Cycles: Remove ccl_fetch and SOA

45 files changed, 702 insertions, 724 deletions

diff --git a/intern/cycles/kernel/closure/alloc.h b/intern/cycles/kernel/closure/alloc.h
index b7abc1ec507..4894ea58dba 100644
--- a/intern/cycles/kernel/closure/alloc.h
+++ b/intern/cycles/kernel/closure/alloc.h
@@ -20,17 +20,17 @@ ccl_device ShaderClosure *closure_alloc(ShaderData *sd, int size, ClosureType ty
 {
 	kernel_assert(size <= sizeof(ShaderClosure));
 
-	int num_closure = ccl_fetch(sd, num_closure);
-	int num_closure_extra = ccl_fetch(sd, num_closure_extra);
+	int num_closure = sd->num_closure;
+	int num_closure_extra = sd->num_closure_extra;
 	if(num_closure + num_closure_extra >= MAX_CLOSURE)
 		return NULL;
 
-	ShaderClosure *sc = &ccl_fetch(sd, closure)[num_closure];
+	ShaderClosure *sc = &sd->closure[num_closure];
 
 	sc->type = type;
 	sc->weight = weight;
 
-	ccl_fetch(sd, num_closure)++;
+	sd->num_closure++;
 
 	return sc;
 }
@@ -44,18 +44,18 @@ ccl_device ccl_addr_space void *closure_alloc_extra(ShaderData *sd, int size)
 	 * This lets us keep the same fast array iteration over closures, as we
 	 * found linked list iteration and iteration with skipping to be slower. */
 	int num_extra = ((size + sizeof(ShaderClosure) - 1) / sizeof(ShaderClosure));
-	int num_closure = ccl_fetch(sd, num_closure);
-	int num_closure_extra = ccl_fetch(sd, num_closure_extra) + num_extra;
+	int num_closure = sd->num_closure;
+	int num_closure_extra = sd->num_closure_extra + num_extra;
 
 	if(num_closure + num_closure_extra > MAX_CLOSURE) {
 		/* Remove previous closure. */
-		ccl_fetch(sd, num_closure)--;
-		ccl_fetch(sd, num_closure_extra)++;
+		sd->num_closure--;
+		sd->num_closure_extra++;
 		return NULL;
 	}
 
-	ccl_fetch(sd, num_closure_extra) = num_closure_extra;
-	return (ccl_addr_space void*)(ccl_fetch(sd, closure) + MAX_CLOSURE - num_closure_extra);
+	sd->num_closure_extra = num_closure_extra;
+	return (ccl_addr_space void*)(sd->closure + MAX_CLOSURE - num_closure_extra);
 }
 
 ccl_device_inline ShaderClosure *bsdf_alloc(ShaderData *sd, int size, float3 weight)
diff --git a/intern/cycles/kernel/closure/bsdf.h b/intern/cycles/kernel/closure/bsdf.h
index 7e4d5fe2e37..a44b9e2d9b9 100644
--- a/intern/cycles/kernel/closure/bsdf.h
+++ b/intern/cycles/kernel/closure/bsdf.h
@@ -51,89 +51,89 @@ ccl_device_forceinline int bsdf_sample(KernelGlobals *kg,
 	switch(sc->type) {
 		case CLOSURE_BSDF_DIFFUSE_ID:
 		case CLOSURE_BSDF_BSSRDF_ID:
-			label = bsdf_diffuse_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_diffuse_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 #ifdef __SVM__
 		case CLOSURE_BSDF_OREN_NAYAR_ID:
-			label = bsdf_oren_nayar_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_oren_nayar_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 #ifdef __OSL__
 		case CLOSURE_BSDF_PHONG_RAMP_ID:
-			label = bsdf_phong_ramp_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_phong_ramp_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_DIFFUSE_RAMP_ID:
-			label = bsdf_diffuse_ramp_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_diffuse_ramp_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 #endif
 		case CLOSURE_BSDF_TRANSLUCENT_ID:
-			label = bsdf_translucent_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_translucent_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_REFLECTION_ID:
-			label = bsdf_reflection_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_reflection_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_REFRACTION_ID:
-			label = bsdf_refraction_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_refraction_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_TRANSPARENT_ID:
-			label = bsdf_transparent_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_transparent_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_MICROFACET_GGX_ID:
 		case CLOSURE_BSDF_MICROFACET_GGX_ANISO_ID:
 		case CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID:
-			label = bsdf_microfacet_ggx_sample(kg, sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_microfacet_ggx_sample(kg, sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_MICROFACET_MULTI_GGX_ID:
-			label = bsdf_microfacet_multi_ggx_sample(kg, sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
-			        eval, omega_in,  &domega_in->dx, &domega_in->dy, pdf, &ccl_fetch(sd, lcg_state));
+			label = bsdf_microfacet_multi_ggx_sample(kg, sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
+			        eval, omega_in,  &domega_in->dx, &domega_in->dy, pdf, &sd->lcg_state);
 			break;
 		case CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_ID:
-			label = bsdf_microfacet_multi_ggx_glass_sample(kg, sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
-			        eval, omega_in,  &domega_in->dx, &domega_in->dy, pdf, &ccl_fetch(sd, lcg_state));
+			label = bsdf_microfacet_multi_ggx_glass_sample(kg, sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
+			        eval, omega_in,  &domega_in->dx, &domega_in->dy, pdf, &sd->lcg_state);
 			break;
 		case CLOSURE_BSDF_MICROFACET_BECKMANN_ID:
 		case CLOSURE_BSDF_MICROFACET_BECKMANN_ANISO_ID:
 		case CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID:
-			label = bsdf_microfacet_beckmann_sample(kg, sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_microfacet_beckmann_sample(kg, sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ID:
 		case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ANISO_ID:
-			label = bsdf_ashikhmin_shirley_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_ashikhmin_shirley_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID:
-			label = bsdf_ashikhmin_velvet_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_ashikhmin_velvet_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_DIFFUSE_TOON_ID:
-			label = bsdf_diffuse_toon_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_diffuse_toon_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_GLOSSY_TOON_ID:
-			label = bsdf_glossy_toon_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_glossy_toon_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_HAIR_REFLECTION_ID:
-			label = bsdf_hair_reflection_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_hair_reflection_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 		case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
-			label = bsdf_hair_transmission_sample(sc, ccl_fetch(sd, Ng), ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv,
+			label = bsdf_hair_transmission_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
 				eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 #endif
 #ifdef __VOLUME__
 		case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
-			label = volume_henyey_greenstein_sample(sc, ccl_fetch(sd, I), ccl_fetch(sd, dI).dx, ccl_fetch(sd, dI).dy, randu, randv, eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
+			label = volume_henyey_greenstein_sample(sc, sd->I, sd->dI.dx, sd->dI.dy, randu, randv, eval, omega_in, &domega_in->dx, &domega_in->dy, pdf);
 			break;
 #endif
 		default:
@@ -157,75 +157,75 @@ float3 bsdf_eval(KernelGlobals *kg,
 {
 	float3 eval;
 
-	if(dot(ccl_fetch(sd, Ng), omega_in) >= 0.0f) {
+	if(dot(sd->Ng, omega_in) >= 0.0f) {
 		switch(sc->type) {
 			case CLOSURE_BSDF_DIFFUSE_ID:
 			case CLOSURE_BSDF_BSSRDF_ID:
-				eval = bsdf_diffuse_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_diffuse_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 #ifdef __SVM__
 			case CLOSURE_BSDF_OREN_NAYAR_ID:
-				eval = bsdf_oren_nayar_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_oren_nayar_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 #ifdef __OSL__
 			case CLOSURE_BSDF_PHONG_RAMP_ID:
-				eval = bsdf_phong_ramp_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_phong_ramp_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_DIFFUSE_RAMP_ID:
-				eval = bsdf_diffuse_ramp_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_diffuse_ramp_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 #endif
 			case CLOSURE_BSDF_TRANSLUCENT_ID:
-				eval = bsdf_translucent_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_translucent_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_REFLECTION_ID:
-				eval = bsdf_reflection_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_reflection_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_REFRACTION_ID:
-				eval = bsdf_refraction_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_refraction_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_TRANSPARENT_ID:
-				eval = bsdf_transparent_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_transparent_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_MICROFACET_GGX_ID:
 			case CLOSURE_BSDF_MICROFACET_GGX_ANISO_ID:
 			case CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID:
-				eval = bsdf_microfacet_ggx_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_microfacet_ggx_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_MICROFACET_MULTI_GGX_ID:
-				eval = bsdf_microfacet_multi_ggx_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf, &ccl_fetch(sd, lcg_state));
+				eval = bsdf_microfacet_multi_ggx_eval_reflect(sc, sd->I, omega_in, pdf, &sd->lcg_state);
 				break;
 			case CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_ID:
-				eval = bsdf_microfacet_multi_ggx_glass_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf, &ccl_fetch(sd, lcg_state));
+				eval = bsdf_microfacet_multi_ggx_glass_eval_reflect(sc, sd->I, omega_in, pdf, &sd->lcg_state);
 				break;
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_ID:
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_ANISO_ID:
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID:
-				eval = bsdf_microfacet_beckmann_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_microfacet_beckmann_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ID:
 			case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ANISO_ID:
-				eval = bsdf_ashikhmin_shirley_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_ashikhmin_shirley_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID:
-				eval = bsdf_ashikhmin_velvet_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_ashikhmin_velvet_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_DIFFUSE_TOON_ID:
-				eval = bsdf_diffuse_toon_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_diffuse_toon_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_GLOSSY_TOON_ID:
-				eval = bsdf_glossy_toon_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_glossy_toon_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_HAIR_REFLECTION_ID:
-				eval = bsdf_hair_reflection_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_hair_reflection_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
-				eval = bsdf_hair_transmission_eval_reflect(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_hair_transmission_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
 #endif
 #ifdef __VOLUME__
 			case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
-				eval = volume_henyey_greenstein_eval_phase(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = volume_henyey_greenstein_eval_phase(sc, sd->I, omega_in, pdf);
 				break;
 #endif
 			default:
@@ -237,63 +237,63 @@ float3 bsdf_eval(KernelGlobals *kg,
 		switch(sc->type) {
 			case CLOSURE_BSDF_DIFFUSE_ID:
 			case CLOSURE_BSDF_BSSRDF_ID:
-				eval = bsdf_diffuse_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_diffuse_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 #ifdef __SVM__
 			case CLOSURE_BSDF_OREN_NAYAR_ID:
-				eval = bsdf_oren_nayar_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_oren_nayar_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_TRANSLUCENT_ID:
-				eval = bsdf_translucent_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_translucent_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_REFLECTION_ID:
-				eval = bsdf_reflection_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_reflection_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_REFRACTION_ID:
-				eval = bsdf_refraction_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_refraction_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_TRANSPARENT_ID:
-				eval = bsdf_transparent_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_transparent_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_MICROFACET_GGX_ID:
 			case CLOSURE_BSDF_MICROFACET_GGX_ANISO_ID:
 			case CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID:
-				eval = bsdf_microfacet_ggx_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_microfacet_ggx_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_MICROFACET_MULTI_GGX_ID:
-				eval = bsdf_microfacet_multi_ggx_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf, &ccl_fetch(sd, lcg_state));
+				eval = bsdf_microfacet_multi_ggx_eval_transmit(sc, sd->I, omega_in, pdf, &sd->lcg_state);
 				break;
 			case CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_ID:
-				eval = bsdf_microfacet_multi_ggx_glass_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf, &ccl_fetch(sd, lcg_state));
+				eval = bsdf_microfacet_multi_ggx_glass_eval_transmit(sc, sd->I, omega_in, pdf, &sd->lcg_state);
 				break;
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_ID:
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_ANISO_ID:
 			case CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID:
-				eval = bsdf_microfacet_beckmann_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_microfacet_beckmann_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ID:
 			case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ANISO_ID:
-				eval = bsdf_ashikhmin_shirley_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_ashikhmin_shirley_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID:
-				eval = bsdf_ashikhmin_velvet_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_ashikhmin_velvet_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_DIFFUSE_TOON_ID:
-				eval = bsdf_diffuse_toon_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_diffuse_toon_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_GLOSSY_TOON_ID:
-				eval = bsdf_glossy_toon_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_glossy_toon_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_HAIR_REFLECTION_ID:
-				eval = bsdf_hair_reflection_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_hair_reflection_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 			case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
-				eval = bsdf_hair_transmission_eval_transmit(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = bsdf_hair_transmission_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
 #endif
 #ifdef __VOLUME__
 			case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
-				eval = volume_henyey_greenstein_eval_phase(sc, ccl_fetch(sd, I), omega_in, pdf);
+				eval = volume_henyey_greenstein_eval_phase(sc, sd->I, omega_in, pdf);
 				break;
 #endif
 			default:
diff --git a/intern/cycles/kernel/geom/geom_attribute.h b/intern/cycles/kernel/geom/geom_attribute.h
index 08ccee56335..cc62192ef21 100644
--- a/intern/cycles/kernel/geom/geom_attribute.h
+++ b/intern/cycles/kernel/geom/geom_attribute.h
@@ -30,7 +30,7 @@ ccl_device_inline uint subd_triangle_patch(KernelGlobals *kg, const ShaderData *
 ccl_device_inline uint attribute_primitive_type(KernelGlobals *kg, const ShaderData *sd)
 {
 #ifdef __HAIR__
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+	if(sd->type & PRIMITIVE_ALL_CURVE) {
 		return ATTR_PRIM_CURVE;
 	}
 	else
@@ -53,12 +53,12 @@ ccl_device_inline AttributeDescriptor attribute_not_found()
 
 ccl_device_inline AttributeDescriptor find_attribute(KernelGlobals *kg, const ShaderData *sd, uint id)
 {
-	if(ccl_fetch(sd, object) == PRIM_NONE) {
+	if(sd->object == PRIM_NONE) {
 		return attribute_not_found();
 	}
 
 	/* for SVM, find attribute by unique id */
-	uint attr_offset = ccl_fetch(sd, object)*kernel_data.bvh.attributes_map_stride;
+	uint attr_offset = sd->object*kernel_data.bvh.attributes_map_stride;
 	attr_offset += attribute_primitive_type(kg, sd);
 	uint4 attr_map = kernel_tex_fetch(__attributes_map, attr_offset);
 	
@@ -73,7 +73,7 @@ ccl_device_inline AttributeDescriptor find_attribute(KernelGlobals *kg, const Sh
 	AttributeDescriptor desc;
 	desc.element = (AttributeElement)attr_map.y;
 	
