Cycles: OpenCL image texture support, fix an attribute node issue and refactor

feature enabling #defines a bit.

19 files changed, 222 insertions, 23 deletions

diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index 87f255e54e7..b17abac2a1b 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -53,6 +53,7 @@ public:
 	int num;
 	bool display_device;
 	bool advanced_shading;
+	bool pack_images;
 	vector<DeviceInfo> multi_devices;
 
 	DeviceInfo()
@@ -62,6 +63,7 @@ public:
 		num = 0;
 		display_device = false;
 		advanced_shading = true;
+		pack_images = false;
 	}
 };
 
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 07988d32aff..5f422332cd2 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -260,6 +260,7 @@ void device_cpu_info(vector<DeviceInfo>& devices)
 	info.id = "CPU";
 	info.num = 0;
 	info.advanced_shading = true;
+	info.pack_images = false;
 
 	devices.insert(devices.begin(), info);
 }
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 0a780e5f576..937dbf2d87c 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -877,6 +877,7 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 		int major, minor;
 		cuDeviceComputeCapability(&major, &minor, num);
 		info.advanced_shading = (major >= 2);
+		info.pack_images = false;
 
 		/* if device has a kernel timeout, assume it is used for display */
 		if(cuDeviceGetAttribute(&attr, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, num) == CUDA_SUCCESS && attr == 1) {
diff --git a/intern/cycles/device/device_multi.cpp b/intern/cycles/device/device_multi.cpp
index 9f7d65e640b..83e69b98f5d 100644
--- a/intern/cycles/device/device_multi.cpp
+++ b/intern/cycles/device/device_multi.cpp
@@ -304,6 +304,7 @@ static bool device_multi_add(vector<DeviceInfo>& devices, DeviceType type, bool
 	int num_added = 0, num_display = 0;
 
 	info.advanced_shading = with_advanced_shading;
+	info.pack_images = false;
 
 	foreach(DeviceInfo& subinfo, devices) {
 		if(subinfo.type == type) {
@@ -326,6 +327,7 @@ static bool device_multi_add(vector<DeviceInfo>& devices, DeviceType type, bool
 			info.multi_devices.push_back(subinfo);
 			if(subinfo.display_device)
 				info.display_device = true;
+			info.pack_images = info.pack_images || subinfo.pack_images;
 			num_added++;
 		}
 	}
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index 322b7ad3eb9..09235506048 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -212,7 +212,7 @@ public:
 	{
 		char version[256];
 
-		int major, minor, req_major = 1, req_minor = 1;
+		int major, minor, req_major = 1, req_minor = 0;
 
 		clGetPlatformInfo(cpPlatform, CL_PLATFORM_VERSION, sizeof(version), &version, NULL);
 
@@ -300,15 +300,15 @@ public:
 		
 		/* Multi Closure for nVidia cards */
 		if(platform_name == "NVIDIA CUDA")
-			build_options += "-D__KERNEL_SHADING__ -D__MULTI_CLOSURE__ -cl-nv-maxrregcount=24 -cl-nv-verbose ";
+			build_options += "-D__KERNEL_SHADING__ -D__KERNEL_OPENCL_NVIDIA__ -cl-nv-maxrregcount=24 -cl-nv-verbose ";
 			
 		/* No Float3 for Apple */
 		else if(platform_name == "Apple")
-			build_options += "-D__CL_NO_FLOAT3__ ";
+			build_options += "-D__CL_NO_FLOAT3__ -D__KERNEL_OPENCL_APPLE__ ";
 			
 		/* Basic shading for AMD cards (non Apple) */
 		else if(platform_name == "AMD Accelerated Parallel Processing")
-			build_options += "-D__KERNEL_SHADING__ -D__CL_NO_FLOAT3__ ";
+			build_options += "-D__CL_NO_FLOAT3__ -D__KERNEL_OPENCL_AMD__ ";
 
 		return build_options;
 	}
@@ -743,6 +743,7 @@ void device_opencl_info(vector<DeviceInfo>& devices)
 		/* we don't know if it's used for display, but assume it is */
 		info.display_device = true;
 		info.advanced_shading = false;
+		info.pack_images = true;
 
