Cycles code internals: add CPU kernel support for 3D image textures.

11 files changed, 201 insertions, 85 deletions

diff --git a/intern/cycles/blender/blender_session.cpp b/intern/cycles/blender/blender_session.cpp
index 09e7472caf2..82434712ca7 100644
--- a/intern/cycles/blender/blender_session.cpp
+++ b/intern/cycles/blender/blender_session.cpp
@@ -93,6 +93,11 @@ void BlenderSession::create_session()
 	/* create scene */
 	scene = new Scene(scene_params, session_params.device);
 
+	/* setup callbacks for builtin image support */
+	scene->image_manager->builtin_image_info_cb = function_bind(&BlenderSession::builtin_image_info, this, _1, _2, _3, _4, _5, _6, _7);
+	scene->image_manager->builtin_image_pixels_cb = function_bind(&BlenderSession::builtin_image_pixels, this, _1, _2, _3);
+	scene->image_manager->builtin_image_float_pixels_cb = function_bind(&BlenderSession::builtin_image_float_pixels, this, _1, _2, _3);
+
 	/* create session */
 	session = new Session(session_params);
 	session->scene = scene;
@@ -121,11 +126,6 @@ void BlenderSession::create_session()
 	session->reset(buffer_params, session_params.samples);
 
 	b_engine.use_highlight_tiles(session_params.progressive_refine == false);
-
-	/* setup callbacks for builtin image support */
-	scene->image_manager->builtin_image_info_cb = function_bind(&BlenderSession::builtin_image_info, this, _1, _2, _3, _4, _5, _6);
-	scene->image_manager->builtin_image_pixels_cb = function_bind(&BlenderSession::builtin_image_pixels, this, _1, _2, _3);
-	scene->image_manager->builtin_image_float_pixels_cb = function_bind(&BlenderSession::builtin_image_float_pixels, this, _1, _2, _3);
 }
 
 void BlenderSession::reset_session(BL::BlendData b_data_, BL::Scene b_scene_)
@@ -724,7 +724,7 @@ int BlenderSession::builtin_image_frame(const string &builtin_name)
 	return atoi(builtin_name.substr(last + 1, builtin_name.size() - last - 1).c_str());
 }
 
-void BlenderSession::builtin_image_info(const string &builtin_name, void *builtin_data, bool &is_float, int &width, int &height, int &channels)
+void BlenderSession::builtin_image_info(const string &builtin_name, void *builtin_data, bool &is_float, int &width, int &height, int &depth, int &channels)
 {
 	PointerRNA ptr;
 	RNA_id_pointer_create((ID*)builtin_data, &ptr);
@@ -734,12 +734,14 @@ void BlenderSession::builtin_image_info(const string &builtin_name, void *builti
 		is_float = b_image.is_float();
 		width = b_image.size()[0];
 		height = b_image.size()[1];
+		depth = 1;
 		channels = b_image.channels();
 	}
 	else {
 		is_float = false;
 		width = 0;
 		height = 0;
+		depth = 0;
 		channels = 0;
 	}
 }
diff --git a/intern/cycles/blender/blender_session.h b/intern/cycles/blender/blender_session.h
index 0568fb291d0..d30e3eda870 100644
--- a/intern/cycles/blender/blender_session.h
+++ b/intern/cycles/blender/blender_session.h
@@ -99,7 +99,7 @@ protected:
 	void do_write_update_render_tile(RenderTile& rtile, bool do_update_only);
 
 	int builtin_image_frame(const string &builtin_name);
-	void builtin_image_info(const string &builtin_name, void *builtin_data, bool &is_float, int &width, int &height, int &channels);
+	void builtin_image_info(const string &builtin_name, void *builtin_data, bool &is_float, int &width, int &height, int &depth, int &channels);
 	bool builtin_image_pixels(const string &builtin_name, void *builtin_data, unsigned char *pixels);
 	bool builtin_image_float_pixels(const string &builtin_name, void *builtin_data, float *pixels);
 };
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index de44feb2deb..e14e403d82b 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -105,7 +105,7 @@ public:
 
