Merge branch 'master' into blender2.8

6 files changed, 204 insertions, 347 deletions

diff --git a/intern/cycles/bvh/bvh.cpp b/intern/cycles/bvh/bvh.cpp
index 0ad3c8a7429..6531ca50d9a 100644
--- a/intern/cycles/bvh/bvh.cpp
+++ b/intern/cycles/bvh/bvh.cpp
@@ -108,6 +108,73 @@ void BVH::refit(Progress& progress)
 	refit_nodes();
 }
 
+void BVH::refit_primitives(int start, int end, BoundBox& bbox, uint& visibility)
+{
+	/* Refit range of primitives. */
+	for(int prim = start; prim < end; prim++) {
+		int pidx = pack.prim_index[prim];
+		int tob = pack.prim_object[prim];
+		Object *ob = objects[tob];
+
+		if(pidx == -1) {
+			/* Object instance. */
+			bbox.grow(ob->bounds);
+		}
+		else {
+			/* Primitives. */
+			const Mesh *mesh = ob->mesh;
+
+			if(pack.prim_type[prim] & PRIMITIVE_ALL_CURVE) {
+				/* Curves. */
+				int str_offset = (params.top_level)? mesh->curve_offset: 0;
+				Mesh::Curve curve = mesh->get_curve(pidx - str_offset);
+				int k = PRIMITIVE_UNPACK_SEGMENT(pack.prim_type[prim]);
+
+				curve.bounds_grow(k, &mesh->curve_keys[0], &mesh->curve_radius[0], bbox);
+
+				visibility |= PATH_RAY_CURVE;
+
+				/* Motion curves. */
+				if(mesh->use_motion_blur) {
+					Attribute *attr = mesh->curve_attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
+
+					if(attr) {
+						size_t mesh_size = mesh->curve_keys.size();
+						size_t steps = mesh->motion_steps - 1;
+						float3 *key_steps = attr->data_float3();
+
+						for(size_t i = 0; i < steps; i++)
+							curve.bounds_grow(k, key_steps + i*mesh_size, &mesh->curve_radius[0], bbox);
+					}
+				}
+			}
+			else {
+				/* Triangles. */
+				int tri_offset = (params.top_level)? mesh->tri_offset: 0;
+				Mesh::Triangle triangle = mesh->get_triangle(pidx - tri_offset);
+				const float3 *vpos = &mesh->verts[0];
+
+				triangle.bounds_grow(vpos, bbox);
+
+				/* Motion triangles. */
+				if(mesh->use_motion_blur) {
+					Attribute *attr = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
+
+					if(attr) {
+						size_t mesh_size = mesh->verts.size();
+						size_t steps = mesh->motion_steps - 1;
+						float3 *vert_steps = attr->data_float3();
+
+						for(size_t i = 0; i < steps; i++)
+							triangle.bounds_grow(vert_steps + i*mesh_size, bbox);
+					}
+				}
+			}
+		}
+		visibility |= ob->visibility_for_tracing();
+	}
+}
+
 /* Triangles */
 
 void BVH::pack_triangle(int idx, float4 tri_verts[3])
diff --git a/intern/cycles/bvh/bvh.h b/intern/cycles/bvh/bvh.h
index 7bac6112fd9..6a82f915692 100644
--- a/intern/cycles/bvh/bvh.h
+++ b/intern/cycles/bvh/bvh.h
@@ -91,6 +91,9 @@ public:
 protected:
 	BVH(const BVHParams& params, const vector<Object*>& objects);
 
