Cycles: per-BSDF normal input and new Bump node.

Each BSDF node now has a Normal input, which can be used to set a custom normal
for the BSDF, for example if you want to have only bump on one of the layers in
a multilayer material.

The Bump node can be used to generate a normal from a scalar value, the same as
what happens when you connect a scalar value to the displacement output.

Documentation has been updated with the latest changes:
http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Nodes

Patch by Agustin Benavidez, some implementation tweaks by me.

18 files changed, 224 insertions, 45 deletions

diff --git a/intern/cycles/blender/blender_shader.cpp b/intern/cycles/blender/blender_shader.cpp
index 56548613647..98ac8692f36 100644
--- a/intern/cycles/blender/blender_shader.cpp
+++ b/intern/cycles/blender/blender_shader.cpp
@@ -402,6 +402,10 @@ static ShaderNode *add_node(BL::BlendData b_data, BL::Scene b_scene, ShaderGraph
 			node = new ParticleInfoNode();
 			break;
 		}
+		case BL::ShaderNode::type_BUMP: {
+			node = new BumpNode();
+			break;
+		}
 		case BL::ShaderNode::type_TEX_IMAGE: {
 			BL::ShaderNodeTexImage b_image_node(b_node);
 			BL::Image b_image(b_image_node.image());
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index ae9b0ac6d99..be49aa54e47 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -370,6 +370,7 @@ typedef struct ShaderClosure {
 #endif
 	float data0;
 	float data1;
+	float3 N;
 
 } ShaderClosure;
 
diff --git a/intern/cycles/kernel/svm/bsdf_ashikhmin_velvet.h b/intern/cycles/kernel/svm/bsdf_ashikhmin_velvet.h
index 40249dbe9c6..785fc038e9e 100644
--- a/intern/cycles/kernel/svm/bsdf_ashikhmin_velvet.h
+++ b/intern/cycles/kernel/svm/bsdf_ashikhmin_velvet.h
@@ -58,7 +58,7 @@ __device void bsdf_ashikhmin_velvet_blur(ShaderClosure *sc, float roughness)
 __device float3 bsdf_ashikhmin_velvet_eval_reflect(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
 	float m_invsigma2 = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cosNO = dot(m_N, I);
 	float cosNI = dot(m_N, omega_in);
@@ -106,7 +106,7 @@ __device float bsdf_ashikhmin_velvet_albedo(const ShaderData *sd, const ShaderCl
 __device int bsdf_ashikhmin_velvet_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
 	float m_invsigma2 = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// we are viewing the surface from above - send a ray out with uniform
 	// distribution over the hemisphere
diff --git a/intern/cycles/kernel/svm/bsdf_diffuse.h b/intern/cycles/kernel/svm/bsdf_diffuse.h
index edf8dd93341..6a67c184854 100644
--- a/intern/cycles/kernel/svm/bsdf_diffuse.h
+++ b/intern/cycles/kernel/svm/bsdf_diffuse.h
@@ -53,7 +53,7 @@ __device void bsdf_diffuse_blur(ShaderClosure *sc, float roughness)
 
 __device float3 bsdf_diffuse_eval_reflect(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cos_pi = fmaxf(dot(m_N, omega_in), 0.0f) * M_1_PI_F;
 	*pdf = cos_pi;
@@ -72,7 +72,7 @@ __device float bsdf_diffuse_albedo(const ShaderData *sd, const ShaderClosure *sc
 
 __device int bsdf_diffuse_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// distribution over the hemisphere
 	sample_cos_hemisphere(m_N, randu, randv, omega_in, pdf);
@@ -116,7 +116,7 @@ __device float3 bsdf_translucent_eval_reflect(const ShaderData *sd, const Shader
 
 __device float3 bsdf_translucent_eval_transmit(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cos_pi = fmaxf(-dot(m_N, omega_in), 0.0f) * M_1_PI_F;
 	*pdf = cos_pi;
@@ -130,7 +130,7 @@ __device float bsdf_translucent_albedo(const ShaderData *sd, const ShaderClosure
 