-	if(ccl_fetch(sd, prim) == PRIM_NONE &&
+	if(sd->prim == PRIM_NONE &&
 	   desc.element != ATTR_ELEMENT_MESH &&
 	   desc.element != ATTR_ELEMENT_VOXEL &&
 	   desc.element != ATTR_ELEMENT_OBJECT)
diff --git a/intern/cycles/kernel/geom/geom_curve.h b/intern/cycles/kernel/geom/geom_curve.h
index 712b67a1b55..7cc840ce78d 100644
--- a/intern/cycles/kernel/geom/geom_curve.h
+++ b/intern/cycles/kernel/geom/geom_curve.h
@@ -32,22 +32,22 @@ ccl_device float curve_attribute_float(KernelGlobals *kg, const ShaderData *sd,
 		if(dy) *dy = 0.0f;
 #endif
 
-		return kernel_tex_fetch(__attributes_float, desc.offset + ccl_fetch(sd, prim));
+		return kernel_tex_fetch(__attributes_float, desc.offset + sd->prim);
 	}
 	else if(desc.element == ATTR_ELEMENT_CURVE_KEY || desc.element == ATTR_ELEMENT_CURVE_KEY_MOTION) {
-		float4 curvedata = kernel_tex_fetch(__curves, ccl_fetch(sd, prim));
-		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(ccl_fetch(sd, type));
+		float4 curvedata = kernel_tex_fetch(__curves, sd->prim);
+		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(sd->type);
 		int k1 = k0 + 1;
 
 		float f0 = kernel_tex_fetch(__attributes_float, desc.offset + k0);
 		float f1 = kernel_tex_fetch(__attributes_float, desc.offset + k1);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*(f1 - f0);
+		if(dx) *dx = sd->du.dx*(f1 - f0);
 		if(dy) *dy = 0.0f;
 #endif
 
-		return (1.0f - ccl_fetch(sd, u))*f0 + ccl_fetch(sd, u)*f1;
+		return (1.0f - sd->u)*f0 + sd->u*f1;
 	}
 	else {
 #ifdef __RAY_DIFFERENTIALS__
@@ -71,22 +71,22 @@ ccl_device float3 curve_attribute_float3(KernelGlobals *kg, const ShaderData *sd
 		if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
 #endif
 
-		return float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + ccl_fetch(sd, prim)));
+		return float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + sd->prim));
 	}
 	else if(desc.element == ATTR_ELEMENT_CURVE_KEY || desc.element == ATTR_ELEMENT_CURVE_KEY_MOTION) {
-		float4 curvedata = kernel_tex_fetch(__curves, ccl_fetch(sd, prim));
-		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(ccl_fetch(sd, type));
+		float4 curvedata = kernel_tex_fetch(__curves, sd->prim);
+		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(sd->type);
 		int k1 = k0 + 1;
 
 		float3 f0 = float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + k0));
 		float3 f1 = float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + k1));
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*(f1 - f0);
+		if(dx) *dx = sd->du.dx*(f1 - f0);
 		if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
 #endif
 
-		return (1.0f - ccl_fetch(sd, u))*f0 + ccl_fetch(sd, u)*f1;
+		return (1.0f - sd->u)*f0 + sd->u*f1;
 	}
 	else {
 #ifdef __RAY_DIFFERENTIALS__
@@ -104,22 +104,22 @@ ccl_device float curve_thickness(KernelGlobals *kg, ShaderData *sd)
 {
 	float r = 0.0f;
 
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
-		float4 curvedata = kernel_tex_fetch(__curves, ccl_fetch(sd, prim));
-		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(ccl_fetch(sd, type));
+	if(sd->type & PRIMITIVE_ALL_CURVE) {
+		float4 curvedata = kernel_tex_fetch(__curves, sd->prim);
+		int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(sd->type);
 		int k1 = k0 + 1;
 
 		float4 P_curve[2];
 
-		if(ccl_fetch(sd, type) & PRIMITIVE_CURVE) {
+		if(sd->type & PRIMITIVE_CURVE) {
 			P_curve[0]= kernel_tex_fetch(__curve_keys, k0);
 			P_curve[1]= kernel_tex_fetch(__curve_keys, k1);
 		}
 		else {
-			motion_curve_keys(kg, ccl_fetch(sd, object), ccl_fetch(sd, prim), ccl_fetch(sd, time), k0, k1, P_curve);
+			motion_curve_keys(kg, sd->object, sd->prim, sd->time, k0, k1, P_curve);
 		}
 
-		r = (P_curve[1].w - P_curve[0].w) * ccl_fetch(sd, u) + P_curve[0].w;
+		r = (P_curve[1].w - P_curve[0].w) * sd->u + P_curve[0].w;
 	}
 
 	return r*2.0f;
@@ -130,8 +130,8 @@ ccl_device float curve_thickness(KernelGlobals *kg, ShaderData *sd)
 
 ccl_device float3 curve_motion_center_location(KernelGlobals *kg, ShaderData *sd)
 {
-	float4 curvedata = kernel_tex_fetch(__curves, ccl_fetch(sd, prim));
-	int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(ccl_fetch(sd, type));
+	float4 curvedata = kernel_tex_fetch(__curves, sd->prim);
+	int k0 = __float_as_int(curvedata.x) + PRIMITIVE_UNPACK_SEGMENT(sd->type);
 	int k1 = k0 + 1;
 
 	float4 P_curve[2];
@@ -139,7 +139,7 @@ ccl_device float3 curve_motion_center_location(KernelGlobals *kg, ShaderData *sd
 	P_curve[0]= kernel_tex_fetch(__curve_keys, k0);
 	P_curve[1]= kernel_tex_fetch(__curve_keys, k1);
 
-	return float4_to_float3(P_curve[1]) * ccl_fetch(sd, u) + float4_to_float3(P_curve[0]) * (1.0f - ccl_fetch(sd, u));
+	return float4_to_float3(P_curve[1]) * sd->u + float4_to_float3(P_curve[0]) * (1.0f - sd->u);
 }
 
 /* Curve tangent normal */
@@ -148,14 +148,14 @@ ccl_device float3 curve_tangent_normal(KernelGlobals *kg, ShaderData *sd)
 {	
 	float3 tgN = make_float3(0.0f,0.0f,0.0f);
 
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+	if(sd->type & PRIMITIVE_ALL_CURVE) {
 
-		tgN = -(-ccl_fetch(sd, I) - ccl_fetch(sd, dPdu) * (dot(ccl_fetch(sd, dPdu),-ccl_fetch(sd, I)) / len_squared(ccl_fetch(sd, dPdu))));
+		tgN = -(-sd->I - sd->dPdu * (dot(sd->dPdu,-sd->I) / len_squared(sd->dPdu)));
 		tgN = normalize(tgN);
 
 		/* need to find suitable scaled gd for corrected normal */
 #if 0
-		tgN = normalize(tgN - gd * ccl_fetch(sd, dPdu));
+		tgN = normalize(tgN - gd * sd->dPdu);
 #endif
 	}
 
@@ -966,7 +966,7 @@ ccl_device_inline float3 bvh_curve_refine(KernelGlobals *kg, ShaderData *sd, con
 
 	if(isect->object != OBJECT_NONE) {
 #ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_itfm);
+		Transform tfm = sd->ob_itfm;
 #else
 		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
 #endif
@@ -979,7 +979,7 @@ ccl_device_inline float3 bvh_curve_refine(KernelGlobals *kg, ShaderData *sd, con
 	int prim = kernel_tex_fetch(__prim_index, isect->prim);
 	float4 v00 = kernel_tex_fetch(__curves, prim);
 
-	int k0 = __float_as_int(v00.x) + PRIMITIVE_UNPACK_SEGMENT(ccl_fetch(sd, type));
+	int k0 = __float_as_int(v00.x) + PRIMITIVE_UNPACK_SEGMENT(sd->type);
 	int k1 = k0 + 1;
 
 	float3 tg;
@@ -990,14 +990,14 @@ ccl_device_inline float3 bvh_curve_refine(KernelGlobals *kg, ShaderData *sd, con
 
 		float4 P_curve[4];
 
-		if(ccl_fetch(sd, type) & PRIMITIVE_CURVE) {
+		if(sd->type & PRIMITIVE_CURVE) {
 			P_curve[0] = kernel_tex_fetch(__curve_keys, ka);
 			P_curve[1] = kernel_tex_fetch(__curve_keys, k0);
 			P_curve[2] = kernel_tex_fetch(__curve_keys, k1);
 			P_curve[3] = kernel_tex_fetch(__curve_keys, kb);
 		}
 		else {
-			motion_cardinal_curve_keys(kg, ccl_fetch(sd, object), ccl_fetch(sd, prim), ccl_fetch(sd, time), ka, k0, k1, kb, P_curve);
+			motion_cardinal_curve_keys(kg, sd->object, sd->prim, sd->time, ka, k0, k1, kb, P_curve);
 		}
 
 		float3 p[4];
@@ -1009,43 +1009,43 @@ ccl_device_inline float3 bvh_curve_refine(KernelGlobals *kg, ShaderData *sd, con
 		P = P + D*t;
 
 #ifdef __UV__
-		ccl_fetch(sd, u) = isect->u;
-		ccl_fetch(sd, v) = 0.0f;
+		sd->u = isect->u;
+		sd->v = 0.0f;
 #endif
 
 		tg = normalize(curvetangent(isect->u, p[0], p[1], p[2], p[3]));
 
 		if(kernel_data.curve.curveflags & CURVE_KN_RIBBONS) {
-			ccl_fetch(sd, Ng) = normalize(-(D - tg * (dot(tg, D))));
+			sd->Ng = normalize(-(D - tg * (dot(tg, D))));
 		}
 		else {
 			/* direction from inside to surface of curve */
 			float3 p_curr = curvepoint(isect->u, p[0], p[1], p[2], p[3]);	
-			ccl_fetch(sd, Ng) = normalize(P - p_curr);
+			sd->Ng = normalize(P - p_curr);
 
 			/* adjustment for changing radius */
 			float gd = isect->v;
 
 			if(gd != 0.0f) {
-				ccl_fetch(sd, Ng) = ccl_fetch(sd, Ng) - gd * tg;
-				ccl_fetch(sd, Ng) = normalize(ccl_fetch(sd, Ng));
+				sd->Ng = sd->Ng - gd * tg;
+				sd->Ng = normalize(sd->Ng);
 			}
 		}
 
 		/* todo: sometimes the normal is still so that this is detected as
 		 * backfacing even if cull backfaces is enabled */
 
-		ccl_fetch(sd, N) = ccl_fetch(sd, Ng);
+		sd->N = sd->Ng;
 	}
 	else {
 		float4 P_curve[2];
 
-		if(ccl_fetch(sd, type) & PRIMITIVE_CURVE) {
+		if(sd->type & PRIMITIVE_CURVE) {
 			P_curve[0]= kernel_tex_fetch(__curve_keys, k0);
 			P_curve[1]= kernel_tex_fetch(__curve_keys, k1);
 		}
 		else {
-			motion_curve_keys(kg, ccl_fetch(sd, object), ccl_fetch(sd, prim), ccl_fetch(sd, time), k0, k1, P_curve);
+			motion_curve_keys(kg, sd->object, sd->prim, sd->time, k0, k1, P_curve);
 		}
 
 		float l = 1.0f;
@@ -1056,39 +1056,39 @@ ccl_device_inline float3 bvh_curve_refine(KernelGlobals *kg, ShaderData *sd, con
 		float3 dif = P - float4_to_float3(P_curve[0]);
 
 #ifdef __UV__
-		ccl_fetch(sd, u) = dot(dif,tg)/l;
-		ccl_fetch(sd, v) = 0.0f;
+		sd->u = dot(dif,tg)/l;
+		sd->v = 0.0f;
 #endif
 
 		if(flag & CURVE_KN_TRUETANGENTGNORMAL) {
-			ccl_fetch(sd, Ng) = -(D - tg * dot(tg, D));
-			ccl_fetch(sd, Ng) = normalize(ccl_fetch(sd, Ng));
+			sd->Ng = -(D - tg * dot(tg, D));
+			sd->Ng = normalize(sd->Ng);
 		}
 		else {
 			float gd = isect->v;
 
 			/* direction from inside to surface of curve */
-			ccl_fetch(sd, Ng) = (dif - tg * ccl_fetch(sd, u) * l) / (P_curve[0].w + ccl_fetch(sd, u) * l * gd);
+			sd->Ng = (dif - tg * sd->u * l) / (P_curve[0].w + sd->u * l * gd);
 
 			/* adjustment for changing radius */
 			if(gd != 0.0f) {
-				ccl_fetch(sd, Ng) = ccl_fetch(sd, Ng) - gd * tg;
-				ccl_fetch(sd, Ng) = normalize(ccl_fetch(sd, Ng));
+				sd->Ng = sd->Ng - gd * tg;
+				sd->Ng = normalize(sd->Ng);
 			}
 		}
 