+	/* Refit range of primitives. */
+	void refit_primitives(int start, int end, BoundBox& bbox, uint& visibility);
+
 	/* triangles and strands */
 	void pack_primitives();
 	void pack_triangle(int idx, float4 storage[3]);
diff --git a/intern/cycles/bvh/bvh2.cpp b/intern/cycles/bvh/bvh2.cpp
index 9aa8e71dfd0..9d89d2b6afb 100644
--- a/intern/cycles/bvh/bvh2.cpp
+++ b/intern/cycles/bvh/bvh2.cpp
@@ -247,73 +247,14 @@ void BVH2::refit_nodes()
 void BVH2::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 {
 	if(leaf) {
+		/* refit leaf node */
 		assert(idx + BVH_NODE_LEAF_SIZE <= pack.leaf_nodes.size());
 		const int4 *data = &pack.leaf_nodes[idx];
 		const int c0 = data[0].x;
 		const int c1 = data[0].y;
-		/* refit leaf node */
-		for(int prim = c0; prim < c1; prim++) {
-			int pidx = pack.prim_index[prim];
-			int tob = pack.prim_object[prim];
-			Object *ob = objects[tob];
-
-			if(pidx == -1) {
-				/* object instance */
-				bbox.grow(ob->bounds);
-			}
-			else {
-				/* primitives */
-				const Mesh *mesh = ob->mesh;
-
-				if(pack.prim_type[prim] & PRIMITIVE_ALL_CURVE) {
-					/* curves */
-					int str_offset = (params.top_level)? mesh->curve_offset: 0;
-					Mesh::Curve curve = mesh->get_curve(pidx - str_offset);
-					int k = PRIMITIVE_UNPACK_SEGMENT(pack.prim_type[prim]);
-
-					curve.bounds_grow(k, &mesh->curve_keys[0], &mesh->curve_radius[0], bbox);
-
-					visibility |= PATH_RAY_CURVE;
 
-					/* motion curves */
-					if(mesh->use_motion_blur) {
-						Attribute *attr = mesh->curve_attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
+		BVH::refit_primitives(c0, c1, bbox, visibility);
 
-						if(attr) {
-							size_t mesh_size = mesh->curve_keys.size();
-							size_t steps = mesh->motion_steps - 1;
-							float3 *key_steps = attr->data_float3();
-
-							for(size_t i = 0; i < steps; i++)
-								curve.bounds_grow(k, key_steps + i*mesh_size, &mesh->curve_radius[0], bbox);
-						}
-					}
-				}
-				else {
-					/* triangles */
-					int tri_offset = (params.top_level)? mesh->tri_offset: 0;
-					Mesh::Triangle triangle = mesh->get_triangle(pidx - tri_offset);
-					const float3 *vpos = &mesh->verts[0];
-
-					triangle.bounds_grow(vpos, bbox);
-
-					/* motion triangles */
-					if(mesh->use_motion_blur) {
-						Attribute *attr = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
-
-						if(attr) {
-							size_t mesh_size = mesh->verts.size();
-							size_t steps = mesh->motion_steps - 1;
-							float3 *vert_steps = attr->data_float3();
-
-							for(size_t i = 0; i < steps; i++)
-								triangle.bounds_grow(vert_steps + i*mesh_size, bbox);
-						}
-					}
-				}
-			}
-			visibility |= ob->visibility_for_tracing();
-		}
 		/* TODO(sergey): De-duplicate with pack_leaf(). */
 		float4 leaf_data[BVH_NODE_LEAF_SIZE];
 		leaf_data[0].x = __int_as_float(c0);
diff --git a/intern/cycles/bvh/bvh4.cpp b/intern/cycles/bvh/bvh4.cpp
index 777de20423b..b910feccec8 100644
--- a/intern/cycles/bvh/bvh4.cpp
+++ b/intern/cycles/bvh/bvh4.cpp
@@ -375,71 +375,12 @@ void BVH4::refit_nodes()
 void BVH4::refit_node(int idx, bool leaf, BoundBox& bbox, uint& visibility)
 {
 	if(leaf) {
+		/* Refit leaf node. */
 		int4 *data = &pack.leaf_nodes[idx];
 		int4 c = data[0];
-		/* Refit leaf node. */
-		for(int prim = c.x; prim < c.y; prim++) {
-			int pidx = pack.prim_index[prim];
-			int tob = pack.prim_object[prim];
-			Object *ob = objects[tob];
-
-			if(pidx == -1) {
-				/* Object instance. */
-				bbox.grow(ob->bounds);
-			}
-			else {
-				/* Primitives. */
-				const Mesh *mesh = ob->mesh;
-
-				if(pack.prim_type[prim] & PRIMITIVE_ALL_CURVE) {
-					/* Curves. */
-					int str_offset = (params.top_level)? mesh->curve_offset: 0;
-					Mesh::Curve curve = mesh->get_curve(pidx - str_offset);
-					int k = PRIMITIVE_UNPACK_SEGMENT(pack.prim_type[prim]);
-
-					curve.bounds_grow(k, &mesh->curve_keys[0], &mesh->curve_radius[0], bbox);
-
-					visibility |= PATH_RAY_CURVE;
 