 __device int bsdf_translucent_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// we are viewing the surface from the right side - send a ray out with cosine
 	// distribution over the hemisphere
diff --git a/intern/cycles/kernel/svm/bsdf_microfacet.h b/intern/cycles/kernel/svm/bsdf_microfacet.h
index d8f0310bd02..60f8e23818c 100644
--- a/intern/cycles/kernel/svm/bsdf_microfacet.h
+++ b/intern/cycles/kernel/svm/bsdf_microfacet.h
@@ -76,7 +76,7 @@ __device float3 bsdf_microfacet_ggx_eval_reflect(const ShaderData *sd, const Sha
 	float m_ag = sc->data0;
 	//float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	if(m_refractive) return make_float3 (0, 0, 0);
 	float cosNO = dot(m_N, I);
@@ -113,7 +113,7 @@ __device float3 bsdf_microfacet_ggx_eval_transmit(const ShaderData *sd, const Sh
 	float m_ag = sc->data0;
 	float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	if(!m_refractive) return make_float3 (0, 0, 0);
 	float cosNO = dot(m_N, I);
@@ -154,7 +154,7 @@ __device int bsdf_microfacet_ggx_sample(const ShaderData *sd, const ShaderClosur
 	float m_ag = sc->data0;
 	float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cosNO = dot(m_N, sd->I);
 	if(cosNO > 0) {
@@ -302,7 +302,7 @@ __device float3 bsdf_microfacet_beckmann_eval_reflect(const ShaderData *sd, cons
 	float m_ab = sc->data0;
 	//float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	if(m_refractive) return make_float3 (0, 0, 0);
 	float cosNO = dot(m_N, I);
@@ -341,7 +341,7 @@ __device float3 bsdf_microfacet_beckmann_eval_transmit(const ShaderData *sd, con
 	float m_ab = sc->data0;
 	float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	if(!m_refractive) return make_float3 (0, 0, 0);
 	float cosNO = dot(m_N, I);
@@ -384,7 +384,7 @@ __device int bsdf_microfacet_beckmann_sample(const ShaderData *sd, const ShaderC
 	float m_ab = sc->data0;
 	float m_eta = sc->data1;
 	int m_refractive = sc->type == CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cosNO = dot(m_N, sd->I);
 	if(cosNO > 0) {
diff --git a/intern/cycles/kernel/svm/bsdf_oren_nayar.h b/intern/cycles/kernel/svm/bsdf_oren_nayar.h
index a7edccdc423..04fbb091fb4 100644
--- a/intern/cycles/kernel/svm/bsdf_oren_nayar.h
+++ b/intern/cycles/kernel/svm/bsdf_oren_nayar.h
@@ -57,9 +57,9 @@ __device void bsdf_oren_nayar_blur(ShaderClosure *sc, float roughness)
 
 __device float3 bsdf_oren_nayar_eval_reflect(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
-	if (dot(sd->N, omega_in) > 0.0f) {
+	if (dot(sc->N, omega_in) > 0.0f) {
 		*pdf = 0.5f * M_1_PI_F;
-		return bsdf_oren_nayar_get_intensity(sc, sd->N, I, omega_in);
+		return bsdf_oren_nayar_get_intensity(sc, sc->N, I, omega_in);
 	}
 	else {
 		*pdf = 0.0f;
@@ -79,15 +79,15 @@ __device float bsdf_oren_nayar_albedo(const ShaderData *sd, const ShaderClosure
 
 __device int bsdf_oren_nayar_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
-	sample_uniform_hemisphere(sd->N, randu, randv, omega_in, pdf);
+	sample_uniform_hemisphere(sc->N, randu, randv, omega_in, pdf);
 
 	if (dot(sd->Ng, *omega_in) > 0.0f) {
-		*eval = bsdf_oren_nayar_get_intensity(sc, sd->N, sd->I, *omega_in);
+		*eval = bsdf_oren_nayar_get_intensity(sc, sc->N, sd->I, *omega_in);
 