 	void tex_alloc(const char *name, device_memory& mem, InterpolationType interpolation, bool periodic)
 	{
-		kernel_tex_copy(&kernel_globals, name, mem.data_pointer, mem.data_width, mem.data_height, interpolation);
+		kernel_tex_copy(&kernel_globals, name, mem.data_pointer, mem.data_width, mem.data_height, mem.data_depth, interpolation);
 		mem.device_pointer = mem.data_pointer;
 
 		stats.mem_alloc(mem.memory_size());
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index a8a3134328f..26d356af458 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -461,6 +461,8 @@ public:
 
 	void tex_alloc(const char *name, device_memory& mem, InterpolationType interpolation, bool periodic)
 	{
+		/* todo: support 3D textures, only CPU for now */
+
 		/* determine format */
 		CUarray_format_enum format;
 		size_t dsize = datatype_size(mem.data_type);
diff --git a/intern/cycles/device/device_memory.h b/intern/cycles/device/device_memory.h
index 1427d12cba2..8d6f4a49a9c 100644
--- a/intern/cycles/device/device_memory.h
+++ b/intern/cycles/device/device_memory.h
@@ -169,6 +169,7 @@ public:
 	size_t data_size;
 	size_t data_width;
 	size_t data_height;
+	size_t data_depth;
 
 	/* device pointer */
 	device_ptr device_pointer;
@@ -195,6 +196,7 @@ public:
 		data_size = 0;
 		data_width = 0;
 		data_height = 0;
+		data_depth = 0;
 
 		assert(data_elements > 0);
 
@@ -204,20 +206,21 @@ public:
 	virtual ~device_vector() {}
 
 	/* vector functions */
-	T *resize(size_t width, size_t height = 0)
+	T *resize(size_t width, size_t height = 0, size_t depth = 0)
 	{
-		data_size = (height == 0)? width: width*height;
+		data_size = width * ((height == 0)? 1: height) * ((depth == 0)? 1: depth);
 		data.resize(data_size);
 		data_pointer = (device_ptr)&data[0];
 		data_width = width;
 		data_height = height;
+		data_depth = depth;
 
 		return &data[0];
 	}
 
-	T *copy(T *ptr, size_t width, size_t height = 0)
+	T *copy(T *ptr, size_t width, size_t height = 0, size_t depth = 0)
 	{
-		T *mem = resize(width, height);
+		T *mem = resize(width, height, depth);
 		memcpy(mem, ptr, memory_size());
 		return mem;
 	}
@@ -230,13 +233,14 @@ public:
 		}
 	}
 
-	void reference(T *ptr, size_t width, size_t height = 0)
+	void reference(T *ptr, size_t width, size_t height = 0, size_t depth = 0)
 	{
 		data.clear();
-		data_size = (height == 0)? width: width*height;
+		data_size = width * ((height == 0)? 1: height) * ((depth == 0)? 1: depth);
 		data_pointer = (device_ptr)ptr;
 		data_width = width;
 		data_height = height;
+		data_depth = depth;
 	}
 
 	void clear()
@@ -245,6 +249,7 @@ public:
 		data_pointer = 0;
 		data_width = 0;
 		data_height = 0;
+		data_depth = 0;
 		data_size = 0;
 	}
 
diff --git a/intern/cycles/device/device_network.h b/intern/cycles/device/device_network.h
index bf8f3c70c49..8596b61b179 100644
--- a/intern/cycles/device/device_network.h
+++ b/intern/cycles/device/device_network.h
@@ -118,7 +118,7 @@ public:
 	void add(const device_memory& mem)
 	{
 		archive & mem.data_type & mem.data_elements & mem.data_size;
-		archive & mem.data_width & mem.data_height & mem.device_pointer;
+		archive & mem.data_width & mem.data_height & mem.data_depth & mem.device_pointer;
 	}
 
 	template<typename T> void add(const T& data)
@@ -261,7 +261,7 @@ public:
 	void read(network_device_memory& mem)
 	{
 		*archive & mem.data_type & mem.data_elements & mem.data_size;
-		*archive & mem.data_width & mem.data_height & mem.device_pointer;
+		*archive & mem.data_width & mem.data_height & mem.data_depth & mem.device_pointer;
 
 		mem.data_pointer = 0;
 	}
diff --git a/intern/cycles/kernel/kernel.cpp b/intern/cycles/kernel/kernel.cpp
index 5d74feed9f4..41cf7a6b66c 100644
--- a/intern/cycles/kernel/kernel.cpp
+++ b/intern/cycles/kernel/kernel.cpp
@@ -37,7 +37,7 @@ void kernel_const_copy(KernelGlobals *kg, const char *name, void *host, size_t s
 		assert(0);
 }
 