-					/* Motion curves. */
-					if(mesh->use_motion_blur) {
-						Attribute *attr = mesh->curve_attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
+		BVH::refit_primitives(c.x, c.y, bbox, visibility);
 
-						if(attr) {
-							size_t mesh_size = mesh->curve_keys.size();
-							size_t steps = mesh->motion_steps - 1;
-							float3 *key_steps = attr->data_float3();
-
-							for(size_t i = 0; i < steps; i++)
-								curve.bounds_grow(k, key_steps + i*mesh_size, &mesh->curve_radius[0], bbox);
-						}
-					}
-				}
-				else {
-					/* Triangles. */
-					int tri_offset = (params.top_level)? mesh->tri_offset: 0;
-					Mesh::Triangle triangle = mesh->get_triangle(pidx - tri_offset);
-					const float3 *vpos = &mesh->verts[0];
-
-					triangle.bounds_grow(vpos, bbox);
-
-					/* Motion triangles. */
-					if(mesh->use_motion_blur) {
-						Attribute *attr = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
-
-						if(attr) {
-							size_t mesh_size = mesh->verts.size();
-							size_t steps = mesh->motion_steps - 1;
-							float3 *vert_steps = attr->data_float3();
-
-							for(size_t i = 0; i < steps; i++)
-								triangle.bounds_grow(vert_steps + i*mesh_size, bbox);
-						}
-					}
-				}
-			}
-			visibility |= ob->visibility_for_tracing();
-		}
 		/* TODO(sergey): This is actually a copy of pack_leaf(),
 		 * but this chunk of code only knows actual data and has
 		 * no idea about BVHNode.
diff --git a/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h b/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
index b7be4fe4409..5ca07eaeb05 100644
--- a/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
+++ b/intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h
@@ -87,7 +87,7 @@ ccl_device T kernel_tex_image_interp_bicubic(const TextureInfo& info, CUtexObjec
 	                       g1x * tex2D<T>(tex, x1, y1));
 }
 
-/* Fast tricubic texture lookup using 8 bilinear lookups. */
+/* Fast tricubic texture lookup using 8 trilinear lookups. */
 template<typename T>
 ccl_device T kernel_tex_image_interp_bicubic_3d(const TextureInfo& info, CUtexObject tex, float x, float y, float z)
 {
diff --git a/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h b/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
index d908af78c7a..faa9dd66d0e 100644
--- a/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
+++ b/intern/cycles/kernel/kernels/opencl/kernel_opencl_image.h
@@ -27,9 +27,21 @@ ccl_device_inline ccl_global TextureInfo* kernel_tex_info(KernelGlobals *kg, uin
 
 #define tex_fetch(type, info, index) ((ccl_global type*)(kg->buffers[info->cl_buffer] + info->data))[(index)]
 
-ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int offset)
+ccl_device_inline int svm_image_texture_wrap_periodic(int x, int width)
+{
+	x %= width;
+	if(x < 0)
+		x += width;
+	return x;
+}
+
+ccl_device_inline int svm_image_texture_wrap_clamp(int x, int width)
+{
+	return clamp(x, 0, width-1);
+}
+
+ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, const ccl_global TextureInfo *info, int id, int offset)
 {
-	const ccl_global TextureInfo *info = kernel_tex_info(kg, id);
 	const int texture_type = kernel_tex_type(id);
 
 	/* Float4 */
@@ -55,19 +67,45 @@ ccl_device_inline float4 svm_image_texture_read(KernelGlobals *kg, int id, int o
 	}
 }
 
-ccl_device_inline int svm_image_texture_wrap_periodic(int x, int width)
+ccl_device_inline float4 svm_image_texture_read_2d(KernelGlobals *kg, int id, int x, int y)
 {
-	x %= width;
-	if(x < 0)
-		x += width;
-	return x;
+	const ccl_global TextureInfo *info = kernel_tex_info(kg, id);
+
+	/* Wrap */
+	if(info->extension == EXTENSION_REPEAT) {
+		x = svm_image_texture_wrap_periodic(x, info->width);
+		y = svm_image_texture_wrap_periodic(y, info->height);
+	}
+	else {
+		x = svm_image_texture_wrap_clamp(x, info->width);
+		y = svm_image_texture_wrap_clamp(y, info->height);
+	}
+
+	int offset = x + info->width * y;
+	return svm_image_texture_read(kg, info, id, offset);
 }
 