 #ifdef __RAY_DIFFERENTIALS__
 		// TODO: find a better approximation for the bounce
-		*domega_in_dx = (2.0f * dot(sd->N, sd->dI.dx)) * sd->N - sd->dI.dx;
-		*domega_in_dy = (2.0f * dot(sd->N, sd->dI.dy)) * sd->N - sd->dI.dy;
+		*domega_in_dx = (2.0f * dot(sc->N, sd->dI.dx)) * sc->N - sd->dI.dx;
+		*domega_in_dy = (2.0f * dot(sc->N, sd->dI.dy)) * sc->N - sd->dI.dy;
 		*domega_in_dx *= 125.0f;
 		*domega_in_dy *= 125.0f;
 #endif
diff --git a/intern/cycles/kernel/svm/bsdf_reflection.h b/intern/cycles/kernel/svm/bsdf_reflection.h
index 09b4e0e48f0..7ce38050a9a 100644
--- a/intern/cycles/kernel/svm/bsdf_reflection.h
+++ b/intern/cycles/kernel/svm/bsdf_reflection.h
@@ -69,7 +69,7 @@ __device float bsdf_reflection_albedo(const ShaderData *sd, const ShaderClosure
 __device int bsdf_reflection_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
 	//const BsdfReflectionClosure *self = (const BsdfReflectionClosure*)sc->data;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// only one direction is possible
 	float cosNO = dot(m_N, sd->I);
diff --git a/intern/cycles/kernel/svm/bsdf_refraction.h b/intern/cycles/kernel/svm/bsdf_refraction.h
index c9c268999c0..7579a4c6276 100644
--- a/intern/cycles/kernel/svm/bsdf_refraction.h
+++ b/intern/cycles/kernel/svm/bsdf_refraction.h
@@ -71,7 +71,7 @@ __device float bsdf_refraction_albedo(const ShaderData *sd, const ShaderClosure
 __device int bsdf_refraction_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
 	float m_eta = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float3 R, T;
 #ifdef __RAY_DIFFERENTIALS__
diff --git a/intern/cycles/kernel/svm/bsdf_ward.h b/intern/cycles/kernel/svm/bsdf_ward.h
index 56b8ad72c89..3c08a3d54d3 100644
--- a/intern/cycles/kernel/svm/bsdf_ward.h
+++ b/intern/cycles/kernel/svm/bsdf_ward.h
@@ -66,7 +66,7 @@ __device float3 bsdf_ward_eval_reflect(const ShaderData *sd, const ShaderClosure
 {
 	float m_ax = sc->data0;
 	float m_ay = sc->data1;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 	float3 m_T = sd->T;
 
 	float cosNO = dot(m_N, I);
@@ -108,7 +108,7 @@ __device int bsdf_ward_sample(const ShaderData *sd, const ShaderClosure *sc, flo
 {
 	float m_ax = sc->data0;
 	float m_ay = sc->data1;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 	float3 m_T = sd->T;
 
 	float cosNO = dot(m_N, sd->I);
diff --git a/intern/cycles/kernel/svm/bsdf_westin.h b/intern/cycles/kernel/svm/bsdf_westin.h
index 3e7c27f44a4..75f935e3267 100644
--- a/intern/cycles/kernel/svm/bsdf_westin.h
+++ b/intern/cycles/kernel/svm/bsdf_westin.h
@@ -62,7 +62,7 @@ __device void bsdf_westin_backscatter_blur(ShaderClosure *sc, float roughness)
 __device float3 bsdf_westin_backscatter_eval_reflect(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
 	float m_invroughness = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// pdf is implicitly 0 (no indirect sampling)
 	float cosNO = dot(m_N, I);
@@ -89,7 +89,7 @@ __device float bsdf_westin_backscatter_albedo(const ShaderData *sd, const Shader
 __device int bsdf_westin_backscatter_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
 	float m_invroughness = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	float cosNO = dot(m_N, sd->I);
 	if(cosNO > 0) {
@@ -151,7 +151,7 @@ __device void bsdf_westin_sheen_blur(ShaderClosure *sc, float roughness)
 __device float3 bsdf_westin_sheen_eval_reflect(const ShaderData *sd, const ShaderClosure *sc, const float3 I, const float3 omega_in, float *pdf)
 {
 	float m_edginess = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// pdf is implicitly 0 (no indirect sampling)
 	float cosNO = dot(m_N, I);
@@ -178,7 +178,7 @@ __device float bsdf_westin_sheen_albedo(const ShaderData *sd, const ShaderClosur
 __device int bsdf_westin_sheen_sample(const ShaderData *sd, const ShaderClosure *sc, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf)
 {
 	float m_edginess = sc->data0;
-	float3 m_N = sd->N;
+	float3 m_N = sc->N;
 