 		devices.push_back(info);
 	}
diff --git a/intern/cycles/kernel/kernel_compat_cpu.h b/intern/cycles/kernel/kernel_compat_cpu.h
index 79f894bfdac..2bd0b61b4fa 100644
--- a/intern/cycles/kernel/kernel_compat_cpu.h
+++ b/intern/cycles/kernel/kernel_compat_cpu.h
@@ -146,6 +146,7 @@ typedef texture<float> texture_float;
 typedef texture<uint> texture_uint;
 typedef texture<int> texture_int;
 typedef texture<uint4> texture_uint4;
+typedef texture<uchar4> texture_uchar4;
 typedef texture_image<float4> texture_image_float4;
 typedef texture_image<uchar4> texture_image_uchar4;
 
diff --git a/intern/cycles/kernel/kernel_compat_cuda.h b/intern/cycles/kernel/kernel_compat_cuda.h
index cc719bfadbc..2f9f2c45e88 100644
--- a/intern/cycles/kernel/kernel_compat_cuda.h
+++ b/intern/cycles/kernel/kernel_compat_cuda.h
@@ -50,6 +50,7 @@ typedef texture<float, 1> texture_float;
 typedef texture<uint, 1> texture_uint;
 typedef texture<int, 1> texture_int;
 typedef texture<uint4, 1> texture_uint4;
+typedef texture<uchar4, 1> texture_uchar4;
 typedef texture<float4, 2> texture_image_float4;
 typedef texture<uchar4, 2, cudaReadModeNormalizedFloat> texture_image_uchar4;
 
diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index cd9557bd0bf..e9e7fbd4ca1 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -59,6 +59,7 @@ __device float3 area_light_sample(float3 axisu, float3 axisv, float randu, float
 	return axisu*randu + axisv*randv;
 }
 
+#ifdef __BACKGROUND_MIS__
 __device float3 background_light_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
 {
 	/* for the following, the CDF values are actually a pair of floats, with the
@@ -165,6 +166,7 @@ __device float background_light_pdf(KernelGlobals *kg, float3 direction)
 
 	return pdf * kernel_data.integrator.pdf_lights;
 }
+#endif
 
 __device void regular_light_sample(KernelGlobals *kg, int point,
 	float randu, float randv, float3 P, LightSample *ls, float *pdf)
diff --git a/intern/cycles/kernel/kernel_textures.h b/intern/cycles/kernel/kernel_textures.h
index f4de4c100c4..4ab2574c8ad 100644
--- a/intern/cycles/kernel/kernel_textures.h
+++ b/intern/cycles/kernel/kernel_textures.h
@@ -7,7 +7,6 @@
 #define KERNEL_IMAGE_TEX(type, ttype, name)
 #endif
 
-
 /* bvh */
 KERNEL_TEX(float4, texture_float4, __bvh_nodes)
 KERNEL_TEX(float4, texture_float4, __tri_woop)
@@ -151,6 +150,10 @@ KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_097)
 KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_098)
 KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float_099)
 
+/* packed image (opencl) */
+KERNEL_TEX(uchar4, texture_uchar4, __tex_image_packed)
+KERNEL_TEX(uint4, texture_uint4, __tex_image_packed_info)
+
 #undef KERNEL_TEX
 #undef KERNEL_IMAGE_TEX
 
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index edca9f8d34d..a64c850d35a 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -49,8 +49,30 @@ CCL_NAMESPACE_BEGIN
 #endif
 
 #ifdef __KERNEL_OPENCL__
-//#define __KERNEL_SHADING__
-//#define __KERNEL_ADV_SHADING__
+
+#ifdef __KERNEL_OPENCL_NVIDIA__
+#define __KERNEL_SHADING__
+#define __MULTI_CLOSURE__
+#endif
+
+#ifdef __KERNEL_OPENCL_APPLE__
+//#define __SVM__
+//#define __EMISSION__
+//#define __IMAGE_TEXTURES__
+//#define __HOLDOUT__
+//#define __PROCEDURAL_TEXTURES__
+//#define __EXTRA_NODES__
+#endif
+
+#ifdef __KERNEL_OPENCL_AMD__
+#define __SVM__
+#define __EMISSION__
+#define __IMAGE_TEXTURES__
+#define __HOLDOUT__
+#define __PROCEDURAL_TEXTURES__
+#define __EXTRA_NODES__
+#endif
+
 #endif
 