-ccl_device_inline int svm_image_texture_wrap_clamp(int x, int width)
+ccl_device_inline float4 svm_image_texture_read_3d(KernelGlobals *kg, int id, int x, int y, int z)
 {
-	return clamp(x, 0, width-1);
+	const ccl_global TextureInfo *info = kernel_tex_info(kg, id);
+
+	/* Wrap */
+	if(info->extension == EXTENSION_REPEAT) {
+		x = svm_image_texture_wrap_periodic(x, info->width);
+		y = svm_image_texture_wrap_periodic(y, info->height);
+		z = svm_image_texture_wrap_periodic(z, info->depth);
+	}
+	else {
+		x = svm_image_texture_wrap_clamp(x, info->width);
+		y = svm_image_texture_wrap_clamp(y, info->height);
+		z = svm_image_texture_wrap_clamp(z, info->depth);
+	}
+
+	int offset = x + info->width * y + info->width * info->height * z;
+	return svm_image_texture_read(kg, info, id, offset);
 }
 
+
 ccl_device_inline float svm_image_texture_frac(float x, int *ix)
 {
 	int i = float_to_int(x) - ((x < 0.0f)? 1: 0);
@@ -87,107 +125,52 @@ ccl_device float4 kernel_tex_image_interp(KernelGlobals *kg, int id, float x, fl
 {
 	const ccl_global TextureInfo *info = kernel_tex_info(kg, id);
 
-	uint width = info->width;
-	uint height = info->height;
-	uint interpolation = info->interpolation;
-	uint extension = info->extension;
+	if(info->extension == EXTENSION_CLIP) {
+		if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
+			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+		}
+	}
 
-	/* Actual sampling. */
-	if(interpolation == INTERPOLATION_CLOSEST) {
+	if(info->interpolation == INTERPOLATION_CLOSEST) {
+		/* Closest interpolation. */
 		int ix, iy;
-		svm_image_texture_frac(x*width, &ix);
-		svm_image_texture_frac(y*height, &iy);
-
-		if(extension == EXTENSION_REPEAT) {
-			ix = svm_image_texture_wrap_periodic(ix, width);
-			iy = svm_image_texture_wrap_periodic(iy, height);
-		}
-		else {
-			if(extension == EXTENSION_CLIP) {
-				if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
-					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-				}
-			}
-			/* Fall through. */
-			/* EXTENSION_EXTEND */
-			ix = svm_image_texture_wrap_clamp(ix, width);
-			iy = svm_image_texture_wrap_clamp(iy, height);
-		}
+		svm_image_texture_frac(x*info->width, &ix);
+		svm_image_texture_frac(y*info->height, &iy);
 