 	// we are viewing the surface from the right side - send a ray out with cosine
 	// distribution over the hemisphere
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index e5167a4f670..5e22edc4696 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -185,7 +185,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 				break;
 			}
 			case NODE_CLOSURE_BSDF:
-				svm_node_closure_bsdf(kg, sd, stack, node, randb, path_flag);
+				svm_node_closure_bsdf(kg, sd, stack, node, randb, path_flag, &offset);
 				break;
 			case NODE_CLOSURE_EMISSION:
 				svm_node_closure_emission(sd, stack, node);
@@ -342,7 +342,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 				svm_node_set_displacement(sd, stack, node.y);
 				break;
 			case NODE_SET_BUMP:
-				svm_node_set_bump(sd, stack, node.y, node.z, node.w);
+				svm_node_set_bump(kg, sd, stack, node);
 				break;
 			case NODE_MATH:
 				svm_node_math(kg, sd, stack, node.y, node.z, node.w, &offset);
@@ -370,6 +370,9 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_TEX_COORD_BUMP_DY:
 				svm_node_tex_coord_bump_dy(kg, sd, stack, node.y, node.z);
 				break;
+			case NODE_CLOSURE_SET_NORMAL:
+				svm_node_set_normal(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);
@@ -384,7 +387,7 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_FALLOFF:
 				svm_node_light_falloff(sd, stack, node);
 				break;
-#endif
+#endif			
 			case NODE_END:
 			default:
 #ifndef __MULTI_CLOSURE__
diff --git a/intern/cycles/kernel/svm/svm_closure.h b/intern/cycles/kernel/svm/svm_closure.h
index c942d50908c..6a0e6915e99 100644
--- a/intern/cycles/kernel/svm/svm_closure.h
+++ b/intern/cycles/kernel/svm/svm_closure.h
@@ -57,7 +57,7 @@ __device_inline void svm_node_closure_set_mix_weight(ShaderClosure *sc, float mi
 #endif
 }
 
-__device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, float randb, int path_flag)
+__device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, float randb, int path_flag, int *offset)
 {
 	uint type, param1_offset, param2_offset;
 
@@ -66,11 +66,19 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 	decode_node_uchar4(node.y, &type, &param1_offset, &param2_offset, &mix_weight_offset);
 	float mix_weight = (stack_valid(mix_weight_offset)? stack_load_float(stack, mix_weight_offset): 1.0f);
 
+	/* note we read this extra node before weight check, so offset is added */
+	uint4 data_node = read_node(kg, offset);
+
 	if(mix_weight == 0.0f)
 		return;
+
+	float3 N = stack_valid(data_node.y)? stack_load_float3(stack, data_node.y): sd->N; 
 #else
 	decode_node_uchar4(node.y, &type, &param1_offset, &param2_offset, NULL);
 	float mix_weight = 1.0f;
+
+	uint4 data_node = read_node(kg, offset);
+	float3 N = stack_valid(data_node.y)? stack_load_float3(stack, data_node.y): sd->N; 
 #endif
 
 	float param1 = (stack_valid(param1_offset))? stack_load_float(stack, param1_offset): __int_as_float(node.z);
@@ -79,6 +87,7 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 	switch(type) {
 		case CLOSURE_BSDF_DIFFUSE_ID: {
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 
 			float roughness = param1;
@@ -90,12 +99,14 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 		}
 		case CLOSURE_BSDF_TRANSLUCENT_ID: {
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 			bsdf_translucent_setup(sd, sc);
 			break;
 		}
 		case CLOSURE_BSDF_TRANSPARENT_ID: {
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 			bsdf_transparent_setup(sd, sc);
 			break;
@@ -108,6 +119,7 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 				break;
 #endif
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 
 			float roughness = param1;
@@ -134,13 +146,14 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 			eta = (sd->flag & SD_BACKFACING)? 1.0f/eta: eta;
 
 			/* fresnel */
-			float cosNO = dot(sd->N, sd->I);
+			float cosNO = dot(N, sd->I);
 			float fresnel = fresnel_dielectric_cos(cosNO, eta);
 			float roughness = param1;
 
 #ifdef __MULTI_CLOSURE__
 			/* reflection */
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 
 			float3 weight = sc->weight;
 			float sample_weight = sc->sample_weight;
@@ -150,6 +163,7 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 
 			/* refraction */
 			sc = svm_node_closure_get(sd);
+			sc->N = N;
 
 			sc->weight = weight;
 			sc->sample_weight = sample_weight;
@@ -158,6 +172,7 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 			svm_node_glossy_setup(sd, sc, type, eta, roughness, true);
 #else
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 
 			bool refract = (randb > fresnel);
 