-void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t width, size_t height, InterpolationType interpolation)
+void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t width, size_t height, size_t depth, InterpolationType interpolation)
 {
 	if(0) {
 	}
@@ -63,6 +63,7 @@ void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t
 			tex->data = (float4*)mem;
 			tex->width = width;
 			tex->height = height;
+			tex->depth = depth;
 			tex->interpolation = interpolation;
 		}
 	}
@@ -79,6 +80,7 @@ void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t
 			tex->data = (uchar4*)mem;
 			tex->width = width;
 			tex->height = height;
+			tex->depth = depth;
 			tex->interpolation = interpolation;
 		}
 	}
diff --git a/intern/cycles/kernel/kernel.h b/intern/cycles/kernel/kernel.h
index a4f13e78006..c4a08646bab 100644
--- a/intern/cycles/kernel/kernel.h
+++ b/intern/cycles/kernel/kernel.h
@@ -32,7 +32,7 @@ void *kernel_osl_memory(KernelGlobals *kg);
 bool kernel_osl_use(KernelGlobals *kg);
 
 void kernel_const_copy(KernelGlobals *kg, const char *name, void *host, size_t size);
-void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t width, size_t height, InterpolationType interpolation=INTERPOLATION_LINEAR);
+void kernel_tex_copy(KernelGlobals *kg, const char *name, device_ptr mem, size_t width, size_t height, size_t depth, InterpolationType interpolation=INTERPOLATION_LINEAR);
 
 void kernel_cpu_path_trace(KernelGlobals *kg, float *buffer, unsigned int *rng_state,
 	int sample, int x, int y, int offset, int stride);
diff --git a/intern/cycles/kernel/kernel_compat_cpu.h b/intern/cycles/kernel/kernel_compat_cpu.h
index 850ef0abed1..feaff503212 100644
--- a/intern/cycles/kernel/kernel_compat_cpu.h
+++ b/intern/cycles/kernel/kernel_compat_cpu.h
@@ -99,6 +99,7 @@ template<typename T> struct texture_image  {
 			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 
 		int ix, iy, nix, niy;
+
 		if(interpolation == INTERPOLATION_CLOSEST) {
 			frac(x*width, &ix);
 			frac(y*height, &iy);
@@ -131,17 +132,84 @@ template<typename T> struct texture_image  {
 				nix = wrap_clamp(ix+1, width);
 				niy = wrap_clamp(iy+1, height);
 			}
+
 			float4 r = (1.0f - ty)*(1.0f - tx)*read(data[ix + iy*width]);
 			r += (1.0f - ty)*tx*read(data[nix + iy*width]);
 			r += ty*(1.0f - tx)*read(data[ix + niy*width]);
 			r += ty*tx*read(data[nix + niy*width]);
+
+			return r;
+		}
+	}
+
+	ccl_always_inline float4 interp_3d(float x, float y, float z, bool periodic = false)
+	{
+		if(!data)
+			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
+
+		int ix, iy, iz, nix, niy, niz;
+
+		if(interpolation == INTERPOLATION_CLOSEST) {
+			frac(x*width, &ix);
+			frac(y*height, &iy);
+			frac(z*depth, &iz);
+
+			if(periodic) {
+				ix = wrap_periodic(ix, width);
+				iy = wrap_periodic(iy, height);
+				iz = wrap_periodic(iz, depth);
+			}
+			else {
+				ix = wrap_clamp(ix, width);
+				iy = wrap_clamp(iy, height);
+				iz = wrap_clamp(iz, depth);
+			}
+
+			return read(data[ix + iy*width + iz*width*height]);
+		}
+		else {
+			float tx = frac(x*width - 0.5f, &ix);
+			float ty = frac(y*height - 0.5f, &iy);
+			float tz = frac(z*depth - 0.5f, &iz);
+
+			if(periodic) {
+				ix = wrap_periodic(ix, width);
+				iy = wrap_periodic(iy, height);
+				iz = wrap_periodic(iz, depth);
+
+				nix = wrap_periodic(ix+1, width);
+				niy = wrap_periodic(iy+1, height);
+				niz = wrap_periodic(iz+1, depth);
+			}
+			else {
+				ix = wrap_clamp(ix, width);
+				iy = wrap_clamp(iy, height);
+				iz = wrap_clamp(iz, depth);
+
+				nix = wrap_clamp(ix+1, width);
+				niy = wrap_clamp(iy+1, height);
+				niz = wrap_clamp(iz+1, depth);
+			}
+
+			float4 r;
+
+			r  = (1.0f - tz)*(1.0f - ty)*(1.0f - tx)*read(data[ix + iy*width + iz*width*height]);
+			r += (1.0f - tz)*(1.0f - ty)*tx*read(data[nix + iy*width + iz*width*height]);
+			r += (1.0f - tz)*ty*(1.0f - tx)*read(data[ix + niy*width + iz*width*height]);
+			r += (1.0f - tz)*ty*tx*read(data[nix + niy*width + iz*width*height]);
+
+			r += tz*(1.0f - ty)*(1.0f - tx)*read(data[ix + iy*width + niz*width*height]);
+			r += tz*(1.0f - ty)*tx*read(data[nix + iy*width + niz*width*height]);
+			r += tz*ty*(1.0f - tx)*read(data[ix + niy*width + niz*width*height]);
+			r += tz*ty*tx*read(data[nix + niy*width + niz*width*height]);
+
 			return r;
 		}
 	}
 