 /* kernel features */
@@ -69,7 +91,9 @@ CCL_NAMESPACE_BEGIN
 #ifdef __KERNEL_SHADING__
 #define __SVM__
 #define __EMISSION__
-#define __TEXTURES__
+#define __PROCEDURAL_TEXTURES__
+#define __IMAGE_TEXTURES__
+#define __EXTRA_NODES__
 #define __HOLDOUT__
 #endif
 
@@ -85,7 +109,6 @@ CCL_NAMESPACE_BEGIN
 //#define __MULTI_LIGHT__
 //#define __OSL__
 //#define __SOBOL_FULL_SCREEN__
-//#define __MODIFY_TP__
 //#define __QBVH__
 
 /* Shader Evaluation */
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index 5f4d7bbd0c4..12ed61673d1 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -216,13 +216,15 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_JUMP:
 				offset = node.y;
 				break;
-#ifdef __TEXTURES__
+#ifdef __IMAGE_TEXTURES__
 			case NODE_TEX_IMAGE:
 				svm_node_tex_image(kg, sd, stack, node);
 				break;
 			case NODE_TEX_ENVIRONMENT:
 				svm_node_tex_environment(kg, sd, stack, node);
 				break;
+#endif
+#ifdef __PROCEDURAL_TEXTURES__
 			case NODE_TEX_SKY:
 				svm_node_tex_sky(kg, sd, stack, node.y, node.z);
 				break;
@@ -254,6 +256,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_GEOMETRY:
 				svm_node_geometry(sd, stack, node.y, node.z);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_GEOMETRY_BUMP_DX:
 				svm_node_geometry_bump_dx(sd, stack, node.y, node.z);
 				break;
@@ -263,6 +266,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_PATH:
 				svm_node_light_path(sd, stack, node.y, node.z, path_flag);
 				break;
+#endif
 			case NODE_CONVERT:
 				svm_node_convert(sd, stack, node.y, node.z, node.w);
 				break;
@@ -272,6 +276,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_VALUE_V:
 				svm_node_value_v(kg, sd, stack, node.y, &offset);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_INVERT:
 				svm_node_invert(sd, stack, node.y, node.z, node.w);
 				break;
@@ -293,21 +298,25 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_HSV:
 				svm_node_hsv(kg, sd, stack, node.y, node.z, node.w, &offset);
 				break;
+#endif
 			case NODE_ATTR:
 				svm_node_attr(kg, sd, stack, node);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_ATTR_BUMP_DX:
 				svm_node_attr_bump_dx(kg, sd, stack, node);
 				break;
 			case NODE_ATTR_BUMP_DY:
 				svm_node_attr_bump_dy(kg, sd, stack, node);
 				break;
+#endif
 			case NODE_FRESNEL:
 				svm_node_fresnel(sd, stack, node.y, node.z, node.w);
 				break;
 			case NODE_LAYER_WEIGHT:
 				svm_node_layer_weight(sd, stack, node);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_SET_DISPLACEMENT:
 				svm_node_set_displacement(sd, stack, node.y);
 				break;
@@ -323,6 +332,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_NORMAL:
 				svm_node_normal(kg, sd, stack, node.y, node.z, node.w, &offset);
 				break;
+#endif
 			case NODE_MAPPING:
 				svm_node_mapping(kg, sd, stack, node.y, node.z, &offset);
 				break;
@@ -332,15 +342,18 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_TEX_COORD:
 				svm_node_tex_coord(kg, sd, stack, node.y, node.z);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_TEX_COORD_BUMP_DX:
 				svm_node_tex_coord_bump_dx(kg, sd, stack, node.y, node.z);
 				break;
 			case NODE_TEX_COORD_BUMP_DY:
 				svm_node_tex_coord_bump_dy(kg, sd, stack, node.y, node.z);
 				break;
+#endif
 			case NODE_EMISSION_SET_WEIGHT_TOTAL:
 				svm_node_emission_set_weight_total(kg, sd, node.y, node.z, node.w);
 				break;
+#ifdef __EXTRA_NODES__
 			case NODE_RGB_RAMP:
 				svm_node_rgb_ramp(kg, sd, stack, node, &offset);
 				break;
@@ -350,6 +363,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_FALLOFF:
 				svm_node_light_falloff(sd, stack, node);
 				break;
+#endif
 			case NODE_END:
 			default:
 #ifndef __MULTI_CLOSURE__
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index 3a94f08d42f..ed70a6dc423 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -22,7 +22,7 @@ CCL_NAMESPACE_BEGIN
 