-		return svm_image_texture_read(kg, id, ix + iy*width);
+		return svm_image_texture_read_2d(kg, id, ix, iy);
+	}
+	else if(info->interpolation == INTERPOLATION_LINEAR) {
+		/* Bilinear interpolation. */
+		int ix, iy;
+		float tx = svm_image_texture_frac(x*info->width - 0.5f, &ix);
+		float ty = svm_image_texture_frac(y*info->height - 0.5f, &iy);
+
+		float4 r;
+		r =  (1.0f - ty)*(1.0f - tx)*svm_image_texture_read_2d(kg, id, ix, iy);
+		r += (1.0f - ty)*tx*svm_image_texture_read_2d(kg, id, ix+1, iy);
+		r += ty*(1.0f - tx)*svm_image_texture_read_2d(kg, id, ix, iy+1);
+		r += ty*tx*svm_image_texture_read_2d(kg, id, ix+1, iy+1);
+		return r;
 	}
 	else {
-		/* Bilinear or bicubic interpolation. */
-		int ix, iy, nix, niy;
-		float tx = svm_image_texture_frac(x*width - 0.5f, &ix);
-		float ty = svm_image_texture_frac(y*height - 0.5f, &iy);
-
-		if(extension == EXTENSION_REPEAT) {
-			ix = svm_image_texture_wrap_periodic(ix, width);
-			iy = svm_image_texture_wrap_periodic(iy, height);
-			nix = svm_image_texture_wrap_periodic(ix+1, width);
-			niy = svm_image_texture_wrap_periodic(iy+1, height);
-		}
-		else {
-			if(extension == EXTENSION_CLIP) {
-				if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
-					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-				}
-			}
-			ix = svm_image_texture_wrap_clamp(ix, width);
-			iy = svm_image_texture_wrap_clamp(iy, height);
-			nix = svm_image_texture_wrap_clamp(ix+1, width);
-			niy = svm_image_texture_wrap_clamp(iy+1, height);
-		}
-
-		if(interpolation == INTERPOLATION_LINEAR) {
-			/* Bilinear interpolation. */
-			float4 r;
-			r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, id, ix + iy*width);
-			r += (1.0f - ty)*tx*svm_image_texture_read(kg, id, nix + iy*width);
-			r += ty*(1.0f - tx)*svm_image_texture_read(kg, id, ix + niy*width);
-			r += ty*tx*svm_image_texture_read(kg, id, nix + niy*width);
-			return r;
-		}
-
 		/* Bicubic interpolation. */
-		int pix, piy, nnix, nniy;
-		if(extension == EXTENSION_REPEAT) {
-			pix = svm_image_texture_wrap_periodic(ix-1, width);
-			piy = svm_image_texture_wrap_periodic(iy-1, height);
-			nnix = svm_image_texture_wrap_periodic(ix+2, width);
-			nniy = svm_image_texture_wrap_periodic(iy+2, height);
-		}
-		else {
-			pix = svm_image_texture_wrap_clamp(ix-1, width);
-			piy = svm_image_texture_wrap_clamp(iy-1, height);
-			nnix = svm_image_texture_wrap_clamp(ix+2, width);
-			nniy = svm_image_texture_wrap_clamp(iy+2, height);
-		}
+		int ix, iy;
+		float tx = svm_image_texture_frac(x*info->width - 0.5f, &ix);
+		float ty = svm_image_texture_frac(y*info->height - 0.5f, &iy);
 
-		const int xc[4] = {pix, ix, nix, nnix};
-		const int yc[4] = {width * piy,
-		                   width * iy,
-		                   width * niy,
-		                   width * nniy};
 		float u[4], v[4];
-		/* Some helper macro to keep code reasonable size,
-		 * let compiler to inline all the matrix multiplications.
-		 */
-#define DATA(x, y) (svm_image_texture_read(kg, id, xc[x] + yc[y]))
-#define TERM(col) \
-		(v[col] * (u[0] * DATA(0, col) + \
-		           u[1] * DATA(1, col) + \
-		           u[2] * DATA(2, col) + \
-		           u[3] * DATA(3, col)))
-
 		SET_CUBIC_SPLINE_WEIGHTS(u, tx);
 		SET_CUBIC_SPLINE_WEIGHTS(v, ty);
 
-		/* Actual interpolation. */
-		return TERM(0) + TERM(1) + TERM(2) + TERM(3);
-#undef TERM
-#undef DATA
+		float4 r = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+
+		for(int y = 0; y < 4; y++) {
+			for(int x = 0; x < 4; x++) {
+				float weight = u[x]*v[y];
+				r += weight*svm_image_texture_read_2d(kg, id, ix+x-1, iy+y-1);
+			}
+		}
+		return r;
 	}
 }
 
@@ -196,145 +179,67 @@ ccl_device float4 kernel_tex_image_interp_3d(KernelGlobals *kg, int id, float x,
 {
 	const ccl_global TextureInfo *info = kernel_tex_info(kg, id);
 
-	uint width = info->width;
-	uint height = info->height;
-	uint depth = info->depth;
+	if(info->extension == EXTENSION_CLIP) {
+		if(x < 0.0f || y < 0.0f || z < 0.0f ||
+		   x > 1.0f || y > 1.0f || z > 1.0f)
+		{
+			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+		}
+	}
+
 	uint interpolation = (interp == INTERPOLATION_NONE)? info->interpolation: interp;
-	uint extension = info->extension;
 