-		ccl_fetch(sd, N) = ccl_fetch(sd, Ng);
+		sd->N = sd->Ng;
 	}
 
 #ifdef __DPDU__
 	/* dPdu/dPdv */
-	ccl_fetch(sd, dPdu) = tg;
-	ccl_fetch(sd, dPdv) = cross(tg, ccl_fetch(sd, Ng));
+	sd->dPdu = tg;
+	sd->dPdv = cross(tg, sd->Ng);
 #endif
 
 	if(isect->object != OBJECT_NONE) {
 #ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_tfm);
+		Transform tfm = sd->ob_tfm;
 #else
 		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
 #endif
diff --git a/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h b/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h
index d57d74ea882..2500228281e 100644
--- a/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h
+++ b/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h
@@ -48,7 +48,7 @@ ccl_device_inline float3 motion_triangle_refine(KernelGlobals *kg,
 			return P;
 		}
 #  ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_itfm);
+		Transform tfm = sd->ob_itfm;
 #  else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
@@ -77,7 +77,7 @@ ccl_device_inline float3 motion_triangle_refine(KernelGlobals *kg,
 
 	if(isect->object != OBJECT_NONE) {
 #  ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_tfm);
+		Transform tfm = sd->ob_tfm;
 #  else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
@@ -116,7 +116,7 @@ float3 motion_triangle_refine_subsurface(KernelGlobals *kg,
 #  ifdef __INTERSECTION_REFINE__
 	if(isect->object != OBJECT_NONE) {
 #    ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_itfm);
+		Transform tfm = sd->ob_itfm;
 #    else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
@@ -144,7 +144,7 @@ float3 motion_triangle_refine_subsurface(KernelGlobals *kg,
 
 	if(isect->object != OBJECT_NONE) {
 #    ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_tfm);
+		Transform tfm = sd->ob_tfm;
 #    else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
diff --git a/intern/cycles/kernel/geom/geom_motion_triangle_shader.h b/intern/cycles/kernel/geom/geom_motion_triangle_shader.h
index 0e024a05db6..cb456056e20 100644
--- a/intern/cycles/kernel/geom/geom_motion_triangle_shader.h
+++ b/intern/cycles/kernel/geom/geom_motion_triangle_shader.h
@@ -39,26 +39,26 @@ ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg,
                                                       bool subsurface)
 {
 	/* Get shader. */
-	ccl_fetch(sd, shader) = kernel_tex_fetch(__tri_shader, ccl_fetch(sd, prim));
+	sd->shader = kernel_tex_fetch(__tri_shader, sd->prim);
 	/* Get motion info. */
 	/* TODO(sergey): This logic is really similar to motion_triangle_vertices(),
 	 * can we de-duplicate something here?
 	 */
 	int numsteps, numverts;
-	object_motion_info(kg, ccl_fetch(sd, object), &numsteps, &numverts, NULL);
+	object_motion_info(kg, sd->object, &numsteps, &numverts, NULL);
 	/* Figure out which steps we need to fetch and their interpolation factor. */
 	int maxstep = numsteps*2;
-	int step = min((int)(ccl_fetch(sd, time)*maxstep), maxstep-1);
-	float t = ccl_fetch(sd, time)*maxstep - step;
+	int step = min((int)(sd->time*maxstep), maxstep-1);
+	float t = sd->time*maxstep - step;
 	/* Find attribute. */
 	AttributeElement elem;
-	int offset = find_attribute_motion(kg, ccl_fetch(sd, object),
+	int offset = find_attribute_motion(kg, sd->object,
 	                                   ATTR_STD_MOTION_VERTEX_POSITION,
 	                                   &elem);
 	kernel_assert(offset != ATTR_STD_NOT_FOUND);
 	/* Fetch vertex coordinates. */
 	float3 verts[3], next_verts[3];
-	uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+	uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, sd->prim);
 	motion_triangle_verts_for_step(kg, tri_vindex, offset, numverts, numsteps, step, verts);
 	motion_triangle_verts_for_step(kg, tri_vindex, offset, numverts, numsteps, step+1, next_verts);
 	/* Interpolate between steps. */
@@ -68,7 +68,7 @@ ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg,
 	/* Compute refined position. */
 #ifdef __SUBSURFACE__
 	if(subsurface) {
-		ccl_fetch(sd, P) = motion_triangle_refine_subsurface(kg,
+		sd->P = motion_triangle_refine_subsurface(kg,
 		                                                     sd,
 		                                                     isect,
 		                                                     ray,
@@ -77,29 +77,29 @@ ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg,
 	else
 #endif  /*  __SUBSURFACE__*/
 	{
-		ccl_fetch(sd, P) = motion_triangle_refine(kg, sd, isect, ray, verts);
+		sd->P = motion_triangle_refine(kg, sd, isect, ray, verts);
 	}
 	/* Compute face normal. */
 	float3 Ng;
-	if(ccl_fetch(sd, object_flag) & SD_OBJECT_NEGATIVE_SCALE_APPLIED) {
+	if(sd->object_flag & SD_OBJECT_NEGATIVE_SCALE_APPLIED) {
 		Ng = normalize(cross(verts[2] - verts[0], verts[1] - verts[0]));
 	}
 	else {
 		Ng = normalize(cross(verts[1] - verts[0], verts[2] - verts[0]));
 	}
-	ccl_fetch(sd, Ng) = Ng;
-	ccl_fetch(sd, N) = Ng;
+	sd->Ng = Ng;
+	sd->N = Ng;
 	/* Compute derivatives of P w.r.t. uv. */
 #ifdef __DPDU__
-	ccl_fetch(sd, dPdu) = (verts[0] - verts[2]);
-	ccl_fetch(sd, dPdv) = (verts[1] - verts[2]);
+	sd->dPdu = (verts[0] - verts[2]);
+	sd->dPdv = (verts[1] - verts[2]);
 #endif
 	/* Compute smooth normal. */
-	if(ccl_fetch(sd, shader) & SHADER_SMOOTH_NORMAL) {
+	if(sd->shader & SHADER_SMOOTH_NORMAL) {
 		/* Find attribute. */
 		AttributeElement elem;
 		int offset = find_attribute_motion(kg,
-		                                   ccl_fetch(sd, object),
+		                                   sd->object,
 		                                   ATTR_STD_MOTION_VERTEX_NORMAL,
 		                                   &elem);
 		kernel_assert(offset != ATTR_STD_NOT_FOUND);
@@ -112,10 +112,10 @@ ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg,
 		normals[1] = (1.0f - t)*normals[1] + t*next_normals[1];
 		normals[2] = (1.0f - t)*normals[2] + t*next_normals[2];
 		/* Interpolate between vertices. */
-		float u = ccl_fetch(sd, u);
-		float v = ccl_fetch(sd, v);
+		float u = sd->u;
+		float v = sd->v;
 		float w = 1.0f - u - v;
-		ccl_fetch(sd, N) = (u*normals[0] + v*normals[1] + w*normals[2]);
+		sd->N = (u*normals[0] + v*normals[1] + w*normals[2]);
 	}
 }
 
diff --git a/intern/cycles/kernel/geom/geom_object.h b/intern/cycles/kernel/geom/geom_object.h
index f51b2d18657..5a04be8b0bf 100644
--- a/intern/cycles/kernel/geom/geom_object.h
+++ b/intern/cycles/kernel/geom/geom_object.h
@@ -137,9 +137,9 @@ ccl_device_inline Transform object_fetch_transform_motion_test(KernelGlobals *kg
 ccl_device_inline void object_position_transform(KernelGlobals *kg, const ShaderData *sd, float3 *P)
 {
 #ifdef __OBJECT_MOTION__
-	*P = transform_point_auto(&ccl_fetch(sd, ob_tfm), *P);
+	*P = transform_point_auto(&sd->ob_tfm, *P);
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	*P = transform_point(&tfm, *P);
 #endif
 }
@@ -149,9 +149,9 @@ ccl_device_inline void object_position_transform(KernelGlobals *kg, const Shader
 ccl_device_inline void object_inverse_position_transform(KernelGlobals *kg, const ShaderData *sd, float3 *P)
 {
 #ifdef __OBJECT_MOTION__
-	*P = transform_point_auto(&ccl_fetch(sd, ob_itfm), *P);
+	*P = transform_point_auto(&sd->ob_itfm, *P);
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	*P = transform_point(&tfm, *P);
 #endif
 }
@@ -161,12 +161,12 @@ ccl_device_inline void object_inverse_position_transform(KernelGlobals *kg, cons
 ccl_device_inline void object_inverse_normal_transform(KernelGlobals *kg, const ShaderData *sd, float3 *N)
 {
 #ifdef __OBJECT_MOTION__
-	if((ccl_fetch(sd, object) != OBJECT_NONE) || (ccl_fetch(sd, type) == PRIMITIVE_LAMP)) {
-		*N = normalize(transform_direction_transposed_auto(&ccl_fetch(sd, ob_tfm), *N));
+	if((sd->object != OBJECT_NONE) || (sd->type == PRIMITIVE_LAMP)) {
+		*N = normalize(transform_direction_transposed_auto(&sd->ob_tfm, *N));
 	}
 #else
-	if(ccl_fetch(sd, object) != OBJECT_NONE) {
-		Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_TRANSFORM);
+	if(sd->object != OBJECT_NONE) {
+		Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 		*N = normalize(transform_direction_transposed(&tfm, *N));
 	}
 #endif
@@ -177,9 +177,9 @@ ccl_device_inline void object_inverse_normal_transform(KernelGlobals *kg, const
 ccl_device_inline void object_normal_transform(KernelGlobals *kg, const ShaderData *sd, float3 *N)
 {
 #ifdef __OBJECT_MOTION__
-	*N = normalize(transform_direction_transposed_auto(&ccl_fetch(sd, ob_itfm), *N));
+	*N = normalize(transform_direction_transposed_auto(&sd->ob_itfm, *N));
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	*N = normalize(transform_direction_transposed(&tfm, *N));
 #endif
 }
@@ -189,9 +189,9 @@ ccl_device_inline void object_normal_transform(KernelGlobals *kg, const ShaderDa
 ccl_device_inline void object_dir_transform(KernelGlobals *kg, const ShaderData *sd, float3 *D)
 {
 #ifdef __OBJECT_MOTION__
-	*D = transform_direction_auto(&ccl_fetch(sd, ob_tfm), *D);
+	*D = transform_direction_auto(&sd->ob_tfm, *D);
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	*D = transform_direction(&tfm, *D);
 #endif
 }
@@ -201,9 +201,9 @@ ccl_device_inline void object_dir_transform(KernelGlobals *kg, const ShaderData
 ccl_device_inline void object_inverse_dir_transform(KernelGlobals *kg, const ShaderData *sd, float3 *D)
 {
 #ifdef __OBJECT_MOTION__
-	*D = transform_direction_auto(&ccl_fetch(sd, ob_itfm), *D);
+	*D = transform_direction_auto(&sd->ob_itfm, *D);
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_INVERSE_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	*D = transform_direction(&tfm, *D);
 #endif
 }
@@ -212,13 +212,13 @@ ccl_device_inline void object_inverse_dir_transform(KernelGlobals *kg, const Sha
 
 ccl_device_inline float3 object_location(KernelGlobals *kg, const ShaderData *sd)
 {
-	if(ccl_fetch(sd, object) == OBJECT_NONE)
+	if(sd->object == OBJECT_NONE)
 		return make_float3(0.0f, 0.0f, 0.0f);
 
 #ifdef __OBJECT_MOTION__
-	return make_float3(ccl_fetch(sd, ob_tfm).x.w, ccl_fetch(sd, ob_tfm).y.w, ccl_fetch(sd, ob_tfm).z.w);
+	return make_float3(sd->ob_tfm.x.w, sd->ob_tfm.y.w, sd->ob_tfm.z.w);
 #else
-	Transform tfm = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_TRANSFORM);
+	Transform tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
 	return make_float3(tfm.x.w, tfm.y.w, tfm.z.w);
 #endif
 }
@@ -326,7 +326,7 @@ ccl_device_inline uint object_patch_map_offset(KernelGlobals *kg, int object)
 
 ccl_device int shader_pass_id(KernelGlobals *kg, const ShaderData *sd)
 {
-	return kernel_tex_fetch(__shader_flag, (ccl_fetch(sd, shader) & SHADER_MASK)*SHADER_SIZE + 1);
+	return kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE + 1);
 }
 
 /* Particle data from which object was instanced */
diff --git a/intern/cycles/kernel/geom/geom_patch.h b/intern/cycles/kernel/geom/geom_patch.h
index 6a0ff5a4a04..5663b598508 100644
--- a/intern/cycles/kernel/geom/geom_patch.h
+++ b/intern/cycles/kernel/geom/geom_patch.h
@@ -267,7 +267,7 @@ ccl_device float patch_eval_float(KernelGlobals *kg, const ShaderData *sd, int o
 	float weights_du[PATCH_MAX_CONTROL_VERTS];
 	float weights_dv[PATCH_MAX_CONTROL_VERTS];
 
-	int num_control = patch_eval_control_verts(kg, ccl_fetch(sd, object), patch, u, v, channel,
+	int num_control = patch_eval_control_verts(kg, sd->object, patch, u, v, channel,
 	                                           indices, weights, weights_du, weights_dv);
 
 	float val = 0.0f;
@@ -294,7 +294,7 @@ ccl_device float3 patch_eval_float3(KernelGlobals *kg, const ShaderData *sd, int
 	float weights_du[PATCH_MAX_CONTROL_VERTS];
 	float weights_dv[PATCH_MAX_CONTROL_VERTS];
 
-	int num_control = patch_eval_control_verts(kg, ccl_fetch(sd, object), patch, u, v, channel,
+	int num_control = patch_eval_control_verts(kg, sd->object, patch, u, v, channel,
 	                                           indices, weights, weights_du, weights_dv);
 
 	float3 val = make_float3(0.0f, 0.0f, 0.0f);
@@ -321,7 +321,7 @@ ccl_device float3 patch_eval_uchar4(KernelGlobals *kg, const ShaderData *sd, int
 	float weights_du[PATCH_MAX_CONTROL_VERTS];
 	float weights_dv[PATCH_MAX_CONTROL_VERTS];
 
-	int num_control = patch_eval_control_verts(kg, ccl_fetch(sd, object), patch, u, v, channel,
+	int num_control = patch_eval_control_verts(kg, sd->object, patch, u, v, channel,
 	                                           indices, weights, weights_du, weights_dv);
 
 	float3 val = make_float3(0.0f, 0.0f, 0.0f);
diff --git a/intern/cycles/kernel/geom/geom_primitive.h b/intern/cycles/kernel/geom/geom_primitive.h
index 8a73bb2f78b..90a9c2147cc 100644
--- a/intern/cycles/kernel/geom/geom_primitive.h
+++ b/intern/cycles/kernel/geom/geom_primitive.h
@@ -28,19 +28,19 @@ ccl_device_inline float primitive_attribute_float(KernelGlobals *kg,
                                                   const AttributeDescriptor desc,
                                                   float *dx, float *dy)
 {
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE) {
+	if(sd->type & PRIMITIVE_ALL_TRIANGLE) {
 		if(subd_triangle_patch(kg, sd) == ~0)
 			return triangle_attribute_float(kg, sd, desc, dx, dy);
 		else
 			return subd_triangle_attribute_float(kg, sd, desc, dx, dy);
 	}
 #ifdef __HAIR__
-	else if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+	else if(sd->type & PRIMITIVE_ALL_CURVE) {
 		return curve_attribute_float(kg, sd, desc, dx, dy);
 	}
 #endif
 #ifdef __VOLUME__
-	else if(ccl_fetch(sd, object) != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
+	else if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
 		return volume_attribute_float(kg, sd, desc, dx, dy);
 	}
 #endif
@@ -56,19 +56,19 @@ ccl_device_inline float3 primitive_attribute_float3(KernelGlobals *kg,
                                                     const AttributeDescriptor desc,
                                                     float3 *dx, float3 *dy)
 {
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE) {
+	if(sd->type & PRIMITIVE_ALL_TRIANGLE) {
 		if(subd_triangle_patch(kg, sd) == ~0)
 			return triangle_attribute_float3(kg, sd, desc, dx, dy);
 		else
 			return subd_triangle_attribute_float3(kg, sd, desc, dx, dy);
 	}
 #ifdef __HAIR__
-	else if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+	else if(sd->type & PRIMITIVE_ALL_CURVE) {
 		return curve_attribute_float3(kg, sd, desc, dx, dy);
 	}
 #endif
 #ifdef __VOLUME__
-	else if(ccl_fetch(sd, object) != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
+	else if(sd->object != OBJECT_NONE && desc.element == ATTR_ELEMENT_VOXEL) {
 		return volume_attribute_float3(kg, sd, desc, dx, dy);
 	}
 #endif
@@ -118,9 +118,9 @@ ccl_device bool primitive_ptex(KernelGlobals *kg, ShaderData *sd, float2 *uv, in
 ccl_device float3 primitive_tangent(KernelGlobals *kg, ShaderData *sd)
 {
 #ifdef __HAIR__
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE)
+	if(sd->type & PRIMITIVE_ALL_CURVE)
 #  ifdef __DPDU__
-		return normalize(ccl_fetch(sd, dPdu));
+		return normalize(sd->dPdu);
 #  else
 		return make_float3(0.0f, 0.0f, 0.0f);
 #  endif
@@ -133,12 +133,12 @@ ccl_device float3 primitive_tangent(KernelGlobals *kg, ShaderData *sd)
 		float3 data = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
 		data = make_float3(-(data.y - 0.5f), (data.x - 0.5f), 0.0f);
 		object_normal_transform(kg, sd, &data);
-		return cross(ccl_fetch(sd, N), normalize(cross(data, ccl_fetch(sd, N))));
+		return cross(sd->N, normalize(cross(data, sd->N)));
 	}
 	else {
 		/* otherwise use surface derivatives */
 #ifdef __DPDU__
-		return normalize(ccl_fetch(sd, dPdu));
+		return normalize(sd->dPdu);
 #else
 		return make_float3(0.0f, 0.0f, 0.0f);
 #endif
@@ -153,17 +153,17 @@ ccl_device_inline float4 primitive_motion_vector(KernelGlobals *kg, ShaderData *
 	float3 center;
 