-	int interpolation;
 	T *data;
-	int width, height;
+	int interpolation;
+	int width, height, depth;
 };
 
 typedef texture<float4> texture_float4;
@@ -161,6 +229,8 @@ typedef texture_image<uchar4> texture_image_uchar4;
 #define kernel_tex_fetch_m128i(tex, index) (kg->tex.fetch_m128i(index))
 #define kernel_tex_lookup(tex, t, offset, size) (kg->tex.lookup(t, offset, size))
 #define kernel_tex_image_interp(tex, x, y) ((tex < MAX_FLOAT_IMAGES) ? kg->texture_float_images[tex].interp(x, y) : kg->texture_byte_images[tex - MAX_FLOAT_IMAGES].interp(x, y))
+#define kernel_tex_image_interp_3d(tex, x, y, z) ((tex < MAX_FLOAT_IMAGES) ? kg->texture_float_images[tex].interp_3d(x, y, z) : kg->texture_byte_images[tex - MAX_FLOAT_IMAGES].interp_3d(x, y, z))
+
 #define kernel_data (kg->__data)
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/render/image.cpp b/intern/cycles/render/image.cpp
index ecccaf54ba9..a0994708ce1 100644
--- a/intern/cycles/render/image.cpp
+++ b/intern/cycles/render/image.cpp
@@ -90,8 +90,8 @@ bool ImageManager::is_float_image(const string& filename, void *builtin_data, bo
 
 	if(builtin_data) {
 		if(builtin_image_info_cb) {
-			int width, height, channels;
-			builtin_image_info_cb(filename, builtin_data, is_float, width, height, channels);
+			int width, height, depth, channels;
+			builtin_image_info_cb(filename, builtin_data, is_float, width, height, depth, channels);
 		}
 
 		if(is_float)
@@ -145,6 +145,13 @@ bool ImageManager::is_float_image(const string& filename, void *builtin_data, bo
 	return is_float;
 }
 