 __device void svm_node_attr_init(KernelGlobals *kg, ShaderData *sd,
 	uint4 node, NodeAttributeType *type,
-	NodeAttributeType *mesh_type, AttributeElement *elem, uint *offset, uint *out_offset)
+	NodeAttributeType *mesh_type, AttributeElement *elem, int *offset, uint *out_offset)
 {
 	if(sd->object != ~0) {
 		/* find attribute by unique id */
@@ -35,7 +35,7 @@ __device void svm_node_attr_init(KernelGlobals *kg, ShaderData *sd,
 
 		/* return result */
 		*elem = (AttributeElement)attr_map.y;
-		*offset = attr_map.z;
+		*offset = as_int(attr_map.z);
 		*mesh_type = (NodeAttributeType)attr_map.w;
 	}
 	else {
@@ -53,7 +53,8 @@ __device void svm_node_attr(KernelGlobals *kg, ShaderData *sd, float *stack, uin
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
@@ -84,7 +85,8 @@ __device void svm_node_attr_bump_dx(KernelGlobals *kg, ShaderData *sd, float *st
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
@@ -119,7 +121,8 @@ __device void svm_node_attr_bump_dy(KernelGlobals *kg, ShaderData *sd, float *st
 {
 	NodeAttributeType type, mesh_type;
 	AttributeElement elem;
-	uint offset, out_offset;
+	uint out_offset;
+	int offset;
 
 	svm_node_attr_init(kg, sd, node, &type, &mesh_type, &elem, &offset, &out_offset);
 
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index eddd0f7034a..388f006c40f 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -18,6 +18,75 @@
 
 CCL_NAMESPACE_BEGIN
 
+#ifdef __KERNEL_OPENCL__
+
+/* For OpenCL all images are packed in a single array, and we do manual lookup
+ * and interpolation. */
+
+__device_inline float4 svm_image_texture_read(KernelGlobals *kg, int offset)
+{
+	uchar4 r = kernel_tex_fetch(__tex_image_packed, offset);
+	float f = 1.0f/255.0f;
+	return make_float4(r.x*f, r.y*f, r.z*f, r.w*f);
+}
+
+__device_inline int svm_image_texture_wrap_periodic(int x, int width)
+{
+	x %= width;
+	if(x < 0)
+		x += width;
+	return x;
+}
+
+__device_inline int svm_image_texture_wrap_clamp(int x, int width)
+{
+	return clamp(x, 0, width-1);
+}
+
+__device_inline float svm_image_texture_frac(float x, int *ix)
+{
+	int i = (int)x - ((x < 0.0f)? 1: 0);
+	*ix = i;
+	return x - (float)i;
+}
+
+__device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
+{
+	uint4 info = kernel_tex_fetch(__tex_image_packed_info, id);
+	uint width = info.x;
+	uint height = info.y;
+	uint offset = info.z;
+	uint periodic = info.w;
+
+	int ix, iy, nix, niy;
+	float tx = svm_image_texture_frac(x*width, &ix);
+	float ty = svm_image_texture_frac(y*height, &iy);
+
+	if(periodic) {
+		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 {
+		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);
+	}
+
+	float4 r = (1.0f - ty)*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + iy*width);
+	r += (1.0f - ty)*tx*svm_image_texture_read(kg, offset + nix + iy*width);
+	r += ty*(1.0f - tx)*svm_image_texture_read(kg, offset + ix + niy*width);
+	r += ty*tx*svm_image_texture_read(kg, offset + nix + niy*width);
+
+	return r;
+}
+
+#else
+
 __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 {
 	float4 r;
@@ -31,9 +100,6 @@ __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 	   also note that cuda has 128 textures limit, we use 100 now, since
 	   we still need some for other storage */
 