@@ -174,17 +189,19 @@ __device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *st
 				break;
 #endif
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 
 			float roughness_u = param1;
 			float roughness_v = param2;
 
-			bsdf_ward_setup(sd, sc, normalize(sd->T), roughness_u, roughness_v);
+			bsdf_ward_setup(sd, sc, sd->T, roughness_u, roughness_v);
 			break;
 		}
 #endif
 		case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID: {
 			ShaderClosure *sc = svm_node_closure_get(sd);
+			sc->N = N;
 			svm_node_closure_set_mix_weight(sc, mix_weight);
 
 			/* sigma */
@@ -425,6 +442,8 @@ __device void svm_node_add_closure(ShaderData *sd, float *stack, uint unused,
 #endif
 }
 
+/* Tangent */
+
 #ifdef __DPDU__
 __device_inline void svm_node_closure_store_tangent(ShaderData *sd, float3 tangent)
 {
@@ -444,5 +463,14 @@ __device void svm_node_closure_tangent(ShaderData *sd, float *stack, uint tangen
 }
 #endif
 
+/* (Bump) normal */
+
+__device void svm_node_set_normal(KernelGlobals *kg, ShaderData *sd, float *stack, uint in_direction, uint out_normal)
+{
+	float3 normal = stack_load_float3(stack, in_direction);
+	sd->N = normal;
+	stack_store_float3(stack, out_normal, normal);
+}
+
 CCL_NAMESPACE_END
 
diff --git a/intern/cycles/kernel/svm/svm_displace.h b/intern/cycles/kernel/svm/svm_displace.h
index b1677f67eca..92f23990ad1 100644
--- a/intern/cycles/kernel/svm/svm_displace.h
+++ b/intern/cycles/kernel/svm/svm_displace.h
@@ -20,23 +20,35 @@ CCL_NAMESPACE_BEGIN
 
 /* Bump Node */
 
-__device void svm_node_set_bump(ShaderData *sd, float *stack, uint c_offset, uint x_offset, uint y_offset)
+__device void svm_node_set_bump(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
 #ifdef __RAY_DIFFERENTIALS__
+	/* get normal input */
+	float3 normal_in = stack_valid(node.y)? stack_load_float3(stack, node.y): sd->N;
+
+	/* get surface tangents from normal */
+	float3 Rx = cross(sd->dP.dy, normal_in);
+	float3 Ry = cross(normal_in, sd->dP.dx);
+
+	/* get bump values */
+	uint c_offset, x_offset, y_offset, intensity_offset;
+	decode_node_uchar4(node.z, &c_offset, &x_offset, &y_offset, &intensity_offset);
+
 	float h_c = stack_load_float(stack, c_offset);
 	float h_x = stack_load_float(stack, x_offset);
 	float h_y = stack_load_float(stack, y_offset);
 
-	float3 Rx = cross(sd->dP.dy, sd->N);
-	float3 Ry = cross(sd->N, sd->dP.dx);
-
+	/* compute surface gradient and determinant */
 	float det = dot(sd->dP.dx, Rx);
 	float3 surfgrad = (h_x - h_c)*Rx + (h_y - h_c)*Ry;
+	float intensity = stack_load_float(stack, intensity_offset);
 
-	surfgrad *= 0.1f; /* todo: remove this factor */
-
+	surfgrad *= intensity;
 	float absdet = fabsf(det);
-	sd->N = normalize(absdet*sd->N - signf(det)*surfgrad);
+
+	/* compute and output perturbed normal */
+	float3 outN = normalize(absdet*normal_in - signf(det)*surfgrad);
+	stack_store_float3(stack, node.w, outN);
 #endif
 }
 
diff --git a/intern/cycles/kernel/svm/svm_types.h b/intern/cycles/kernel/svm/svm_types.h
index 22afa304432..327d9cfa014 100644
--- a/intern/cycles/kernel/svm/svm_types.h
+++ b/intern/cycles/kernel/svm/svm_types.h
@@ -95,6 +95,7 @@ typedef enum NodeType {
 	NODE_TEX_BRICK,
 	NODE_CLOSURE_SET_TANGENT,
 	NODE_CLOSURE_TANGENT,
+	NODE_CLOSURE_SET_NORMAL,
 } NodeType;
 
 typedef enum NodeAttributeType {
diff --git a/intern/cycles/render/graph.cpp b/intern/cycles/render/graph.cpp
index c13102e9567..c1c976dc193 100644
--- a/intern/cycles/render/graph.cpp
+++ b/intern/cycles/render/graph.cpp
@@ -229,6 +229,8 @@ void ShaderGraph::finalize(bool do_bump, bool do_osl)
 	if(!finalized) {
 		clean();
 		default_inputs(do_osl);
+		refine_bump_nodes();
+
 		if(do_bump)
 			bump_from_displacement();
 
