.cpp
@@ -151,7 +151,8 @@ void ObjectManager::device_update_transforms(Device *device, DeviceScene *dscene
 	uint *object_flag = dscene->object_flag.resize(scene->objects.size());
 	int i = 0;
 	map<Mesh*, float> surface_area_map;
-	Scene::MotionType need_motion = scene->need_motion();
+	Scene::MotionType need_motion = scene->need_motion(device->info.advanced_shading);
+	bool have_motion = false;
 
 	foreach(Object *ob, scene->objects) {
 		Mesh *mesh = ob->mesh;
@@ -229,6 +230,7 @@ void ObjectManager::device_update_transforms(Device *device, DeviceScene *dscene
 				transform_motion_decompose(&decomp, &ob->motion);
 				memcpy(&objects[offset+8], &decomp, sizeof(float4)*8);
 				flag |= SD_OBJECT_MOTION;
+				have_motion = true;
 			}
 			else {
 				float4 no_motion = make_float4(FLT_MAX);
@@ -253,6 +255,8 @@ void ObjectManager::device_update_transforms(Device *device, DeviceScene *dscene
 
 	device->tex_alloc("__objects", dscene->objects);
 	device->tex_alloc("__object_flag", dscene->object_flag);
+
+	dscene->data.bvh.have_motion = have_motion;
 }
 
 void ObjectManager::device_update(Device *device, DeviceScene *dscene, Scene *scene, Progress& progress)
@@ -300,7 +304,8 @@ void ObjectManager::apply_static_transforms(Scene *scene, Progress& progress)
 	/* counter mesh users */
 	map<Mesh*, int> mesh_users;
 #ifdef __OBJECT_MOTION__
-	bool motion_blur = scene->need_motion() == Scene::MOTION_BLUR;
+	Scene::MotionType need_motion = scene->need_motion();
+	bool motion_blur = need_motion == Scene::MOTION_BLUR;
 #else
 	bool motion_blur = false;
 #endif
diff --git a/intern/cycles/render/scene.cpp b/intern/cycles/render/scene.cpp
index 071338d49c2..15031b9500c 100644
--- a/intern/cycles/render/scene.cpp
+++ b/intern/cycles/render/scene.cpp
@@ -183,10 +183,10 @@ void Scene::device_update(Device *device_, Progress& progress)
 	device->const_copy_to("__data", &dscene.data, sizeof(dscene.data));
 }
 
-Scene::MotionType Scene::need_motion()
+Scene::MotionType Scene::need_motion(bool advanced_shading)
 {
 	if(integrator->motion_blur)
-		return MOTION_BLUR;
+		return (advanced_shading)? MOTION_BLUR: MOTION_NONE;
 	else if(Pass::contains(film->passes, PASS_MOTION))
 		return MOTION_PASS;
 	else
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index 09087fb2970..bd45c1c04e6 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -194,7 +194,7 @@ public:
 	void need_global_attributes(AttributeRequestSet& attributes);
 
 	enum MotionType { MOTION_NONE = 0, MOTION_PASS, MOTION_BLUR };
-	MotionType need_motion();
+	MotionType need_motion(bool advanced_shading = true);
 
 	bool need_update();
 	bool need_reset();
diff --git a/source/blender/editors/render/render_update.c b/source/blender/editors/render/render_update.c
index cd55b91cb6b..5b1c03f65df 100644
--- a/source/blender/editors/render/render_update.c
+++ b/source/blender/editors/render/render_update.c
@@ -74,12 +74,19 @@ void ED_render_scene_update(Main *bmain, Scene *scene, int updated)
 	bScreen *sc;
 	ScrArea *sa;
 	ARegion *ar;
+	static int recursive_check = FALSE;
 
 	/* don't do this render engine update if we're updating the scene from
 	 * other threads doing e.g. rendering or baking jobs */
 	if (!BLI_thread_is_main())
 		return;
 
+	/* don't call this recursively for frame updates */
+	if(recursive_check)
+		return;
+	
+	recursive_check = TRUE;
+
 	C = CTX_create();
 	CTX_data_main_set(C, bmain);
 	CTX_data_scene_set(C, scene);
@@ -114,6 +121,8 @@ void ED_render_scene_update(Main *bmain, Scene *scene, int updated)
 	}
 
 	CTX_free(C);
+
+	recursive_check = FALSE;
 }
 
 void ED_render_engine_area_exit(ScrArea *sa)
diff --git a/source/blender/makesrna/intern/rna_scene.c b/source/blender/makesrna/intern/rna_scene.c
index a3616e0845a..537dab73340 100644
--- a/source/blender/makesrna/intern/rna_scene.c
+++ b/source/blender/makesrna/intern/rna_scene.c
@@ -3600,14 +3600,14 @@ static void rna_def_scene_render_data(BlenderRNA *brna)
 	RNA_def_property_boolean_sdna(prop, NULL, "mode", R_MBLUR);
 	RNA_def_property_ui_text(prop, "Motion Blur", "Use multi-sampled 3D scene motion blur");
 	RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, NULL);
+	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, "rna_Scene_glsl_update");
 	
 	prop = RNA_def_property(srna, "motion_blur_samples", PROP_INT, PROP_NONE);
 	RNA_def_property_int_sdna(prop, NULL, "mblur_samples");
 	RNA_def_property_range(prop, 1, 32);
 	RNA_def_property_ui_text(prop, "Motion Samples", "Number of scene samples to take with motion blur");
 	RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, NULL);
+	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, "rna_Scene_glsl_update");
 	
 	prop = RNA_def_property(srna, "motion_blur_shutter", PROP_FLOAT, PROP_NONE);
 	RNA_def_property_float_sdna(prop, NULL, "blurfac");
@@ -3615,7 +3615,7 @@ static void rna_def_scene_render_data(BlenderRNA *brna)
 	RNA_def_property_ui_range(prop, 0.01, 2.0f, 1, 0);
 	RNA_def_property_ui_text(prop, "Shutter", "Time taken in frames between shutter open and close");
 	RNA_def_property_clear_flag(prop, PROP_ANIMATABLE);
-	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, NULL);
+	RNA_def_property_update(prop, NC_SCENE | ND_RENDER_OPTIONS, "rna_Scene_glsl_update");
 	
 	/* border */
 	prop = RNA_def_property(srna, "use_border", PROP_BOOLEAN, PROP_NONE);