-#ifdef __KERNEL_OPENCL__
-	r = make_float4(0.0f, 0.0f, 0.0f, 0.0f); /* todo */
-#else
 	switch(id) {
 		case 0: r = kernel_tex_image_interp(__tex_image_000, x, y); break;
 		case 1: r = kernel_tex_image_interp(__tex_image_001, x, y); break;
@@ -139,11 +205,12 @@ __device float4 svm_image_texture(KernelGlobals *kg, int id, float x, float y)
 			kernel_assert(0);
 			return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 	}
-#endif
 
 	return r;
 }
 
+#endif
+
 __device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
 	uint id = node.y;
diff --git a/intern/cycles/render/image.cpp b/intern/cycles/render/image.cpp
index b9e02467450..6417d0e2103 100644
--- a/intern/cycles/render/image.cpp
+++ b/intern/cycles/render/image.cpp
@@ -34,6 +34,7 @@ CCL_NAMESPACE_BEGIN
 ImageManager::ImageManager()
 {
 	need_update = true;
+	pack_images = false;
 	osl_texture_system = NULL;
 }
 
@@ -45,6 +46,11 @@ ImageManager::~ImageManager()
 		assert(!float_images[slot]);
 }
 
+void ImageManager::set_pack_images(bool pack_images_)
+{
+	pack_images = pack_images_;
+}
+
 void ImageManager::set_osl_texture_system(void *texture_system)
 {
 	osl_texture_system = texture_system;
@@ -84,7 +90,7 @@ int ImageManager::add_image(const string& filename, bool& is_float)
 	size_t slot;
 
 	/* load image info and find out if we need a float texture */
-	is_float = is_float_image(filename);
+	is_float = (pack_images)? false: is_float_image(filename);
 
 	if(is_float) {
 		/* find existing image */
@@ -361,7 +367,8 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, int sl
 		if(slot >= 10) name = string_printf("__tex_image_float_0%d", slot);
 		else name = string_printf("__tex_image_float_00%d", slot);
 
-		device->tex_alloc(name.c_str(), tex_img, true, true);
+		if(!pack_images)
+			device->tex_alloc(name.c_str(), tex_img, true, true);
 	}
 	else {
 		string filename = path_filename(images[slot]->filename);
@@ -387,7 +394,8 @@ void ImageManager::device_load_image(Device *device, DeviceScene *dscene, int sl
 		if(slot >= 10) name = string_printf("__tex_image_0%d", slot);
 		else name = string_printf("__tex_image_00%d", slot);
 
-		device->tex_alloc(name.c_str(), tex_img, true, true);
+		if(!pack_images)
+			device->tex_alloc(name.c_str(), tex_img, true, true);
 	}
 
 	img->need_load = false;
@@ -466,9 +474,49 @@ void ImageManager::device_update(Device *device, DeviceScene *dscene, Progress&
 
 	pool.wait_work();
 
+	if(pack_images)
+		device_pack_images(device, dscene, progress);
+
 	need_update = false;
 }
 
+void ImageManager::device_pack_images(Device *device, DeviceScene *dscene, Progress& progess)
+{
+	/* for OpenCL, we pack all image textures inside a single big texture, and
+	   will do our own interpolation in the kernel */
+	size_t size = 0;
+
+	for(size_t slot = 0; slot < images.size(); slot++) {
+		if(!images[slot])
+			continue;
+
+		device_vector<uchar4>& tex_img = dscene->tex_image[slot];
+		size += tex_img.size();
+	}
+
+	uint4 *info = dscene->tex_image_packed_info.resize(images.size());
+	uchar4 *pixels = dscene->tex_image_packed.resize(size);
+
+	size_t offset = 0;
+
+	for(size_t slot = 0; slot < images.size(); slot++) {
+		if(!images[slot])
+			continue;
+
+		device_vector<uchar4>& tex_img = dscene->tex_image[slot];
+
+		info[slot] = make_uint4(tex_img.data_width, tex_img.data_height, offset, 1);
+
+		memcpy(pixels+offset, (void*)tex_img.data_pointer, tex_img.memory_size());
+		offset += tex_img.size();
+	}
+
+	if(dscene->tex_image_packed.size())
+		device->tex_alloc("__tex_image_packed", dscene->tex_image_packed);
+	if(dscene->tex_image_packed_info.size())
+		device->tex_alloc("__tex_image_packed_info", dscene->tex_image_packed_info);
+}
+
 void ImageManager::device_free(Device *device, DeviceScene *dscene)
 {
 	for(size_t slot = 0; slot < images.size(); slot++)
@@ -476,6 +524,12 @@ void ImageManager::device_free(Device *device, DeviceScene *dscene)
 	for(size_t slot = 0; slot < float_images.size(); slot++)
 		device_free_image(device, dscene, slot + TEX_IMAGE_FLOAT_START);
 