@@ -484,6 +486,61 @@ void ShaderGraph::default_inputs(bool do_osl)
 		add(texco);
 }
 
+void ShaderGraph::refine_bump_nodes()
+{
+	/* we transverse the node graph looking for bump nodes, when we find them,
+	 * like in bump_from_displacement(), we copy the sub-graph defined from "bump"
+	 * input to the inputs "center","dx" and "dy" What is in "bump" input is moved
+	 * to "center" input. */
+
+	foreach(ShaderNode *node, nodes) {
+		if(node->name == ustring("bump") && node->input("Height")->link) {
+			ShaderInput *bump_input = node->input("Height");
+			set<ShaderNode*> nodes_bump;
+
+			/* make 2 extra copies of the subgraph defined in Bump input */
+			map<ShaderNode*, ShaderNode*> nodes_dx;
+			map<ShaderNode*, ShaderNode*> nodes_dy;
+
+			/* find dependencies for the given input */
+			find_dependencies(nodes_bump, bump_input );
+
+			copy_nodes(nodes_bump, nodes_dx);
+			copy_nodes(nodes_bump, nodes_dy);
+	
+			/* mark nodes to indicate they are use for bump computation, so
+			   that any texture coordinates are shifted by dx/dy when sampling */
+			foreach(ShaderNode *node, nodes_bump)
+				node->bump = SHADER_BUMP_CENTER;
+			foreach(NodePair& pair, nodes_dx)
+				pair.second->bump = SHADER_BUMP_DX;
+			foreach(NodePair& pair, nodes_dy)
+				pair.second->bump = SHADER_BUMP_DY;
+
+			ShaderOutput *out = bump_input->link;
+			ShaderOutput *out_dx = nodes_dx[out->parent]->output(out->name);
+			ShaderOutput *out_dy = nodes_dy[out->parent]->output(out->name);
+
+			connect(out_dx, node->input("SampleX"));
+			connect(out_dy, node->input("SampleY"));
+			
+			/* add generated nodes */
+			foreach(NodePair& pair, nodes_dx)
+				add(pair.second);
+			foreach(NodePair& pair, nodes_dy)
+				add(pair.second);
+			
+			/* connect what is conected is bump to samplecenter input*/
+			connect(out , node->input("SampleCenter"));
+
+			/* bump input is just for connectivity purpose for the graph input,
+			 * we reconected this input to samplecenter, so lets disconnect it
+			 * from bump input */
+			disconnect(bump_input);
+		}
+	}
+}
+
 void ShaderGraph::bump_from_displacement()
 {
 	/* generate bump mapping automatically from displacement. bump mapping is
@@ -497,7 +554,7 @@ void ShaderGraph::bump_from_displacement()
 	 * different shifted coordinates.
 	 *
 	 * these 3 displacement values are then fed into the bump node, which will
-	 * modify the normal. */
+	 * output the the perturbed normal. */
 
 	ShaderInput *displacement_in = output()->input("Displacement");
 
@@ -526,6 +583,12 @@ void ShaderGraph::bump_from_displacement()
 	foreach(NodePair& pair, nodes_dy)
 		pair.second->bump = SHADER_BUMP_DY;
 
+	/* add set normal node and connect the bump normal ouput to the set normal
+	 * output, so it can finally set the shader normal, note we are only doing
+	 * this for bump from displacement, this will be the only bump allowed to
+	 * overwrite the shader normal */
+	ShaderNode *set_normal = add(new SetNormalNode());
+	
 	/* add bump node and connect copied graphs to it */
 	ShaderNode *bump = add(new BumpNode());
 
@@ -537,6 +600,9 @@ void ShaderGraph::bump_from_displacement()
 	connect(out_center, bump->input("SampleCenter"));
 	connect(out_dx, bump->input("SampleX"));
 	connect(out_dy, bump->input("SampleY"));
+	
+	/* connect the bump out to the set normal in: */
+	connect(bump->output("Normal"), set_normal->input("Direction"));
 