 #ifdef __HAIR__
-	bool is_curve_primitive = ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE;
+	bool is_curve_primitive = sd->type & PRIMITIVE_ALL_CURVE;
 	if(is_curve_primitive) {
 		center = curve_motion_center_location(kg, sd);
 
-		if(!(ccl_fetch(sd, object_flag) & SD_OBJECT_TRANSFORM_APPLIED)) {
+		if(!(sd->object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
 			object_position_transform(kg, sd, &center);
 		}
 	}
 	else
 #endif
-		center = ccl_fetch(sd, P);
+		center = sd->P;
 
 	float3 motion_pre = center, motion_post = center;
 
@@ -173,16 +173,16 @@ ccl_device_inline float4 primitive_motion_vector(KernelGlobals *kg, ShaderData *
 	if(desc.offset != ATTR_STD_NOT_FOUND) {
 		/* get motion info */
 		int numverts, numkeys;
-		object_motion_info(kg, ccl_fetch(sd, object), NULL, &numverts, &numkeys);
+		object_motion_info(kg, sd->object, NULL, &numverts, &numkeys);
 
 		/* lookup attributes */
 		motion_pre = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
 
-		desc.offset += (ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE)? numverts: numkeys;
+		desc.offset += (sd->type & PRIMITIVE_ALL_TRIANGLE)? numverts: numkeys;
 		motion_post = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
 
 #ifdef __HAIR__
-		if(is_curve_primitive && (ccl_fetch(sd, object_flag) & SD_OBJECT_HAS_VERTEX_MOTION) == 0) {
+		if(is_curve_primitive && (sd->object_flag & SD_OBJECT_HAS_VERTEX_MOTION) == 0) {
 			object_position_transform(kg, sd, &motion_pre);
 			object_position_transform(kg, sd, &motion_post);
 		}
@@ -193,10 +193,10 @@ ccl_device_inline float4 primitive_motion_vector(KernelGlobals *kg, ShaderData *
 	 * transformation was set match the world/object space of motion_pre/post */
 	Transform tfm;
 	
-	tfm = object_fetch_vector_transform(kg, ccl_fetch(sd, object), OBJECT_VECTOR_MOTION_PRE);
+	tfm = object_fetch_vector_transform(kg, sd->object, OBJECT_VECTOR_MOTION_PRE);
 	motion_pre = transform_point(&tfm, motion_pre);
 
-	tfm = object_fetch_vector_transform(kg, ccl_fetch(sd, object), OBJECT_VECTOR_MOTION_POST);
+	tfm = object_fetch_vector_transform(kg, sd->object, OBJECT_VECTOR_MOTION_POST);
 	motion_post = transform_point(&tfm, motion_post);
 
 	float3 motion_center;
diff --git a/intern/cycles/kernel/geom/geom_subd_triangle.h b/intern/cycles/kernel/geom/geom_subd_triangle.h
index 647840dc696..044e82f03d4 100644
--- a/intern/cycles/kernel/geom/geom_subd_triangle.h
+++ b/intern/cycles/kernel/geom/geom_subd_triangle.h
@@ -22,14 +22,14 @@ CCL_NAMESPACE_BEGIN
 
 ccl_device_inline uint subd_triangle_patch(KernelGlobals *kg, const ShaderData *sd)
 {
-	return (ccl_fetch(sd, prim) != PRIM_NONE) ? kernel_tex_fetch(__tri_patch, ccl_fetch(sd, prim)) : ~0;
+	return (sd->prim != PRIM_NONE) ? kernel_tex_fetch(__tri_patch, sd->prim) : ~0;
 }
 
 /* UV coords of triangle within patch */
 
 ccl_device_inline void subd_triangle_patch_uv(KernelGlobals *kg, const ShaderData *sd, float2 uv[3])
 {
-	uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+	uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, sd->prim);
 
 	uv[0] = kernel_tex_fetch(__tri_patch_uv, tri_vindex.x);
 	uv[1] = kernel_tex_fetch(__tri_patch_uv, tri_vindex.y);
@@ -110,7 +110,7 @@ ccl_device_noinline float subd_triangle_attribute_float(KernelGlobals *kg, const
 		float2 dpdv = uv[1] - uv[2];
 
 		/* p is [s, t] */
-		float2 p = dpdu * ccl_fetch(sd, u) + dpdv * ccl_fetch(sd, v) + uv[2];
+		float2 p = dpdu * sd->u + dpdv * sd->v + uv[2];
 
 		float a, dads, dadt;
 		a = patch_eval_float(kg, sd, desc.offset, patch, p.x, p.y, 0, &dads, &dadt);
@@ -123,8 +123,8 @@ ccl_device_noinline float subd_triangle_attribute_float(KernelGlobals *kg, const
 			float dtdv = dpdv.y;
 
 			if(dx) {
-				float dudx = ccl_fetch(sd, du).dx;
-				float dvdx = ccl_fetch(sd, dv).dx;
+				float dudx = sd->du.dx;
+				float dvdx = sd->dv.dx;
 
 				float dsdx = dsdu*dudx + dsdv*dvdx;
 				float dtdx = dtdu*dudx + dtdv*dvdx;
@@ -132,8 +132,8 @@ ccl_device_noinline float subd_triangle_attribute_float(KernelGlobals *kg, const
 				*dx = dads*dsdx + dadt*dtdx;
 			}
 			if(dy) {
-				float dudy = ccl_fetch(sd, du).dy;
-				float dvdy = ccl_fetch(sd, dv).dy;
+				float dudy = sd->du.dy;
+				float dvdy = sd->dv.dy;
 
 				float dsdy = dsdu*dudy + dsdv*dvdy;
 				float dtdy = dtdu*dudy + dtdv*dvdy;
@@ -174,11 +174,11 @@ ccl_device_noinline float subd_triangle_attribute_float(KernelGlobals *kg, const
 		float c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
-		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+		if(dx) *dx = sd->du.dx*a + sd->dv.dx*b - (sd->du.dx + sd->dv.dx)*c;
+		if(dy) *dy = sd->du.dy*a + sd->dv.dy*b - (sd->du.dy + sd->dv.dy)*c;
 #endif
 
-		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+		return sd->u*a + sd->v*b + (1.0f - sd->u - sd->v)*c;
 	}
 	else if(desc.element == ATTR_ELEMENT_CORNER) {
 		float2 uv[3];
@@ -202,11 +202,11 @@ ccl_device_noinline float subd_triangle_attribute_float(KernelGlobals *kg, const
 		float c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
-		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+		if(dx) *dx = sd->du.dx*a + sd->dv.dx*b - (sd->du.dx + sd->dv.dx)*c;
+		if(dy) *dy = sd->du.dy*a + sd->dv.dy*b - (sd->du.dy + sd->dv.dy)*c;
 #endif
 
-		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+		return sd->u*a + sd->v*b + (1.0f - sd->u - sd->v)*c;
 	}
 	else {
 		if(dx) *dx = 0.0f;
@@ -229,7 +229,7 @@ ccl_device_noinline float3 subd_triangle_attribute_float3(KernelGlobals *kg, con
 		float2 dpdv = uv[1] - uv[2];
 
 		/* p is [s, t] */
-		float2 p = dpdu * ccl_fetch(sd, u) + dpdv * ccl_fetch(sd, v) + uv[2];
+		float2 p = dpdu * sd->u + dpdv * sd->v + uv[2];
 
 		float3 a, dads, dadt;
 
@@ -248,8 +248,8 @@ ccl_device_noinline float3 subd_triangle_attribute_float3(KernelGlobals *kg, con
 			float dtdv = dpdv.y;
 
 			if(dx) {
-				float dudx = ccl_fetch(sd, du).dx;
-				float dvdx = ccl_fetch(sd, dv).dx;
+				float dudx = sd->du.dx;
+				float dvdx = sd->dv.dx;
 
 				float dsdx = dsdu*dudx + dsdv*dvdx;
 				float dtdx = dtdu*dudx + dtdv*dvdx;
@@ -257,8 +257,8 @@ ccl_device_noinline float3 subd_triangle_attribute_float3(KernelGlobals *kg, con
 				*dx = dads*dsdx + dadt*dtdx;
 			}
 			if(dy) {
-				float dudy = ccl_fetch(sd, du).dy;
-				float dvdy = ccl_fetch(sd, dv).dy;
+				float dudy = sd->du.dy;
+				float dvdy = sd->dv.dy;
 
 				float dsdy = dsdu*dudy + dsdv*dvdy;
 				float dtdy = dtdu*dudy + dtdv*dvdy;
@@ -299,11 +299,11 @@ ccl_device_noinline float3 subd_triangle_attribute_float3(KernelGlobals *kg, con
 		float3 c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
-		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+		if(dx) *dx = sd->du.dx*a + sd->dv.dx*b - (sd->du.dx + sd->dv.dx)*c;
+		if(dy) *dy = sd->du.dy*a + sd->dv.dy*b - (sd->du.dy + sd->dv.dy)*c;
 #endif
 
-		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+		return sd->u*a + sd->v*b + (1.0f - sd->u - sd->v)*c;
 	}
 	else if(desc.element == ATTR_ELEMENT_CORNER || desc.element == ATTR_ELEMENT_CORNER_BYTE) {
 		float2 uv[3];
@@ -337,11 +337,11 @@ ccl_device_noinline float3 subd_triangle_attribute_float3(KernelGlobals *kg, con
 		float3 c = mix(mix(f0, f1, uv[2].x), mix(f3, f2, uv[2].x), uv[2].y);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*a + ccl_fetch(sd, dv).dx*b - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*c;
-		if(dy) *dy = ccl_fetch(sd, du).dy*a + ccl_fetch(sd, dv).dy*b - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*c;
+		if(dx) *dx = sd->du.dx*a + sd->dv.dx*b - (sd->du.dx + sd->dv.dx)*c;
+		if(dy) *dy = sd->du.dy*a + sd->dv.dy*b - (sd->du.dy + sd->dv.dy)*c;
 #endif
 
-		return ccl_fetch(sd, u)*a + ccl_fetch(sd, v)*b + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*c;
+		return sd->u*a + sd->v*b + (1.0f - sd->u - sd->v)*c;
 	}
 	else {
 		if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
diff --git a/intern/cycles/kernel/geom/geom_triangle.h b/intern/cycles/kernel/geom/geom_triangle.h
index 3229091bbb0..47778553b94 100644
--- a/intern/cycles/kernel/geom/geom_triangle.h
+++ b/intern/cycles/kernel/geom/geom_triangle.h
@@ -26,13 +26,13 @@ CCL_NAMESPACE_BEGIN
 ccl_device_inline float3 triangle_normal(KernelGlobals *kg, ShaderData *sd)
 {
 	/* load triangle vertices */
-	const uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+	const uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, sd->prim);
 	const float3 v0 = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w+0));
 	const float3 v1 = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w+1));
 	const float3 v2 = float4_to_float3(kernel_tex_fetch(__prim_tri_verts, tri_vindex.w+2));
 
 	/* return normal */
-	if(ccl_fetch(sd, object_flag) & SD_OBJECT_NEGATIVE_SCALE_APPLIED) {
+	if(sd->object_flag & SD_OBJECT_NEGATIVE_SCALE_APPLIED) {
 		return normalize(cross(v2 - v0, v1 - v0));
 	}
 	else {
@@ -110,34 +110,34 @@ ccl_device float triangle_attribute_float(KernelGlobals *kg, const ShaderData *s
 		if(dx) *dx = 0.0f;
 		if(dy) *dy = 0.0f;
 
-		return kernel_tex_fetch(__attributes_float, desc.offset + ccl_fetch(sd, prim));
+		return kernel_tex_fetch(__attributes_float, desc.offset + sd->prim);
 	}
 	else if(desc.element == ATTR_ELEMENT_VERTEX || desc.element == ATTR_ELEMENT_VERTEX_MOTION) {
-		uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+		uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, sd->prim);
 
 		float f0 = kernel_tex_fetch(__attributes_float, desc.offset + tri_vindex.x);
 		float f1 = kernel_tex_fetch(__attributes_float, desc.offset + tri_vindex.y);
 		float f2 = kernel_tex_fetch(__attributes_float, desc.offset + tri_vindex.z);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*f0 + ccl_fetch(sd, dv).dx*f1 - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*f2;
-		if(dy) *dy = ccl_fetch(sd, du).dy*f0 + ccl_fetch(sd, dv).dy*f1 - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*f2;
+		if(dx) *dx = sd->du.dx*f0 + sd->dv.dx*f1 - (sd->du.dx + sd->dv.dx)*f2;
+		if(dy) *dy = sd->du.dy*f0 + sd->dv.dy*f1 - (sd->du.dy + sd->dv.dy)*f2;
 #endif
 