-	/* Actual sampling. */
 	if(interpolation == INTERPOLATION_CLOSEST) {
+		/* Closest interpolation. */
 		int ix, iy, iz;
-		svm_image_texture_frac(x*width, &ix);
-		svm_image_texture_frac(y*height, &iy);
-		svm_image_texture_frac(z*depth, &iz);
-
-		if(extension == EXTENSION_REPEAT) {
-			ix = svm_image_texture_wrap_periodic(ix, width);
-			iy = svm_image_texture_wrap_periodic(iy, height);
-			iz = svm_image_texture_wrap_periodic(iz, depth);
-		}
-		else {
-			if(extension == EXTENSION_CLIP) {
-				if(x < 0.0f || y < 0.0f || z < 0.0f ||
-				   x > 1.0f || y > 1.0f || z > 1.0f)
-				{
-					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-				}
-			}
-			/* Fall through. */
-			/* EXTENSION_EXTEND */
-			ix = svm_image_texture_wrap_clamp(ix, width);
-			iy = svm_image_texture_wrap_clamp(iy, height);
-			iz = svm_image_texture_wrap_clamp(iz, depth);
-		}
-		return svm_image_texture_read(kg, id, ix + iy*width + iz*width*height);
+		svm_image_texture_frac(x*info->width, &ix);
+		svm_image_texture_frac(y*info->height, &iy);
+		svm_image_texture_frac(z*info->depth, &iz);
+
+		return svm_image_texture_read_3d(kg, id, ix, iy, iz);
+	}
+	else if(interpolation == INTERPOLATION_LINEAR) {
+		/* Bilinear interpolation. */
+		int ix, iy, iz;
+		float tx = svm_image_texture_frac(x*info->width - 0.5f, &ix);
+		float ty = svm_image_texture_frac(y*info->height - 0.5f, &iy);
+		float tz = svm_image_texture_frac(z*info->depth - 0.5f, &iz);
+
+		float4 r;
+		r  = (1.0f - tz)*(1.0f - ty)*(1.0f - tx)*svm_image_texture_read_3d(kg, id, ix, iy, iz);
+		r += (1.0f - tz)*(1.0f - ty)*tx*svm_image_texture_read_3d(kg, id, ix+1, iy, iz);
+		r += (1.0f - tz)*ty*(1.0f - tx)*svm_image_texture_read_3d(kg, id, ix, iy+1, iz);
+		r += (1.0f - tz)*ty*tx*svm_image_texture_read_3d(kg, id, ix+1, iy+1, iz);
+
+		r += tz*(1.0f - ty)*(1.0f - tx)*svm_image_texture_read_3d(kg, id, ix, iy, iz+1);
+		r += tz*(1.0f - ty)*tx*svm_image_texture_read_3d(kg, id, ix+1, iy, iz+1);
+		r += tz*ty*(1.0f - tx)*svm_image_texture_read_3d(kg, id, ix, iy+1, iz+1);
+		r += tz*ty*tx*svm_image_texture_read_3d(kg, id, ix+1, iy+1, iz+1);
+		return r;
 	}
 	else {
-		/* Bilinear or bicubic interpolation. */
-		int ix, iy, iz, nix, niy, niz;
-		float tx = svm_image_texture_frac(x*(float)width - 0.5f, &ix);
-		float ty = svm_image_texture_frac(y*(float)height - 0.5f, &iy);
-		float tz = svm_image_texture_frac(z*(float)depth - 0.5f, &iz);
-
-		if(extension == EXTENSION_REPEAT) {
-			ix = svm_image_texture_wrap_periodic(ix, width);
-			iy = svm_image_texture_wrap_periodic(iy, height);
-			iz = svm_image_texture_wrap_periodic(iz, depth);
-
-			nix = svm_image_texture_wrap_periodic(ix+1, width);
-			niy = svm_image_texture_wrap_periodic(iy+1, height);
-			niz = svm_image_texture_wrap_periodic(iz+1, depth);
-		}
-		else {
-			if(extension == EXTENSION_CLIP) {
-				if(x < 0.0f || y < 0.0f || z < 0.0f ||
-				   x > 1.0f || y > 1.0f || z > 1.0f)
-				{
-					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
-				}
-			}
-			/* Fall through. */
-			/*  EXTENSION_EXTEND */
-			nix = svm_image_texture_wrap_clamp(ix+1, width);
-			niy = svm_image_texture_wrap_clamp(iy+1, height);
-			niz = svm_image_texture_wrap_clamp(iz+1, depth);
-
-			ix = svm_image_texture_wrap_clamp(ix, width);
-			iy = svm_image_texture_wrap_clamp(iy, height);
-			iz = svm_image_texture_wrap_clamp(iz, depth);
-		}
-
-		if(interpolation == INTERPOLATION_LINEAR) {
-			/* Bilinear interpolation. */
-			float4 r;
-			r  = (1.0f - tz)*(1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, id, ix + iy*width + iz*width*height);
-			r += (1.0f - tz)*(1.0f - ty)*tx*svm_image_texture_read(kg, id, nix + iy*width + iz*width*height);
-			r += (1.0f - tz)*ty*(1.0f - tx)*svm_image_texture_read(kg, id, ix + niy*width + iz*width*height);
-			r += (1.0f - tz)*ty*tx*svm_image_texture_read(kg, id, nix + niy*width + iz*width*height);
-
-			r += tz*(1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, id, ix + iy*width + niz*width*height);
-			r += tz*(1.0f - ty)*tx*svm_image_texture_read(kg, id, nix + iy*width + niz*width*height);
-			r += tz*ty*(1.0f - tx)*svm_image_texture_read(kg, id, ix + niy*width + niz*width*height);
-			r += tz*ty*tx*svm_image_texture_read(kg, id, nix + niy*width + niz*width*height);
-			return r;
-		}
-
 		/* Bicubic interpolation. */
-		int pix, piy, piz, nnix, nniy, nniz;
-		if(extension == EXTENSION_REPEAT) {
-			pix = svm_image_texture_wrap_periodic(ix-1, width);
-			piy = svm_image_texture_wrap_periodic(iy-1, height);
-			piz = svm_image_texture_wrap_periodic(iz-1, depth);
-			nnix = svm_image_texture_wrap_periodic(ix+2, width);
-			nniy = svm_image_texture_wrap_periodic(iy+2, height);
-			nniz = svm_image_texture_wrap_periodic(iz+2, depth);
-		}
-		else {
-			pix = svm_image_texture_wrap_clamp(ix-1, width);
-			piy = svm_image_texture_wrap_clamp(iy-1, height);
-			piz = svm_image_texture_wrap_clamp(iz-1, depth);
-			nnix = svm_image_texture_wrap_clamp(ix+2, width);
-			nniy = svm_image_texture_wrap_clamp(iy+2, height);
-			nniz = svm_image_texture_wrap_clamp(iz+2, depth);
-		}
+		int ix, iy, iz;
+		float tx = svm_image_texture_frac(x*info->width - 0.5f, &ix);
+		float ty = svm_image_texture_frac(y*info->height - 0.5f, &iy);
+		float tz = svm_image_texture_frac(z*info->depth - 0.5f, &iz);
 