+static bool image_equals(ImageManager::Image *image, const string& filename, void *builtin_data, bool interpolation)
+{
+	return image->filename == filename &&
+	       image->builtin_data == builtin_data &&
+	       image->interpolation == interpolation;
+}
+
 int ImageManager::add_image(const string& filename, void *builtin_data, bool animated, bool& is_float, bool& is_linear, InterpolationType interpolation)
 {
 	Image *img;
@@ -156,7 +163,7 @@ int ImageManager::add_image(const string& filename, void *builtin_data, bool ani
 	if(is_float) {
 		/* find existing image */
 		for(slot = 0; slot < float_images.size(); slot++) {
-			if(float_images[slot] && float_images[slot]->filename == filename && float_images[slot]->interpolation == interpolation) {
+			if(float_images[slot] && image_equals(float_images[slot], filename, builtin_data, interpolation)) {
 				float_images[slot]->users++;
 				return slot;
 			}
@@ -192,7 +199,7 @@ int ImageManager::add_image(const string& filename, void *builtin_data, bool ani
 	}
 	else {
 		for(slot = 0; slot < images.size(); slot++) {
-			if(images[slot] && images[slot]->filename == filename && images[slot]->interpolation == interpolation) {
+			if(images[slot] && image_equals(images[slot], filename, builtin_data, interpolation)) {
 				images[slot]->users++;
 				return slot+tex_image_byte_start;
 			}
@@ -233,22 +240,43 @@ int ImageManager::add_image(const string& filename, void *builtin_data, bool ani
 	return slot;
 }
 
+void ImageManager::remove_image(int slot)
+{
+	if(slot >= tex_image_byte_start) {
+		slot -= tex_image_byte_start;
+
+		assert(images[slot] != NULL);
+
+		/* decrement user count */
+		images[slot]->users--;
+		assert(images[slot]->users >= 0);
+
+		/* don't remove immediately, rather do it all together later on. one of
+		 * the reasons for this is that on shader changes we add and remove nodes
+		 * that use them, but we do not want to reload the image all the time. */
+		if(images[slot]->users == 0)
+			need_update = true;
+	}
+	else {
+		/* decrement user count */
+		float_images[slot]->users--;
+		assert(float_images[slot]->users >= 0);
+
+		/* don't remove immediately, rather do it all together later on. one of
+		 * the reasons for this is that on shader changes we add and remove nodes
+		 * that use them, but we do not want to reload the image all the time. */
+		if(float_images[slot]->users == 0)
+			need_update = true;
+	}
+}
+
 void ImageManager::remove_image(const string& filename, void *builtin_data, InterpolationType interpolation)
 {
 	size_t slot;
 
 	for(slot = 0; slot < images.size(); slot++) {
-		if(images[slot] && images[slot]->filename == filename && images[slot]->interpolation == interpolation && images[slot]->builtin_data == builtin_data) {
-			/* decrement user count */
-			images[slot]->users--;
-			assert(images[slot]->users >= 0);
-
-			/* don't remove immediately, rather do it all together later on. one of
-			 * the reasons for this is that on shader changes we add and remove nodes
-			 * that use them, but we do not want to reload the image all the time. */
-			if(images[slot]->users == 0)
-				need_update = true;
-
+		if(images[slot] && image_equals(images[slot], filename, builtin_data, interpolation)) {
+			remove_image(slot+tex_image_byte_start);
 			break;
 		}
 	}
@@ -256,19 +284,8 @@ void ImageManager::remove_image(const string& filename, void *builtin_data, Inte
 	if(slot == images.size()) {
 		/* see if it's in a float texture slot */
 		for(slot = 0; slot < float_images.size(); slot++) {
-			if(float_images[slot] && float_images[slot]->filename == filename
-								  && float_images[slot]->interpolation == interpolation
-								  && float_images[slot]->builtin_data == builtin_data) {
-				/* decrement user count */
-				float_images[slot]->users--;
-				assert(float_images[slot]->users >= 0);
-
-				/* don't remove immediately, rather do it all together later on. one of
-				 * the reasons for this is that on shader changes we add and remove nodes
-				 * that use them, but we do not want to reload the image all the time. */
-				if(float_images[slot]->users == 0)
-					need_update = true;
-
+			if(float_images[slot] && image_equals(float_images[slot], filename, builtin_data, interpolation)) {
+				remove_image(slot);
 				break;
 			}
 		}
@@ -281,7 +298,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 		return false;
 
 	ImageInput *in = NULL;
-	int width, height, components;
+	int width, height, depth, components;
 
 	if(!img->builtin_data) {
 		/* load image from file through OIIO */
@@ -299,6 +316,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 
 		width = spec.width;
 		height = spec.height;
+		depth = spec.depth;
 		components = spec.nchannels;
 	}
 	else {
@@ -307,7 +325,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 			return false;
 
 		bool is_float;
-		builtin_image_info_cb(img->filename, img->builtin_data, is_float, width, height, components);
+		builtin_image_info_cb(img->filename, img->builtin_data, is_float, width, height, depth, components);
 	}
 