-		return ccl_fetch(sd, u)*f0 + ccl_fetch(sd, v)*f1 + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*f2;
+		return sd->u*f0 + sd->v*f1 + (1.0f - sd->u - sd->v)*f2;
 	}
 	else if(desc.element == ATTR_ELEMENT_CORNER) {
-		int tri = desc.offset + ccl_fetch(sd, prim)*3;
+		int tri = desc.offset + sd->prim*3;
 		float f0 = kernel_tex_fetch(__attributes_float, tri + 0);
 		float f1 = kernel_tex_fetch(__attributes_float, tri + 1);
 		float f2 = kernel_tex_fetch(__attributes_float, tri + 2);
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*f0 + ccl_fetch(sd, dv).dx*f1 - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*f2;
-		if(dy) *dy = ccl_fetch(sd, du).dy*f0 + ccl_fetch(sd, dv).dy*f1 - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*f2;
+		if(dx) *dx = sd->du.dx*f0 + sd->dv.dx*f1 - (sd->du.dx + sd->dv.dx)*f2;
+		if(dy) *dy = sd->du.dy*f0 + sd->dv.dy*f1 - (sd->du.dy + sd->dv.dy)*f2;
 #endif
 
-		return ccl_fetch(sd, u)*f0 + ccl_fetch(sd, v)*f1 + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*f2;
+		return sd->u*f0 + sd->v*f1 + (1.0f - sd->u - sd->v)*f2;
 	}
 	else {
 		if(dx) *dx = 0.0f;
@@ -153,24 +153,24 @@ ccl_device float3 triangle_attribute_float3(KernelGlobals *kg, const ShaderData
 		if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
 		if(dy) *dy = make_float3(0.0f, 0.0f, 0.0f);
 
-		return float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + ccl_fetch(sd, prim)));
+		return float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + sd->prim));
 	}
 	else if(desc.element == ATTR_ELEMENT_VERTEX || desc.element == ATTR_ELEMENT_VERTEX_MOTION) {
-		uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, ccl_fetch(sd, prim));
+		uint4 tri_vindex = kernel_tex_fetch(__tri_vindex, sd->prim);
 
 		float3 f0 = float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + tri_vindex.x));
 		float3 f1 = float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + tri_vindex.y));
 		float3 f2 = float4_to_float3(kernel_tex_fetch(__attributes_float3, desc.offset + tri_vindex.z));
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*f0 + ccl_fetch(sd, dv).dx*f1 - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*f2;
-		if(dy) *dy = ccl_fetch(sd, du).dy*f0 + ccl_fetch(sd, dv).dy*f1 - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*f2;
+		if(dx) *dx = sd->du.dx*f0 + sd->dv.dx*f1 - (sd->du.dx + sd->dv.dx)*f2;
+		if(dy) *dy = sd->du.dy*f0 + sd->dv.dy*f1 - (sd->du.dy + sd->dv.dy)*f2;
 #endif
 
-		return ccl_fetch(sd, u)*f0 + ccl_fetch(sd, v)*f1 + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*f2;
+		return sd->u*f0 + sd->v*f1 + (1.0f - sd->u - sd->v)*f2;
 	}
 	else if(desc.element == ATTR_ELEMENT_CORNER || desc.element == ATTR_ELEMENT_CORNER_BYTE) {
-		int tri = desc.offset + ccl_fetch(sd, prim)*3;
+		int tri = desc.offset + sd->prim*3;
 		float3 f0, f1, f2;
 
 		if(desc.element == ATTR_ELEMENT_CORNER) {
@@ -185,11 +185,11 @@ ccl_device float3 triangle_attribute_float3(KernelGlobals *kg, const ShaderData
 		}
 
 #ifdef __RAY_DIFFERENTIALS__
-		if(dx) *dx = ccl_fetch(sd, du).dx*f0 + ccl_fetch(sd, dv).dx*f1 - (ccl_fetch(sd, du).dx + ccl_fetch(sd, dv).dx)*f2;
-		if(dy) *dy = ccl_fetch(sd, du).dy*f0 + ccl_fetch(sd, dv).dy*f1 - (ccl_fetch(sd, du).dy + ccl_fetch(sd, dv).dy)*f2;
+		if(dx) *dx = sd->du.dx*f0 + sd->dv.dx*f1 - (sd->du.dx + sd->dv.dx)*f2;
+		if(dy) *dy = sd->du.dy*f0 + sd->dv.dy*f1 - (sd->du.dy + sd->dv.dy)*f2;
 #endif
 
-		return ccl_fetch(sd, u)*f0 + ccl_fetch(sd, v)*f1 + (1.0f - ccl_fetch(sd, u) - ccl_fetch(sd, v))*f2;
+		return sd->u*f0 + sd->v*f1 + (1.0f - sd->u - sd->v)*f2;
 	}
 	else {
 		if(dx) *dx = make_float3(0.0f, 0.0f, 0.0f);
diff --git a/intern/cycles/kernel/geom/geom_triangle_intersect.h b/intern/cycles/kernel/geom/geom_triangle_intersect.h
index 4db121d94f4..4d234dd62bd 100644
--- a/intern/cycles/kernel/geom/geom_triangle_intersect.h
+++ b/intern/cycles/kernel/geom/geom_triangle_intersect.h
@@ -457,7 +457,7 @@ ccl_device_inline float3 triangle_refine(KernelGlobals *kg,
 			return P;
 		}
 #  ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_itfm);
+		Transform tfm = sd->ob_itfm;
 #  else
 		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_INVERSE_TRANSFORM);
 #  endif
@@ -491,7 +491,7 @@ ccl_device_inline float3 triangle_refine(KernelGlobals *kg,
 
 	if(isect->object != OBJECT_NONE) {
 #  ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_tfm);
+		Transform tfm = sd->ob_tfm;
 #  else
 		Transform tfm = object_fetch_transform(kg, isect->object, OBJECT_TRANSFORM);
 #  endif
@@ -519,7 +519,7 @@ ccl_device_inline float3 triangle_refine_subsurface(KernelGlobals *kg,
 
 	if(isect->object != OBJECT_NONE) {
 #ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_itfm);
+		Transform tfm = sd->ob_itfm;
 #else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
@@ -557,7 +557,7 @@ ccl_device_inline float3 triangle_refine_subsurface(KernelGlobals *kg,
 
 	if(isect->object != OBJECT_NONE) {
 #ifdef __OBJECT_MOTION__
-		Transform tfm = ccl_fetch(sd, ob_tfm);
+		Transform tfm = sd->ob_tfm;
 #else
 		Transform tfm = object_fetch_transform(kg,
 		                                       isect->object,
diff --git a/intern/cycles/kernel/geom/geom_volume.h b/intern/cycles/kernel/geom/geom_volume.h
index 03724c955be..1e0ef5201c9 100644
--- a/intern/cycles/kernel/geom/geom_volume.h
+++ b/intern/cycles/kernel/geom/geom_volume.h
@@ -64,7 +64,7 @@ ccl_device_inline float3 volume_normalized_position(KernelGlobals *kg,
 
 ccl_device float volume_attribute_float(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc, float *dx, float *dy)
 {
-	float3 P = volume_normalized_position(kg, sd, ccl_fetch(sd, P));
+	float3 P = volume_normalized_position(kg, sd, sd->P);
 #ifdef __KERNEL_CUDA__
 #  if __CUDA_ARCH__ >= 300
 	CUtexObject tex = kernel_tex_fetch(__bindless_mapping, desc.offset);
@@ -91,7 +91,7 @@ ccl_device float volume_attribute_float(KernelGlobals *kg, const ShaderData *sd,
 
 ccl_device float3 volume_attribute_float3(KernelGlobals *kg, const ShaderData *sd, const AttributeDescriptor desc, float3 *dx, float3 *dy)
 {
-	float3 P = volume_normalized_position(kg, sd, ccl_fetch(sd, P));
+	float3 P = volume_normalized_position(kg, sd, sd->P);
 #ifdef __KERNEL_CUDA__
 #  if __CUDA_ARCH__ >= 300
 	CUtexObject tex = kernel_tex_fetch(__bindless_mapping, desc.offset);
diff --git a/intern/cycles/kernel/kernel_camera.h b/intern/cycles/kernel/kernel_camera.h
index dedac6b1465..0df5217d97a 100644
--- a/intern/cycles/kernel/kernel_camera.h
+++ b/intern/cycles/kernel/kernel_camera.h
@@ -457,7 +457,7 @@ ccl_device_inline float3 camera_world_to_ndc(KernelGlobals *kg, ShaderData *sd,
 {
 	if(kernel_data.cam.type != CAMERA_PANORAMA) {
 		/* perspective / ortho */
-		if(ccl_fetch(sd, object) == PRIM_NONE && kernel_data.cam.type == CAMERA_PERSPECTIVE)
+		if(sd->object == PRIM_NONE && kernel_data.cam.type == CAMERA_PERSPECTIVE)
 			P += camera_position(kg);
 
 		Transform tfm = kernel_data.cam.worldtondc;
@@ -467,7 +467,7 @@ ccl_device_inline float3 camera_world_to_ndc(KernelGlobals *kg, ShaderData *sd,
 		/* panorama */
 		Transform tfm = kernel_data.cam.worldtocamera;
 
-		if(ccl_fetch(sd, object) != OBJECT_NONE)
+		if(sd->object != OBJECT_NONE)
 			P = normalize(transform_point(&tfm, P));
 		else
 			P = normalize(transform_direction(&tfm, P));
diff --git a/intern/cycles/kernel/kernel_emission.h b/intern/cycles/kernel/kernel_emission.h
index 8c7c651a053..bc2d9604122 100644
--- a/intern/cycles/kernel/kernel_emission.h
+++ b/intern/cycles/kernel/kernel_emission.h
@@ -67,7 +67,7 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
 		                         ls->shader, ls->object, ls->prim,
 		                         ls->u, ls->v, t, time, false, ls->lamp);
 
-		ls->Ng = ccl_fetch(emission_sd, Ng);
+		ls->Ng = emission_sd->Ng;
 
 		/* no path flag, we're evaluating this for all closures. that's weak but
 		 * we'd have to do multiple evaluations otherwise */
@@ -76,7 +76,7 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
 		path_state_modify_bounce(state, false);
 
 		/* evaluate emissive closure */
-		if(ccl_fetch(emission_sd, flag) & SD_EMISSION)
+		if(emission_sd->flag & SD_EMISSION)
 			eval = shader_emissive_eval(kg, emission_sd);
 		else
 			eval = make_float3(0.0f, 0.0f, 0.0f);
@@ -112,7 +112,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg,
 	                                         -ls->D,
 	                                         dD,
 	                                         ls->t,
-	                                         ccl_fetch(sd, time));
+	                                         sd->time);
 
 	if(is_zero(light_eval))
 		return false;
@@ -120,7 +120,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg,
 	/* evaluate BSDF at shading point */
 
 #ifdef __VOLUME__
-	if(ccl_fetch(sd, prim) != PRIM_NONE)
+	if(sd->prim != PRIM_NONE)
 		shader_bsdf_eval(kg, sd, ls->D, eval, ls->pdf, ls->shader & SHADER_USE_MIS);
 	else {
 		float bsdf_pdf;
@@ -168,8 +168,8 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg,
 
 	if(ls->shader & SHADER_CAST_SHADOW) {
 		/* setup ray */
-		bool transmit = (dot(ccl_fetch(sd, Ng), ls->D) < 0.0f);
-		ray->P = ray_offset(ccl_fetch(sd, P), (transmit)? -ccl_fetch(sd, Ng): ccl_fetch(sd, Ng));
+		bool transmit = (dot(sd->Ng, ls->D) < 0.0f);
+		ray->P = ray_offset(sd->P, (transmit)? -sd->Ng: sd->Ng);
 
 		if(ls->t == FLT_MAX) {
 			/* distant light */
@@ -182,7 +182,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg,
 			ray->D = normalize_len(ray->D, &ray->t);
 		}
 
-		ray->dP = ccl_fetch(sd, dP);
+		ray->dP = sd->dP;
 		ray->dD = differential3_zero();
 	}
 	else {
@@ -204,14 +204,14 @@ ccl_device_noinline float3 indirect_primitive_emission(KernelGlobals *kg, Shader
 	float3 L = shader_emissive_eval(kg, sd);
 
 #ifdef __HAIR__
-	if(!(path_flag & PATH_RAY_MIS_SKIP) && (ccl_fetch(sd, flag) & SD_USE_MIS) && (ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE))
+	if(!(path_flag & PATH_RAY_MIS_SKIP) && (sd->flag & SD_USE_MIS) && (sd->type & PRIMITIVE_ALL_TRIANGLE))
 #else
-	if(!(path_flag & PATH_RAY_MIS_SKIP) && (ccl_fetch(sd, flag) & SD_USE_MIS))
+	if(!(path_flag & PATH_RAY_MIS_SKIP) && (sd->flag & SD_USE_MIS))
 #endif
 	{
 		/* multiple importance sampling, get triangle light pdf,
 		 * and compute weight with respect to BSDF pdf */
-		float pdf = triangle_light_pdf(kg, ccl_fetch(sd, Ng), ccl_fetch(sd, I), t);
+		float pdf = triangle_light_pdf(kg, sd->Ng, sd->I, t);
 		float mis_weight = power_heuristic(bsdf_pdf, pdf);
 
 		return L*mis_weight;
diff --git a/intern/cycles/kernel/kernel_passes.h b/intern/cycles/kernel/kernel_passes.h
index 7790cce067b..ed523696571 100644
--- a/intern/cycles/kernel/kernel_passes.h
+++ b/intern/cycles/kernel/kernel_passes.h
@@ -75,18 +75,18 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
 		return;
 	
 	if(!(path_flag & PATH_RAY_SINGLE_PASS_DONE)) {
-		if(!(ccl_fetch(sd, flag) & SD_TRANSPARENT) ||
+		if(!(sd->flag & SD_TRANSPARENT) ||
 		   kernel_data.film.pass_alpha_threshold == 0.0f ||
 		   average(shader_bsdf_alpha(kg, sd)) >= kernel_data.film.pass_alpha_threshold)
 		{
 
 			if(sample == 0) {
 				if(flag & PASS_DEPTH) {
-					float depth = camera_distance(kg, ccl_fetch(sd, P));
+					float depth = camera_distance(kg, sd->P);
 					kernel_write_pass_float(buffer + kernel_data.film.pass_depth, sample, depth);
 				}
 				if(flag & PASS_OBJECT_ID) {
-					float id = object_pass_id(kg, ccl_fetch(sd, object));
+					float id = object_pass_id(kg, sd->object);
 					kernel_write_pass_float(buffer + kernel_data.film.pass_object_id, sample, id);
 				}
 				if(flag & PASS_MATERIAL_ID) {
@@ -96,7 +96,7 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
 			}
 
 			if(flag & PASS_NORMAL) {
-				float3 normal = ccl_fetch(sd, N);
+				float3 normal = sd->N;
 				kernel_write_pass_float3(buffer + kernel_data.film.pass_normal, sample, normal);
 			}
 			if(flag & PASS_UV) {
@@ -127,7 +127,7 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
 		float mist_start = kernel_data.film.mist_start;
 		float mist_inv_depth = kernel_data.film.mist_inv_depth;
 