+	device->tex_free(dscene->tex_image_packed);
+	dscene->tex_image_packed.clear();
+
+	device->tex_free(dscene->tex_image_packed_info);
+	dscene->tex_image_packed_info.clear();
+
 	images.clear();
 	float_images.clear();
 }
diff --git a/intern/cycles/render/image.h b/intern/cycles/render/image.h
index cc01b4a8e4c..2b5e53cabe1 100644
--- a/intern/cycles/render/image.h
+++ b/intern/cycles/render/image.h
@@ -47,6 +47,7 @@ public:
 	void device_free(Device *device, DeviceScene *dscene);
 
 	void set_osl_texture_system(void *texture_system);
+	void set_pack_images(bool pack_images_);
 
 	bool need_update;
 
@@ -61,12 +62,15 @@ private:
 	vector<Image*> images;
 	vector<Image*> float_images;
 	void *osl_texture_system;
+	bool pack_images;
 
 	bool file_load_image(Image *img, device_vector<uchar4>& tex_img);
 	bool file_load_float_image(Image *img, device_vector<float4>& tex_img);
 
 	void device_load_image(Device *device, DeviceScene *dscene, int slot, Progress *progess);
 	void device_free_image(Device *device, DeviceScene *dscene, int slot);
+
+	void device_pack_images(Device *device, DeviceScene *dscene, Progress& progess);
 };
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 0422f97a706..cabbd5760c2 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -421,7 +421,7 @@ void MeshManager::update_svm_attributes(Device *device, DeviceScene *dscene, Sce
 
 			attr_map[index].x = id;
 			attr_map[index].y = req.element;
-			attr_map[index].z = req.offset;
+			attr_map[index].z = as_uint(req.offset);
 
 			if(req.type == TypeDesc::TypeFloat)
 				attr_map[index].w = NODE_ATTR_FLOAT;
diff --git a/intern/cycles/render/scene.cpp b/intern/cycles/render/scene.cpp
index b6453339d41..a5f90bfe34b 100644
--- a/intern/cycles/render/scene.cpp
+++ b/intern/cycles/render/scene.cpp
@@ -111,6 +111,8 @@ void Scene::device_update(Device *device_, Progress& progress)
 	 * - Displacement shader must have all shader data available.
 	 * - Light manager needs final mesh data to compute emission CDF.
 	 */
+	
+	image_manager->set_pack_images(device->info.pack_images);
 
 	progress.set_status("Updating Background");
 	background->device_update(device, &dscene, this);
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index ca4d9fc9625..90bc47d5c8e 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -97,6 +97,10 @@ public:
 	device_vector<uchar4> tex_image[TEX_NUM_IMAGES];
 	device_vector<float4> tex_float_image[TEX_NUM_FLOAT_IMAGES];
 
+	/* opencl images */
+	device_vector<uchar4> tex_image_packed;
+	device_vector<uint4> tex_image_packed_info;
+
 	KernelData data;
 };
 
diff --git a/intern/cycles/util/util_math.h b/intern/cycles/util/util_math.h
index 8c0e7105b22..fee2f10085b 100644
--- a/intern/cycles/util/util_math.h
+++ b/intern/cycles/util/util_math.h
@@ -965,6 +965,20 @@ __device_inline void print_int4(const char *label, const int4& a)
 
 #ifndef __KERNEL_OPENCL__
 
+__device_inline unsigned int as_int(uint i)
+{
+	union { unsigned int ui; int i; } u;
+	u.ui = i;
+	return u.i;
+}
+
+__device_inline unsigned int as_uint(int i)
+{
+	union { unsigned int ui; int i; } u;
+	u.i = i;
+	return u.ui;
+}
+
 __device_inline unsigned int as_uint(float f)
 {
 	union { unsigned int i; float f; } u;