 	/* we only handle certain number of components */
@@ -321,15 +339,21 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 	}
 
 	/* read RGBA pixels */
-	uchar *pixels = (uchar*)tex_img.resize(width, height);
-	int scanlinesize = width*components*sizeof(uchar);
+	uchar *pixels = (uchar*)tex_img.resize(width, height, depth);
 
 	if(in) {
-		in->read_image(TypeDesc::UINT8,
-			(uchar*)pixels + (height-1)*scanlinesize,
-			AutoStride,
-			-scanlinesize,
-			AutoStride);
+		if(depth <= 1) {
+			int scanlinesize = width*components*sizeof(uchar);
+
+			in->read_image(TypeDesc::UINT8,
+				(uchar*)pixels + (height-1)*scanlinesize,
+				AutoStride,
+				-scanlinesize,
+				AutoStride);
+		}
+		else {
+			in->read_image(TypeDesc::UINT8, (uchar*)pixels);
+		}
 
 		in->close();
 		delete in;
@@ -339,7 +363,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 	}
 
 	if(components == 2) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = pixels[i*2+1];
 			pixels[i*4+2] = pixels[i*2+0];
 			pixels[i*4+1] = pixels[i*2+0];
@@ -347,7 +371,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 		}
 	}
 	else if(components == 3) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = 255;
 			pixels[i*4+2] = pixels[i*3+2];
 			pixels[i*4+1] = pixels[i*3+1];
@@ -355,7 +379,7 @@ bool ImageManager::file_load_image(Image *img, device_vector<uchar4>& tex_img)
 		}
 	}
 	else if(components == 1) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = 255;
 			pixels[i*4+2] = pixels[i];
 			pixels[i*4+1] = pixels[i];
@@ -372,7 +396,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 		return false;
 
 	ImageInput *in = NULL;
-	int width, height, components;
+	int width, height, depth, components;
 
 	if(!img->builtin_data) {
 		/* load image from file through OIIO */
@@ -391,6 +415,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 		/* we only handle certain number of components */
 		width = spec.width;
 		height = spec.height;
+		depth = spec.depth;
 		components = spec.nchannels;
 	}
 	else {
@@ -399,7 +424,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 			return false;
 
 		bool is_float;
-		builtin_image_info_cb(img->filename, img->builtin_data, is_float, width, height, components);
+		builtin_image_info_cb(img->filename, img->builtin_data, is_float, width, height, depth, components);
 	}
 
 	if(!(components >= 1 && components <= 4)) {
@@ -411,15 +436,21 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 	}
 
 	/* read RGBA pixels */
-	float *pixels = (float*)tex_img.resize(width, height);
-	int scanlinesize = width*components*sizeof(float);
+	float *pixels = (float*)tex_img.resize(width, height, depth);
 
 	if(in) {
-		in->read_image(TypeDesc::FLOAT,
-			(uchar*)pixels + (height-1)*scanlinesize,
-			AutoStride,
-			-scanlinesize,
-			AutoStride);
+		if(depth <= 1) {
+			int scanlinesize = width*components*sizeof(float);
+
+			in->read_image(TypeDesc::FLOAT,
+				(uchar*)pixels + (height-1)*scanlinesize,
+				AutoStride,
+				-scanlinesize,
+				AutoStride);
+		}
+		else {
+			in->read_image(TypeDesc::FLOAT, (uchar*)pixels);
+		}
 
 		in->close();
 		delete in;
@@ -429,7 +460,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 	}
 
 	if(components == 2) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = pixels[i*2+1];
 			pixels[i*4+2] = pixels[i*2+0];
 			pixels[i*4+1] = pixels[i*2+0];
@@ -437,7 +468,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 		}
 	}
 	else if(components == 3) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = 1.0f;
 			pixels[i*4+2] = pixels[i*3+2];
 			pixels[i*4+1] = pixels[i*3+1];
@@ -445,7 +476,7 @@ bool ImageManager::file_load_float_image(Image *img, device_vector<float4>& tex_
 		}
 	}
 	else if(components == 1) {
-		for(int i = width*height-1; i >= 0; i--) {
+		for(int i = width*height*depth-1; i >= 0; i--) {
 			pixels[i*4+3] = 1.0f;
 			pixels[i*4+2] = pixels[i];
 			pixels[i*4+1] = pixels[i];
@@ -460,9 +491,7 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, int sl
 {
 	if(progress->get_cancel())
 		return;
-	if(osl_texture_system)
-		return;
-
+	
 	Image *img;
 	bool is_float;
 