-		float depth = camera_distance(kg, ccl_fetch(sd, P));
+		float depth = camera_distance(kg, sd->P);
 		float mist = saturate((depth - mist_start)*mist_inv_depth);
 
 		/* falloff */
diff --git a/intern/cycles/kernel/kernel_path.h b/intern/cycles/kernel/kernel_path.h
index f90701a8260..95c27850513 100644
--- a/intern/cycles/kernel/kernel_path.h
+++ b/intern/cycles/kernel/kernel_path.h
@@ -75,17 +75,17 @@ ccl_device_noinline void kernel_path_ao(KernelGlobals *kg,
 
 	sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
-	if(dot(ccl_fetch(sd, Ng), ao_D) > 0.0f && ao_pdf != 0.0f) {
+	if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
 		Ray light_ray;
 		float3 ao_shadow;
 
-		light_ray.P = ray_offset(ccl_fetch(sd, P), ccl_fetch(sd, Ng));
+		light_ray.P = ray_offset(sd->P, sd->Ng);
 		light_ray.D = ao_D;
 		light_ray.t = kernel_data.background.ao_distance;
 #ifdef __OBJECT_MOTION__
-		light_ray.time = ccl_fetch(sd, time);
+		light_ray.time = sd->time;
 #endif  /* __OBJECT_MOTION__ */
-		light_ray.dP = ccl_fetch(sd, dP);
+		light_ray.dP = sd->dP;
 		light_ray.dD = differential3_zero();
 
 		if(!shadow_blocked(kg, emission_sd, state, &light_ray, &ao_shadow)) {
@@ -459,7 +459,7 @@ bool kernel_path_subsurface_scatter(
 #  ifdef __VOLUME__
 		ss_indirect->need_update_volume_stack =
 		        kernel_data.integrator.use_volumes &&
-		        ccl_fetch(sd, object_flag) & SD_OBJECT_INTERSECTS_VOLUME;
+		        sd->object_flag & SD_OBJECT_INTERSECTS_VOLUME;
 #  endif  /* __VOLUME__ */
 
 		/* compute lighting with the BSDF closure */
diff --git a/intern/cycles/kernel/kernel_path_branched.h b/intern/cycles/kernel/kernel_path_branched.h
index ff2b828795d..d58960cae4e 100644
--- a/intern/cycles/kernel/kernel_path_branched.h
+++ b/intern/cycles/kernel/kernel_path_branched.h
@@ -42,17 +42,17 @@ ccl_device_inline void kernel_branched_path_ao(KernelGlobals *kg,
 
 		sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
-		if(dot(ccl_fetch(sd, Ng), ao_D) > 0.0f && ao_pdf != 0.0f) {
+		if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
 			Ray light_ray;
 			float3 ao_shadow;
 
-			light_ray.P = ray_offset(ccl_fetch(sd, P), ccl_fetch(sd, Ng));
+			light_ray.P = ray_offset(sd->P, sd->Ng);
 			light_ray.D = ao_D;
 			light_ray.t = kernel_data.background.ao_distance;
 #ifdef __OBJECT_MOTION__
-			light_ray.time = ccl_fetch(sd, time);
+			light_ray.time = sd->time;
 #endif  /* __OBJECT_MOTION__ */
-			light_ray.dP = ccl_fetch(sd, dP);
+			light_ray.dP = sd->dP;
 			light_ray.dD = differential3_zero();
 
 			if(!shadow_blocked(kg, emission_sd, state, &light_ray, &ao_shadow))
@@ -67,8 +67,8 @@ ccl_device_noinline void kernel_branched_path_surface_indirect_light(KernelGloba
 	RNG *rng, ShaderData *sd, ShaderData *indirect_sd, ShaderData *emission_sd,
 	float3 throughput, float num_samples_adjust, PathState *state, PathRadiance *L)
 {
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		const ShaderClosure *sc = &ccl_fetch(sd, closure)[i];
+	for(int i = 0; i < sd->num_closure; i++) {
+		const ShaderClosure *sc = &sd->closure[i];
 
 		if(!CLOSURE_IS_BSDF(sc->type))
 			continue;
@@ -140,8 +140,8 @@ ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg,
                                                         Ray *ray,
                                                         float3 throughput)
 {
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = &ccl_fetch(sd, closure)[i];
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(!CLOSURE_IS_BSSRDF(sc->type))
 			continue;
@@ -169,7 +169,7 @@ ccl_device void kernel_branched_path_subsurface_scatter(KernelGlobals *kg,
 			Ray volume_ray = *ray;
 			bool need_update_volume_stack =
 			        kernel_data.integrator.use_volumes &&
-			        ccl_fetch(sd, object_flag) & SD_OBJECT_INTERSECTS_VOLUME;
+			        sd->object_flag & SD_OBJECT_INTERSECTS_VOLUME;
 #endif  /* __VOLUME__ */
 
 			/* compute lighting with the BSDF closure */
diff --git a/intern/cycles/kernel/kernel_path_surface.h b/intern/cycles/kernel/kernel_path_surface.h
index fea503d06e5..34a78552c1d 100644
--- a/intern/cycles/kernel/kernel_path_surface.h
+++ b/intern/cycles/kernel/kernel_path_surface.h
@@ -25,7 +25,7 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 {
 #ifdef __EMISSION__
 	/* sample illumination from lights to find path contribution */
-	if(!(ccl_fetch(sd, flag) & SD_BSDF_HAS_EVAL))
+	if(!(sd->flag & SD_BSDF_HAS_EVAL))
 		return;
 
 	Ray light_ray;
@@ -33,7 +33,7 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 	bool is_lamp;
 
 #  ifdef __OBJECT_MOTION__
-	light_ray.time = ccl_fetch(sd, time);
+	light_ray.time = sd->time;
 #  endif
 
 	if(sample_all_lights) {
@@ -52,7 +52,7 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 				float terminate = path_branched_rng_light_termination(kg, &lamp_rng, state, j, num_samples);
 
 				LightSample ls;
-				if(lamp_light_sample(kg, i, light_u, light_v, ccl_fetch(sd, P), &ls)) {
+				if(lamp_light_sample(kg, i, light_u, light_v, sd->P, &ls)) {
 					/* The sampling probability returned by lamp_light_sample assumes that all lights were sampled.
 					 * However, this code only samples lamps, so if the scene also had mesh lights, the real probability is twice as high. */
 					if(kernel_data.integrator.pdf_triangles != 0.0f)
@@ -87,7 +87,7 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 					light_t = 0.5f*light_t;
 
 				LightSample ls;
-				if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+				if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 					/* Same as above, probability needs to be corrected since the sampling was forced to select a mesh light. */
 					if(kernel_data.integrator.num_all_lights)
 						ls.pdf *= 2.0f;
@@ -113,7 +113,7 @@ ccl_device_noinline void kernel_branched_path_surface_connect_light(KernelGlobal
 		float terminate = path_state_rng_light_termination(kg, rng, state);
 
 		LightSample ls;
-		if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+		if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 			/* sample random light */
 			if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
 				/* trace shadow ray */
@@ -156,15 +156,15 @@ ccl_device bool kernel_branched_path_surface_bounce(KernelGlobals *kg, RNG *rng,
 	path_state_next(kg, state, label);
 
 	/* setup ray */
-	ray->P = ray_offset(ccl_fetch(sd, P), (label & LABEL_TRANSMIT)? -ccl_fetch(sd, Ng): ccl_fetch(sd, Ng));
+	ray->P = ray_offset(sd->P, (label & LABEL_TRANSMIT)? -sd->Ng: sd->Ng);
 	ray->D = normalize(bsdf_omega_in);
 	ray->t = FLT_MAX;
 #ifdef __RAY_DIFFERENTIALS__
-	ray->dP = ccl_fetch(sd, dP);
+	ray->dP = sd->dP;
 	ray->dD = bsdf_domega_in;
 #endif
 #ifdef __OBJECT_MOTION__
-	ray->time = ccl_fetch(sd, time);
+	ray->time = sd->time;
 #endif
 
 #ifdef __VOLUME__
@@ -195,7 +195,7 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg, ccl_
 	PathRadiance *L)
 {
 #ifdef __EMISSION__
-	if(!(kernel_data.integrator.use_direct_light && (ccl_fetch(sd, flag) & SD_BSDF_HAS_EVAL)))
+	if(!(kernel_data.integrator.use_direct_light && (sd->flag & SD_BSDF_HAS_EVAL)))
 		return;
 
 	/* sample illumination from lights to find path contribution */
@@ -208,11 +208,11 @@ ccl_device_inline void kernel_path_surface_connect_light(KernelGlobals *kg, ccl_
 	bool is_lamp;
 
 #ifdef __OBJECT_MOTION__
-	light_ray.time = ccl_fetch(sd, time);
+	light_ray.time = sd->time;
 #endif
 
 	LightSample ls;
-	if(light_sample(kg, light_t, light_u, light_v, ccl_fetch(sd, time), ccl_fetch(sd, P), state->bounce, &ls)) {
+	if(light_sample(kg, light_t, light_u, light_v, sd->time, sd->P, state->bounce, &ls)) {
 		float terminate = path_state_rng_light_termination(kg, rng, state);
 		if(direct_emission(kg, sd, emission_sd, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
 			/* trace shadow ray */
@@ -238,7 +238,7 @@ ccl_device bool kernel_path_surface_bounce(KernelGlobals *kg,
                                            ccl_addr_space Ray *ray)
 {
 	/* no BSDF? we can stop here */
-	if(ccl_fetch(sd, flag) & SD_BSDF) {
+	if(sd->flag & SD_BSDF) {
 		/* sample BSDF */
 		float bsdf_pdf;
 		BsdfEval bsdf_eval;
@@ -270,16 +270,16 @@ ccl_device bool kernel_path_surface_bounce(KernelGlobals *kg,
 		path_state_next(kg, state, label);
 
 		/* setup ray */
-		ray->P = ray_offset(ccl_fetch(sd, P), (label & LABEL_TRANSMIT)? -ccl_fetch(sd, Ng): ccl_fetch(sd, Ng));
+		ray->P = ray_offset(sd->P, (label & LABEL_TRANSMIT)? -sd->Ng: sd->Ng);
 		ray->D = normalize(bsdf_omega_in);
 
 		if(state->bounce == 0)
-			ray->t -= ccl_fetch(sd, ray_length); /* clipping works through transparent */
+			ray->t -= sd->ray_length; /* clipping works through transparent */
 		else
 			ray->t = FLT_MAX;
 
 #ifdef __RAY_DIFFERENTIALS__
-		ray->dP = ccl_fetch(sd, dP);
+		ray->dP = sd->dP;
 		ray->dD = bsdf_domega_in;
 #endif
 
@@ -291,21 +291,21 @@ ccl_device bool kernel_path_surface_bounce(KernelGlobals *kg,
 		return true;
 	}
 #ifdef __VOLUME__
-	else if(ccl_fetch(sd, flag) & SD_HAS_ONLY_VOLUME) {
+	else if(sd->flag & SD_HAS_ONLY_VOLUME) {
 		/* no surface shader but have a volume shader? act transparent */
 
 		/* update path state, count as transparent */
 		path_state_next(kg, state, LABEL_TRANSPARENT);
 
 		if(state->bounce == 0)
-			ray->t -= ccl_fetch(sd, ray_length); /* clipping works through transparent */
+			ray->t -= sd->ray_length; /* clipping works through transparent */
 		else
 			ray->t = FLT_MAX;
 
 		/* setup ray position, direction stays unchanged */
-		ray->P = ray_offset(ccl_fetch(sd, P), -ccl_fetch(sd, Ng));
+		ray->P = ray_offset(sd->P, -sd->Ng);
 #ifdef __RAY_DIFFERENTIALS__
-		ray->dP = ccl_fetch(sd, dP);
+		ray->dP = sd->dP;
 #endif
 
 		/* enter/exit volume */
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index 59c1331a63c..a2ab96b35e2 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -38,13 +38,13 @@ CCL_NAMESPACE_BEGIN
 #ifdef __OBJECT_MOTION__
 ccl_device void shader_setup_object_transforms(KernelGlobals *kg, ShaderData *sd, float time)
 {
-	if(ccl_fetch(sd, object_flag) & SD_OBJECT_MOTION) {
-		ccl_fetch(sd, ob_tfm) = object_fetch_transform_motion(kg, ccl_fetch(sd, object), time);
-		ccl_fetch(sd, ob_itfm) = transform_quick_inverse(ccl_fetch(sd, ob_tfm));
+	if(sd->object_flag & SD_OBJECT_MOTION) {
+		sd->ob_tfm = object_fetch_transform_motion(kg, sd->object, time);
+		sd->ob_itfm = transform_quick_inverse(sd->ob_tfm);
 	}
 	else {
-		ccl_fetch(sd, ob_tfm) = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_TRANSFORM);
-		ccl_fetch(sd, ob_itfm) = object_fetch_transform(kg, ccl_fetch(sd, object), OBJECT_INVERSE_TRANSFORM);
+		sd->ob_tfm = object_fetch_transform(kg, sd->object, OBJECT_TRANSFORM);
+		sd->ob_itfm = object_fetch_transform(kg, sd->object, OBJECT_INVERSE_TRANSFORM);
 	}
 }
 #endif
@@ -55,55 +55,55 @@ ccl_device_noinline void shader_setup_from_ray(KernelGlobals *kg,
                                                const Ray *ray)
 {
 #ifdef __INSTANCING__
-	ccl_fetch(sd, object) = (isect->object == PRIM_NONE)? kernel_tex_fetch(__prim_object, isect->prim): isect->object;
+	sd->object = (isect->object == PRIM_NONE)? kernel_tex_fetch(__prim_object, isect->prim): isect->object;
 #endif
 