 	/* connect bump output to normal input nodes that aren't set yet. actually
 	 * this will only set the normal input to the geometry node that we created
@@ -544,8 +610,14 @@ void ShaderGraph::bump_from_displacement()
 	foreach(ShaderNode *node, nodes)
 		foreach(ShaderInput *input, node->inputs)
 			if(!input->link && input->default_value == ShaderInput::NORMAL)
-				connect(bump->output("Normal"), input);
-	
+				connect(set_normal->output("Normal"), input);
+
+	/* for displacement bump, clear the normal input in case the above loop
+	 * connected the setnormal out to the bump normalin */
+	ShaderInput *bump_normal_in = bump->input("NormalIn");
+	if(bump_normal_in)
+		bump_normal_in->link = NULL;
+
 	/* finally, add the copied nodes to the graph. we can't do this earlier
 	 * because we would create dependency cycles in the above loop */
 	foreach(NodePair& pair, nodes_center)
diff --git a/intern/cycles/render/graph.h b/intern/cycles/render/graph.h
index d485ed02150..acafe1d5278 100644
--- a/intern/cycles/render/graph.h
+++ b/intern/cycles/render/graph.h
@@ -235,6 +235,7 @@ protected:
 	void break_cycles(ShaderNode *node, vector<bool>& visited, vector<bool>& on_stack);
 	void clean();
 	void bump_from_displacement();
+	void refine_bump_nodes();
 	void default_inputs(bool do_osl);
 };
 
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index da31a528310..b4ef93a4ff8 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -1175,7 +1175,7 @@ BsdfNode::BsdfNode()
 	closure = ccl::CLOSURE_BSDF_DIFFUSE_ID;
 
 	add_input("Color", SHADER_SOCKET_COLOR, make_float3(0.8f, 0.8f, 0.8f));
-	add_input("Normal", SHADER_SOCKET_NORMAL, ShaderInput::NORMAL, true);
+	add_input("Normal", SHADER_SOCKET_NORMAL, ShaderInput::NORMAL);
 
 	add_output("BSDF", SHADER_SOCKET_CLOSURE);
 }
@@ -1183,6 +1183,7 @@ BsdfNode::BsdfNode()
 void BsdfNode::compile(SVMCompiler& compiler, ShaderInput *param1, ShaderInput *param2)
 {
 	ShaderInput *color_in = input("Color");
+	ShaderInput *normal_in = input("Normal");
 
 	if(color_in->link) {
 		compiler.stack_assign(color_in);
@@ -1203,6 +1204,10 @@ void BsdfNode::compile(SVMCompiler& compiler, ShaderInput *param1, ShaderInput *
 			compiler.closure_mix_weight_offset()),
 		__float_as_int((param1)? param1->value.x: 0.0f),
 		__float_as_int((param2)? param2->value.x: 0.0f));
+
+	if(normal_in->link)
+		compiler.stack_assign(normal_in);
+	compiler.add_node(NODE_CLOSURE_BSDF, normal_in->stack_offset);
 }
 
 void BsdfNode::compile(SVMCompiler& compiler)
@@ -2215,7 +2220,7 @@ static ShaderEnum mix_type_init()
 	enm.insert("Burn", NODE_MIX_BURN);
 	enm.insert("Hue", NODE_MIX_HUE);
 	enm.insert("Saturation", NODE_MIX_SAT);
-	enm.insert("Value", NODE_MIX_VAL );
+	enm.insert("Value", NODE_MIX_VAL);
 	enm.insert("Color", NODE_MIX_COLOR);
 	enm.insert("Soft Light", NODE_MIX_SOFT);
 	enm.insert("Linear Light", NODE_MIX_LINEAR);
@@ -2586,6 +2591,7 @@ OutputNode::OutputNode()
 	add_input("Surface", SHADER_SOCKET_CLOSURE);
 	add_input("Volume", SHADER_SOCKET_CLOSURE);
 	add_input("Displacement", SHADER_SOCKET_FLOAT);
+	add_input("Normal", SHADER_SOCKET_NORMAL);
 }
 