@@ -475,6 +504,9 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, int sl
 		is_float = true;
 	}
 
+	if(osl_texture_system && !img->builtin_data)
+		return;
+
 	if(is_float) {
 		string filename = path_filename(float_images[slot]->filename);
 		progress->set_status("Updating Images", "Loading " + filename);
@@ -556,7 +588,7 @@ void ImageManager::device_free_image(Device *device, DeviceScene *dscene, int sl
 	}
 
 	if(img) {
-		if(osl_texture_system) {
+		if(osl_texture_system && !img->builtin_data) {
 #ifdef WITH_OSL
 			ustring filename(images[slot]->filename);
 			((OSL::TextureSystem*)osl_texture_system)->invalidate(filename);
@@ -606,7 +638,7 @@ void ImageManager::device_update(Device *device, DeviceScene *dscene, Progress&
 			device_free_image(device, dscene, slot + tex_image_byte_start);
 		}
 		else if(images[slot]->need_load) {
-			if(!osl_texture_system) 
+			if(!osl_texture_system || images[slot]->builtin_data) 
 				pool.push(function_bind(&ImageManager::device_load_image, this, device, dscene, slot + tex_image_byte_start, &progress));
 		}
 	}
@@ -619,7 +651,7 @@ void ImageManager::device_update(Device *device, DeviceScene *dscene, Progress&
 			device_free_image(device, dscene, slot);
 		}
 		else if(float_images[slot]->need_load) {
-			if(!osl_texture_system) 
+			if(!osl_texture_system || float_images[slot]->builtin_data) 
 				pool.push(function_bind(&ImageManager::device_load_image, this, device, dscene, slot, &progress));
 		}
 	}
@@ -657,6 +689,7 @@ void ImageManager::device_pack_images(Device *device, DeviceScene *dscene, Progr
 
 		device_vector<uchar4>& tex_img = dscene->tex_image[slot];
 
+		/* todo: support 3D textures, only CPU for now */
 
 		/* The image options are packed
 		   bit 0 -> periodic
diff --git a/intern/cycles/render/image.h b/intern/cycles/render/image.h
index 85b6b512bae..a52ab2853b9 100644
--- a/intern/cycles/render/image.h
+++ b/intern/cycles/render/image.h
@@ -50,6 +50,7 @@ public:
 	~ImageManager();
 
 	int add_image(const string& filename, void *builtin_data, bool animated, bool& is_float, bool& is_linear, InterpolationType interpolation);
+	void remove_image(int slot);
 	void remove_image(const string& filename, void *builtin_data, InterpolationType interpolation);
 	bool is_float_image(const string& filename, void *builtin_data, bool& is_linear);
 
@@ -63,15 +64,9 @@ public:
 
 	bool need_update;
 
-	boost::function<void(const string &filename, void *data, bool &is_float, int &width, int &height, int &channels)> builtin_image_info_cb;
+	boost::function<void(const string &filename, void *data, bool &is_float, int &width, int &height, int &depth, int &channels)> builtin_image_info_cb;
 	boost::function<bool(const string &filename, void *data, unsigned char *pixels)> builtin_image_pixels_cb;
 	boost::function<bool(const string &filename, void *data, float *pixels)> builtin_image_float_pixels_cb;
-private:
-	int tex_num_images;
-	int tex_num_float_images;
-	int tex_image_byte_start;
-	thread_mutex device_mutex;
-	int animation_frame;
 
 	struct Image {
 		string filename;
@@ -84,6 +79,13 @@ private:
 		int users;
 	};
 
+private:
+	int tex_num_images;
+	int tex_num_float_images;
+	int tex_image_byte_start;
+	thread_mutex device_mutex;
+	int animation_frame;
+
 	vector<Image*> images;
 	vector<Image*> float_images;
 	void *osl_texture_system;