-	ccl_fetch(sd, type) = isect->type;
-	ccl_fetch(sd, flag) = 0;
-	ccl_fetch(sd, object_flag) = kernel_tex_fetch(__object_flag,
-	                                              ccl_fetch(sd, object));
+	sd->type = isect->type;
+	sd->flag = 0;
+	sd->object_flag = kernel_tex_fetch(__object_flag,
+	                                              sd->object);
 
 	/* matrices and time */
 #ifdef __OBJECT_MOTION__
 	shader_setup_object_transforms(kg, sd, ray->time);
-	ccl_fetch(sd, time) = ray->time;
+	sd->time = ray->time;
 #endif
 
-	ccl_fetch(sd, prim) = kernel_tex_fetch(__prim_index, isect->prim);
-	ccl_fetch(sd, ray_length) = isect->t;
+	sd->prim = kernel_tex_fetch(__prim_index, isect->prim);
+	sd->ray_length = isect->t;
 
 #ifdef __UV__
-	ccl_fetch(sd, u) = isect->u;
-	ccl_fetch(sd, v) = isect->v;
+	sd->u = isect->u;
+	sd->v = isect->v;
 #endif
 
 #ifdef __HAIR__
-	if(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+	if(sd->type & PRIMITIVE_ALL_CURVE) {
 		/* curve */
-		float4 curvedata = kernel_tex_fetch(__curves, ccl_fetch(sd, prim));
+		float4 curvedata = kernel_tex_fetch(__curves, sd->prim);
 
-		ccl_fetch(sd, shader) = __float_as_int(curvedata.z);
-		ccl_fetch(sd, P) = bvh_curve_refine(kg, sd, isect, ray);
+		sd->shader = __float_as_int(curvedata.z);
+		sd->P = bvh_curve_refine(kg, sd, isect, ray);
 	}
 	else
 #endif
-	if(ccl_fetch(sd, type) & PRIMITIVE_TRIANGLE) {
+	if(sd->type & PRIMITIVE_TRIANGLE) {
 		/* static triangle */
 		float3 Ng = triangle_normal(kg, sd);
-		ccl_fetch(sd, shader) = kernel_tex_fetch(__tri_shader, ccl_fetch(sd, prim));
+		sd->shader = kernel_tex_fetch(__tri_shader, sd->prim);
 
 		/* vectors */
-		ccl_fetch(sd, P) = triangle_refine(kg, sd, isect, ray);
-		ccl_fetch(sd, Ng) = Ng;
-		ccl_fetch(sd, N) = Ng;
+		sd->P = triangle_refine(kg, sd, isect, ray);
+		sd->Ng = Ng;
+		sd->N = Ng;
 		
 		/* smooth normal */
-		if(ccl_fetch(sd, shader) & SHADER_SMOOTH_NORMAL)
-			ccl_fetch(sd, N) = triangle_smooth_normal(kg, ccl_fetch(sd, prim), ccl_fetch(sd, u), ccl_fetch(sd, v));
+		if(sd->shader & SHADER_SMOOTH_NORMAL)
+			sd->N = triangle_smooth_normal(kg, sd->prim, sd->u, sd->v);
 
 #ifdef __DPDU__
 		/* dPdu/dPdv */
-		triangle_dPdudv(kg, ccl_fetch(sd, prim), &ccl_fetch(sd, dPdu), &ccl_fetch(sd, dPdv));
+		triangle_dPdudv(kg, sd->prim, &sd->dPdu, &sd->dPdv);
 #endif
 	}
 	else {
@@ -111,40 +111,40 @@ ccl_device_noinline void shader_setup_from_ray(KernelGlobals *kg,
 		motion_triangle_shader_setup(kg, sd, isect, ray, false);
 	}
 
-	ccl_fetch(sd, I) = -ray->D;
+	sd->I = -ray->D;
 
-	ccl_fetch(sd, flag) |= kernel_tex_fetch(__shader_flag, (ccl_fetch(sd, shader) & SHADER_MASK)*SHADER_SIZE);
+	sd->flag |= kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE);
 
 #ifdef __INSTANCING__
 	if(isect->object != OBJECT_NONE) {
 		/* instance transform */
-		object_normal_transform_auto(kg, sd, &ccl_fetch(sd, N));
-		object_normal_transform_auto(kg, sd, &ccl_fetch(sd, Ng));
+		object_normal_transform_auto(kg, sd, &sd->N);
+		object_normal_transform_auto(kg, sd, &sd->Ng);
 #  ifdef __DPDU__
-		object_dir_transform_auto(kg, sd, &ccl_fetch(sd, dPdu));
-		object_dir_transform_auto(kg, sd, &ccl_fetch(sd, dPdv));
+		object_dir_transform_auto(kg, sd, &sd->dPdu);
+		object_dir_transform_auto(kg, sd, &sd->dPdv);
 #  endif
 	}
 #endif
 
 	/* backfacing test */
-	bool backfacing = (dot(ccl_fetch(sd, Ng), ccl_fetch(sd, I)) < 0.0f);
+	bool backfacing = (dot(sd->Ng, sd->I) < 0.0f);
 
 	if(backfacing) {
-		ccl_fetch(sd, flag) |= SD_BACKFACING;
-		ccl_fetch(sd, Ng) = -ccl_fetch(sd, Ng);
-		ccl_fetch(sd, N) = -ccl_fetch(sd, N);
+		sd->flag |= SD_BACKFACING;
+		sd->Ng = -sd->Ng;
+		sd->N = -sd->N;
 #ifdef __DPDU__
-		ccl_fetch(sd, dPdu) = -ccl_fetch(sd, dPdu);
-		ccl_fetch(sd, dPdv) = -ccl_fetch(sd, dPdv);
+		sd->dPdu = -sd->dPdu;
+		sd->dPdv = -sd->dPdv;
 #endif
 	}
 
 #ifdef __RAY_DIFFERENTIALS__
 	/* differentials */
-	differential_transfer(&ccl_fetch(sd, dP), ray->dP, ray->D, ray->dD, ccl_fetch(sd, Ng), isect->t);
-	differential_incoming(&ccl_fetch(sd, dI), ray->dD);
-	differential_dudv(&ccl_fetch(sd, du), &ccl_fetch(sd, dv), ccl_fetch(sd, dPdu), ccl_fetch(sd, dPdv), ccl_fetch(sd, dP), ccl_fetch(sd, Ng));
+	differential_transfer(&sd->dP, ray->dP, ray->D, ray->dD, sd->Ng, isect->t);
+	differential_incoming(&sd->dI, ray->dD);
+	differential_dudv(&sd->du, &sd->dv, sd->dPdu, sd->dPdv, sd->dP, sd->Ng);
 #endif
 }
 
@@ -249,106 +249,106 @@ ccl_device_inline void shader_setup_from_sample(KernelGlobals *kg,
                                                 int lamp)
 {
 	/* vectors */
-	ccl_fetch(sd, P) = P;
-	ccl_fetch(sd, N) = Ng;
-	ccl_fetch(sd, Ng) = Ng;
-	ccl_fetch(sd, I) = I;
-	ccl_fetch(sd, shader) = shader;
+	sd->P = P;
+	sd->N = Ng;
+	sd->Ng = Ng;
+	sd->I = I;
+	sd->shader = shader;
 	if(prim != PRIM_NONE)
-		ccl_fetch(sd, type) = PRIMITIVE_TRIANGLE;
+		sd->type = PRIMITIVE_TRIANGLE;
 	else if(lamp != LAMP_NONE)
-		ccl_fetch(sd, type) = PRIMITIVE_LAMP;
+		sd->type = PRIMITIVE_LAMP;
 	else
-		ccl_fetch(sd, type) = PRIMITIVE_NONE;
+		sd->type = PRIMITIVE_NONE;
 
 	/* primitive */
 #ifdef __INSTANCING__
-	ccl_fetch(sd, object) = object;
+	sd->object = object;
 #endif
 	/* currently no access to bvh prim index for strand sd->prim*/
-	ccl_fetch(sd, prim) = prim;
+	sd->prim = prim;
 #ifdef __UV__
-	ccl_fetch(sd, u) = u;
-	ccl_fetch(sd, v) = v;
+	sd->u = u;
+	sd->v = v;
 #endif
-	ccl_fetch(sd, ray_length) = t;
+	sd->ray_length = t;
 
-	ccl_fetch(sd, flag) = kernel_tex_fetch(__shader_flag, (ccl_fetch(sd, shader) & SHADER_MASK)*SHADER_SIZE);
-	ccl_fetch(sd, object_flag) = 0;
-	if(ccl_fetch(sd, object) != OBJECT_NONE) {
-		ccl_fetch(sd, object_flag) |= kernel_tex_fetch(__object_flag,
-		                                               ccl_fetch(sd, object));
+	sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE);
+	sd->object_flag = 0;
+	if(sd->object != OBJECT_NONE) {
+		sd->object_flag |= kernel_tex_fetch(__object_flag,
+		                                               sd->object);
 
 #ifdef __OBJECT_MOTION__
 		shader_setup_object_transforms(kg, sd, time);
-		ccl_fetch(sd, time) = time;
+		sd->time = time;
 	}
 	else if(lamp != LAMP_NONE) {
-		ccl_fetch(sd, ob_tfm)  = lamp_fetch_transform(kg, lamp, false);
-		ccl_fetch(sd, ob_itfm) = lamp_fetch_transform(kg, lamp, true);
+		sd->ob_tfm  = lamp_fetch_transform(kg, lamp, false);
+		sd->ob_itfm = lamp_fetch_transform(kg, lamp, true);
 #endif
 	}
 
 	/* transform into world space */
 	if(object_space) {
-		object_position_transform_auto(kg, sd, &ccl_fetch(sd, P));
-		object_normal_transform_auto(kg, sd, &ccl_fetch(sd, Ng));
-		ccl_fetch(sd, N) = ccl_fetch(sd, Ng);
-		object_dir_transform_auto(kg, sd, &ccl_fetch(sd, I));
+		object_position_transform_auto(kg, sd, &sd->P);
+		object_normal_transform_auto(kg, sd, &sd->Ng);
+		sd->N = sd->Ng;
+		object_dir_transform_auto(kg, sd, &sd->I);
 	}
 
-	if(ccl_fetch(sd, type) & PRIMITIVE_TRIANGLE) {
+	if(sd->type & PRIMITIVE_TRIANGLE) {
 		/* smooth normal */
-		if(ccl_fetch(sd, shader) & SHADER_SMOOTH_NORMAL) {
-			ccl_fetch(sd, N) = triangle_smooth_normal(kg, ccl_fetch(sd, prim), ccl_fetch(sd, u), ccl_fetch(sd, v));
+		if(sd->shader & SHADER_SMOOTH_NORMAL) {
+			sd->N = triangle_smooth_normal(kg, sd->prim, sd->u, sd->v);
 
 #ifdef __INSTANCING__
-			if(!(ccl_fetch(sd, object_flag) & SD_OBJECT_TRANSFORM_APPLIED)) {
-				object_normal_transform_auto(kg, sd, &ccl_fetch(sd, N));
+			if(!(sd->object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
+				object_normal_transform_auto(kg, sd, &sd->N);
 			}
 #endif
 		}
 
 		/* dPdu/dPdv */
 #ifdef __DPDU__
-		triangle_dPdudv(kg, ccl_fetch(sd, prim), &ccl_fetch(sd, dPdu), &ccl_fetch(sd, dPdv));
+		triangle_dPdudv(kg, sd->prim, &sd->dPdu, &sd->dPdv);
 
 #  ifdef __INSTANCING__
-		if(!(ccl_fetch(sd, object_flag) & SD_OBJECT_TRANSFORM_APPLIED)) {
-			object_dir_transform_auto(kg, sd, &ccl_fetch(sd, dPdu));
-			object_dir_transform_auto(kg, sd, &ccl_fetch(sd, dPdv));
+		if(!(sd->object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
+			object_dir_transform_auto(kg, sd, &sd->dPdu);
+			object_dir_transform_auto(kg, sd, &sd->dPdv);
 		}
 #  endif
 #endif
 	}
 	else {
 #ifdef __DPDU__
-		ccl_fetch(sd, dPdu) = make_float3(0.0f, 0.0f, 0.0f);
-		ccl_fetch(sd, dPdv) = make_float3(0.0f, 0.0f, 0.0f);
+		sd->dPdu = make_float3(0.0f, 0.0f, 0.0f);
+		sd->dPdv = make_float3(0.0f, 0.0f, 0.0f);
 #endif
 	}
 
 	/* backfacing test */
-	if(ccl_fetch(sd, prim) != PRIM_NONE) {
-		bool backfacing = (dot(ccl_fetch(sd, Ng), ccl_fetch(sd, I)) < 0.0f);
+	if(sd->prim != PRIM_NONE) {
+		bool backfacing = (dot(sd->Ng, sd->I) < 0.0f);
 
 		if(backfacing) {
-			ccl_fetch(sd, flag) |= SD_BACKFACING;
-			ccl_fetch(sd, Ng) = -ccl_fetch(sd, Ng);
-			ccl_fetch(sd, N) = -ccl_fetch(sd, N);
+			sd->flag |= SD_BACKFACING;
+			sd->Ng = -sd->Ng;
+			sd->N = -sd->N;
 #ifdef __DPDU__
-			ccl_fetch(sd, dPdu) = -ccl_fetch(sd, dPdu);
-			ccl_fetch(sd, dPdv) = -ccl_fetch(sd, dPdv);
+			sd->dPdu = -sd->dPdu;
+			sd->dPdv = -sd->dPdv;
 #endif
 		}
 	}
 
 #ifdef __RAY_DIFFERENTIALS__
 	/* no ray differentials here yet */
-	ccl_fetch(sd, dP) = differential3_zero();
-	ccl_fetch(sd, dI) = differential3_zero();
-	ccl_fetch(sd, du) = differential_zero();
-	ccl_fetch(sd, dv) = differential_zero();
+	sd->dP = differential3_zero();
+	sd->dI = differential3_zero();
+	sd->du = differential_zero();
+	sd->dv = differential_zero();
 #endif
 }
 