 void OutputNode::compile(SVMCompiler& compiler)
@@ -2733,9 +2739,15 @@ void VectorMathNode::compile(OSLCompiler& compiler)
 BumpNode::BumpNode()
 : ShaderNode("bump")
 {
+	/* this input is used by the user, but after graph transform it is no longer
+	 * used and moved to sampler center/x/y instead */
+	add_input("Height", SHADER_SOCKET_NORMAL);
+
 	add_input("SampleCenter", SHADER_SOCKET_FLOAT);
 	add_input("SampleX", SHADER_SOCKET_FLOAT);
 	add_input("SampleY", SHADER_SOCKET_FLOAT);
+	add_input("NormalIn", SHADER_SOCKET_NORMAL, ShaderInput::NORMAL);
+	add_input("Strength", SHADER_SOCKET_FLOAT, 0.1f);
 
 	add_output("Normal", SHADER_SOCKET_NORMAL);
 }
@@ -2745,12 +2757,25 @@ void BumpNode::compile(SVMCompiler& compiler)
 	ShaderInput *center_in = input("SampleCenter");
 	ShaderInput *dx_in = input("SampleX");
 	ShaderInput *dy_in = input("SampleY");
+	ShaderInput *normal_in = input("NormalIn");
+	ShaderInput *intensity_in = input("Strength");
+	ShaderOutput *normal_out = output("Normal");
 
 	compiler.stack_assign(center_in);
 	compiler.stack_assign(dx_in);
 	compiler.stack_assign(dy_in);
+	compiler.stack_assign(intensity_in);
+	compiler.stack_assign(normal_out);
 
-	compiler.add_node(NODE_SET_BUMP, center_in->stack_offset, dx_in->stack_offset, dy_in->stack_offset);
+	if(normal_in->link)
+		compiler.stack_assign(normal_in);
+	
+	/* pack all parameters in the node */
+	compiler.add_node(NODE_SET_BUMP,
+		normal_in->stack_offset,
+		compiler.encode_uchar4(center_in->stack_offset, dx_in->stack_offset,
+			dy_in->stack_offset, intensity_in->stack_offset),
+		normal_out->stack_offset);
 }
 
 void BumpNode::compile(OSLCompiler& compiler)
@@ -2819,17 +2844,44 @@ void RGBRampNode::compile(OSLCompiler& compiler)
 	/* NB: cycles float3 type is actually 4 floats! need to use an explicit array */
 	float ramp_color[RAMP_TABLE_SIZE][3];
 	float ramp_alpha[RAMP_TABLE_SIZE];
+
 	for (int i = 0; i < RAMP_TABLE_SIZE; ++i) {
 		ramp_color[i][0] = ramp[i].x;
 		ramp_color[i][1] = ramp[i].y;
 		ramp_color[i][2] = ramp[i].z;
 		ramp_alpha[i] = ramp[i].w;
 	}
+
 	compiler.parameter_color_array("ramp_color", ramp_color, RAMP_TABLE_SIZE);
 	compiler.parameter_array("ramp_alpha", ramp_alpha, RAMP_TABLE_SIZE);
 	
 	compiler.add(this, "node_rgb_ramp");
 }
 
+/* Set Normal Node */
+
+SetNormalNode::SetNormalNode()
+: ShaderNode("set_normal")
+{
+	add_input("Direction", SHADER_SOCKET_VECTOR);
+	add_output("Normal", SHADER_SOCKET_NORMAL);
+}
+
+void SetNormalNode::compile(SVMCompiler& compiler)
+{
+	ShaderInput  *direction_in = input("Direction");
+	ShaderOutput *normal_out = output("Normal");
+
+	compiler.stack_assign(direction_in);
+	compiler.stack_assign(normal_out);
+
+	compiler.add_node(NODE_CLOSURE_SET_NORMAL, direction_in->stack_offset, normal_out->stack_offset);
+}
+
+void SetNormalNode::compile(OSLCompiler& compiler)
+{
+	compiler.add(this, "set_normal"); 
+}
+
 CCL_NAMESPACE_END
 
diff --git a/intern/cycles/render/nodes.h b/intern/cycles/render/nodes.h
index 6e5d7be0091..fbc61e12fd4 100644
--- a/intern/cycles/render/nodes.h
+++ b/intern/cycles/render/nodes.h
@@ -440,6 +440,11 @@ public:
 	float4 ramp[RAMP_TABLE_SIZE];
 };
 
+class SetNormalNode : public ShaderNode {
+public:
+	SHADER_NODE_CLASS(SetNormalNode)
+};
+
 CCL_NAMESPACE_END
 
 #endif /* __NODES_H__ */