-		const int xc[4] = {pix, ix, nix, nnix};
-		const int yc[4] = {width * piy,
-		                   width * iy,
-		                   width * niy,
-		                   width * nniy};
-		const int zc[4] = {width * height * piz,
-		                   width * height * iz,
-		                   width * height * niz,
-		                   width * height * nniz};
 		float u[4], v[4], w[4];
-
-		/* Some helper macro to keep code reasonable size,
-		 * let compiler to inline all the matrix multiplications.
-		 */
-#define DATA(x, y, z) (svm_image_texture_read(kg, id, xc[x] + yc[y] + zc[z]))
-#define COL_TERM(col, row) \
-		(v[col] * (u[0] * DATA(0, col, row) + \
-		           u[1] * DATA(1, col, row) + \
-		           u[2] * DATA(2, col, row) + \
-		           u[3] * DATA(3, col, row)))
-#define ROW_TERM(row) \
-		(w[row] * (COL_TERM(0, row) + \
-		           COL_TERM(1, row) + \
-		           COL_TERM(2, row) + \
-		           COL_TERM(3, row)))
-
 		SET_CUBIC_SPLINE_WEIGHTS(u, tx);
 		SET_CUBIC_SPLINE_WEIGHTS(v, ty);
 		SET_CUBIC_SPLINE_WEIGHTS(w, tz);
 
-		/* Actual interpolation. */
-		return ROW_TERM(0) + ROW_TERM(1) + ROW_TERM(2) + ROW_TERM(3);
+		float4 r = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 
-#undef COL_TERM
-#undef ROW_TERM
-#undef DATA
+		for(int z = 0; z < 4; z++) {
+			for(int y = 0; y < 4; y++) {
+				for(int x = 0; x < 4; x++) {
+					float weight = u[x]*v[y]*w[z];
+					r += weight*svm_image_texture_read_3d(kg, id, ix+x-1, iy+y-1, iz+z-1);
+				}
+			}
+		}
+		return r;
 	}
 }