@@ -378,39 +378,39 @@ ccl_device void shader_setup_from_displace(KernelGlobals *kg, ShaderData *sd,
 ccl_device_inline void shader_setup_from_background(KernelGlobals *kg, ShaderData *sd, const Ray *ray)
 {
 	/* vectors */
-	ccl_fetch(sd, P) = ray->D;
-	ccl_fetch(sd, N) = -ray->D;
-	ccl_fetch(sd, Ng) = -ray->D;
-	ccl_fetch(sd, I) = -ray->D;
-	ccl_fetch(sd, shader) = kernel_data.background.surface_shader;
-	ccl_fetch(sd, flag) = kernel_tex_fetch(__shader_flag, (ccl_fetch(sd, shader) & SHADER_MASK)*SHADER_SIZE);
-	ccl_fetch(sd, object_flag) = 0;
+	sd->P = ray->D;
+	sd->N = -ray->D;
+	sd->Ng = -ray->D;
+	sd->I = -ray->D;
+	sd->shader = kernel_data.background.surface_shader;
+	sd->flag = kernel_tex_fetch(__shader_flag, (sd->shader & SHADER_MASK)*SHADER_SIZE);
+	sd->object_flag = 0;
 #ifdef __OBJECT_MOTION__
-	ccl_fetch(sd, time) = ray->time;
+	sd->time = ray->time;
 #endif
-	ccl_fetch(sd, ray_length) = 0.0f;
+	sd->ray_length = 0.0f;
 
 #ifdef __INSTANCING__
-	ccl_fetch(sd, object) = PRIM_NONE;
+	sd->object = PRIM_NONE;
 #endif
-	ccl_fetch(sd, prim) = PRIM_NONE;
+	sd->prim = PRIM_NONE;
 #ifdef __UV__
-	ccl_fetch(sd, u) = 0.0f;
-	ccl_fetch(sd, v) = 0.0f;
+	sd->u = 0.0f;
+	sd->v = 0.0f;
 #endif
 
 #ifdef __DPDU__
 	/* dPdu/dPdv */
-	ccl_fetch(sd, dPdu) = make_float3(0.0f, 0.0f, 0.0f);
-	ccl_fetch(sd, dPdv) = make_float3(0.0f, 0.0f, 0.0f);
+	sd->dPdu = make_float3(0.0f, 0.0f, 0.0f);
+	sd->dPdv = make_float3(0.0f, 0.0f, 0.0f);
 #endif
 
 #ifdef __RAY_DIFFERENTIALS__
 	/* differentials */
-	ccl_fetch(sd, dP) = ray->dD;
-	differential_incoming(&ccl_fetch(sd, dI), ccl_fetch(sd, dP));
-	ccl_fetch(sd, du) = differential_zero();
-	ccl_fetch(sd, dv) = differential_zero();
+	sd->dP = ray->dD;
+	differential_incoming(&sd->dI, sd->dP);
+	sd->du = differential_zero();
+	sd->dv = differential_zero();
 #endif
 }
 
@@ -505,11 +505,11 @@ ccl_device_inline void _shader_bsdf_multi_eval(KernelGlobals *kg, ShaderData *sd
 {
 	/* this is the veach one-sample model with balance heuristic, some pdf
 	 * factors drop out when using balance heuristic weighting */
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
+	for(int i = 0; i < sd->num_closure; i++) {
 		if(i == skip_bsdf)
 			continue;
 
-		const ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+		const ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF(sc->type)) {
 			float bsdf_pdf = 0.0f;
@@ -535,8 +535,8 @@ ccl_device_inline void _shader_bsdf_multi_eval_branched(KernelGlobals *kg,
                                                         float light_pdf,
                                                         bool use_mis)
 {
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		const ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		const ShaderClosure *sc = &sd->closure[i];
 		if(CLOSURE_IS_BSDF(sc->type)) {
 			float bsdf_pdf = 0.0f;
 			float3 eval = bsdf_eval(kg, sd, sc, omega_in, &bsdf_pdf);
@@ -591,22 +591,22 @@ ccl_device_inline int shader_bsdf_sample(KernelGlobals *kg,
 {
 	int sampled = 0;
 
-	if(ccl_fetch(sd, num_closure) > 1) {
+	if(sd->num_closure > 1) {
 		/* pick a BSDF closure based on sample weights */
 		float sum = 0.0f;
 
-		for(sampled = 0; sampled < ccl_fetch(sd, num_closure); sampled++) {
-			const ShaderClosure *sc = ccl_fetch_array(sd, closure, sampled);
+		for(sampled = 0; sampled < sd->num_closure; sampled++) {
+			const ShaderClosure *sc = &sd->closure[sampled];
 			
 			if(CLOSURE_IS_BSDF(sc->type))
 				sum += sc->sample_weight;
 		}
 
-		float r = ccl_fetch(sd, randb_closure)*sum;
+		float r = sd->randb_closure*sum;
 		sum = 0.0f;
 
-		for(sampled = 0; sampled < ccl_fetch(sd, num_closure); sampled++) {
-			const ShaderClosure *sc = ccl_fetch_array(sd, closure, sampled);
+		for(sampled = 0; sampled < sd->num_closure; sampled++) {
+			const ShaderClosure *sc = &sd->closure[sampled];
 			
 			if(CLOSURE_IS_BSDF(sc->type)) {
 				sum += sc->sample_weight;
@@ -616,13 +616,13 @@ ccl_device_inline int shader_bsdf_sample(KernelGlobals *kg,
 			}
 		}
 
-		if(sampled == ccl_fetch(sd, num_closure)) {
+		if(sampled == sd->num_closure) {
 			*pdf = 0.0f;
 			return LABEL_NONE;
 		}
 	}
 
-	const ShaderClosure *sc = ccl_fetch_array(sd, closure, sampled);
+	const ShaderClosure *sc = &sd->closure[sampled];
 
 	int label;
 	float3 eval;
@@ -633,7 +633,7 @@ ccl_device_inline int shader_bsdf_sample(KernelGlobals *kg,
 	if(*pdf != 0.0f) {
 		bsdf_eval_init(bsdf_eval, sc->type, eval*sc->weight, kernel_data.film.use_light_pass);
 
-		if(ccl_fetch(sd, num_closure) > 1) {
+		if(sd->num_closure > 1) {
 			float sweight = sc->sample_weight;
 			_shader_bsdf_multi_eval(kg, sd, *omega_in, pdf, sampled, bsdf_eval, *pdf*sweight, sweight);
 		}
@@ -660,8 +660,8 @@ ccl_device int shader_bsdf_sample_closure(KernelGlobals *kg, ShaderData *sd,
 
 ccl_device void shader_bsdf_blur(KernelGlobals *kg, ShaderData *sd, float roughness)
 {
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF(sc->type))
 			bsdf_blur(kg, sc, roughness);
@@ -670,13 +670,13 @@ ccl_device void shader_bsdf_blur(KernelGlobals *kg, ShaderData *sd, float roughn
 
 ccl_device float3 shader_bsdf_transparency(KernelGlobals *kg, ShaderData *sd)
 {
-	if(ccl_fetch(sd, flag) & SD_HAS_ONLY_VOLUME)
+	if(sd->flag & SD_HAS_ONLY_VOLUME)
 		return make_float3(1.0f, 1.0f, 1.0f);
 
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(sc->type == CLOSURE_BSDF_TRANSPARENT_ID) // todo: make this work for osl
 			eval += sc->weight;
@@ -687,8 +687,8 @@ ccl_device float3 shader_bsdf_transparency(KernelGlobals *kg, ShaderData *sd)
 
 ccl_device void shader_bsdf_disable_transparency(KernelGlobals *kg, ShaderData *sd)
 {
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(sc->type == CLOSURE_BSDF_TRANSPARENT_ID) {
 			sc->sample_weight = 0.0f;
@@ -711,8 +711,8 @@ ccl_device float3 shader_bsdf_diffuse(KernelGlobals *kg, ShaderData *sd)
 {
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF_DIFFUSE(sc->type))
 			eval += sc->weight;
@@ -725,8 +725,8 @@ ccl_device float3 shader_bsdf_glossy(KernelGlobals *kg, ShaderData *sd)
 {
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF_GLOSSY(sc->type))
 			eval += sc->weight;
@@ -739,8 +739,8 @@ ccl_device float3 shader_bsdf_transmission(KernelGlobals *kg, ShaderData *sd)
 {
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF_TRANSMISSION(sc->type))
 			eval += sc->weight;
@@ -753,8 +753,8 @@ ccl_device float3 shader_bsdf_subsurface(KernelGlobals *kg, ShaderData *sd)
 {
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSSRDF(sc->type) || CLOSURE_IS_BSDF_BSSRDF(sc->type))
 			eval += sc->weight;
@@ -768,8 +768,8 @@ ccl_device float3 shader_bsdf_ao(KernelGlobals *kg, ShaderData *sd, float ao_fac
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 	float3 N = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSDF_DIFFUSE(sc->type)) {
 			const DiffuseBsdf *bsdf = (const DiffuseBsdf*)sc;
@@ -778,12 +778,12 @@ ccl_device float3 shader_bsdf_ao(KernelGlobals *kg, ShaderData *sd, float ao_fac
 		}
 		else if(CLOSURE_IS_AMBIENT_OCCLUSION(sc->type)) {
 			eval += sc->weight;
-			N += ccl_fetch(sd, N)*average(sc->weight);
+			N += sd->N*average(sc->weight);
 		}
 	}
 
 	if(is_zero(N))
-		N = ccl_fetch(sd, N);
+		N = sd->N;
 	else
 		N = normalize(N);
 
@@ -798,8 +798,8 @@ ccl_device float3 shader_bssrdf_sum(ShaderData *sd, float3 *N_, float *texture_b
 	float3 N = make_float3(0.0f, 0.0f, 0.0f);
 	float texture_blur = 0.0f, weight_sum = 0.0f;
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BSSRDF(sc->type)) {
 			const Bssrdf *bssrdf = (const Bssrdf*)sc;
@@ -813,7 +813,7 @@ ccl_device float3 shader_bssrdf_sum(ShaderData *sd, float3 *N_, float *texture_b
 	}
 
 	if(N_)
-		*N_ = (is_zero(N))? ccl_fetch(sd, N): normalize(N);
+		*N_ = (is_zero(N))? sd->N: normalize(N);
 
 	if(texture_blur_)
 		*texture_blur_ = texture_blur/weight_sum;
@@ -826,7 +826,7 @@ ccl_device float3 shader_bssrdf_sum(ShaderData *sd, float3 *N_, float *texture_b
 
 ccl_device float3 emissive_eval(KernelGlobals *kg, ShaderData *sd, ShaderClosure *sc)
 {
-	return emissive_simple_eval(ccl_fetch(sd, Ng), ccl_fetch(sd, I));
+	return emissive_simple_eval(sd->Ng, sd->I);
 }
 
 ccl_device float3 shader_emissive_eval(KernelGlobals *kg, ShaderData *sd)
@@ -834,8 +834,8 @@ ccl_device float3 shader_emissive_eval(KernelGlobals *kg, ShaderData *sd)
 	float3 eval;
 	eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_EMISSION(sc->type))
 			eval += emissive_eval(kg, sd, sc)*sc->weight;
@@ -850,8 +850,8 @@ ccl_device float3 shader_holdout_eval(KernelGlobals *kg, ShaderData *sd)
 {
 	float3 weight = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_HOLDOUT(sc->type))
 			weight += sc->weight;
@@ -865,9 +865,9 @@ ccl_device float3 shader_holdout_eval(KernelGlobals *kg, ShaderData *sd)
 ccl_device void shader_eval_surface(KernelGlobals *kg, ShaderData *sd, ccl_addr_space RNG *rng,
 	ccl_addr_space PathState *state, float randb, int path_flag, ShaderContext ctx)
 {
-	ccl_fetch(sd, num_closure) = 0;
-	ccl_fetch(sd, num_closure_extra) = 0;
-	ccl_fetch(sd, randb_closure) = randb;
+	sd->num_closure = 0;
+	sd->num_closure_extra = 0;
+	sd->randb_closure = randb;
 
 #ifdef __OSL__
 	if(kg->osl)
@@ -881,13 +881,13 @@ ccl_device void shader_eval_surface(KernelGlobals *kg, ShaderData *sd, ccl_addr_
 		DiffuseBsdf *bsdf = (DiffuseBsdf*)bsdf_alloc(sd,
 		                                             sizeof(DiffuseBsdf),
 		                                             make_float3(0.8f, 0.8f, 0.8f));
-		bsdf->N = ccl_fetch(sd, N);
-		ccl_fetch(sd, flag) |= bsdf_diffuse_setup(bsdf);
+		bsdf->N = sd->N;
+		sd->flag |= bsdf_diffuse_setup(bsdf);
 #endif
 	}
 
-	if(rng && (ccl_fetch(sd, flag) & SD_BSDF_NEEDS_LCG)) {
-		ccl_fetch(sd, lcg_state) = lcg_state_init_addrspace(rng, state, 0xb4bc3953);
+	if(rng && (sd->flag & SD_BSDF_NEEDS_LCG)) {
+		sd->lcg_state = lcg_state_init_addrspace(rng, state, 0xb4bc3953);
 	}
 }
 
@@ -896,9 +896,9 @@ ccl_device void shader_eval_surface(KernelGlobals *kg, ShaderData *sd, ccl_addr_
 ccl_device float3 shader_eval_background(KernelGlobals *kg, ShaderData *sd,
 	ccl_addr_space PathState *state, int path_flag, ShaderContext ctx)
 {
-	ccl_fetch(sd, num_closure) = 0;
-	ccl_fetch(sd, num_closure_extra) = 0;
-	ccl_fetch(sd, randb_closure) = 0.0f;
+	sd->num_closure = 0;
+	sd->num_closure_extra = 0;
+	sd->randb_closure = 0.0f;
 
 #ifdef __SVM__
 #ifdef __OSL__
@@ -913,8 +913,8 @@ ccl_device float3 shader_eval_background(KernelGlobals *kg, ShaderData *sd,
 
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
 
-	for(int i = 0; i < ccl_fetch(sd, num_closure); i++) {
-		const ShaderClosure *sc = ccl_fetch_array(sd, closure, i);
+	for(int i = 0; i < sd->num_closure; i++) {
+		const ShaderClosure *sc = &sd->closure[i];
 
 		if(CLOSURE_IS_BACKGROUND(sc->type))
 			eval += sc->weight;
@@ -1093,9 +1093,9 @@ ccl_device_inline void shader_eval_volume(KernelGlobals *kg,
 
 ccl_device void shader_eval_displacement(KernelGlobals *kg, ShaderData *sd, ccl_addr_space PathState *state, ShaderContext ctx)
 {
-	ccl_fetch(sd, num_closure) = 0;
-	ccl_fetch(sd, num_closure_extra) = 0;
-	ccl_fetch(sd, randb_closure) = 0.0f;
+	sd->num_closure = 0;
+	sd->num_closure_extra = 0;
+	sd->randb_closure = 0.0f;
 
 	/* this will modify sd->P */
 #ifdef __SVM__
diff --git a/intern/cycles/kernel/kernel_shadow.h b/intern/cycles/kernel/kernel_shadow.h
index 6abfa9c3873..2483c5f9ae1 100644
--- a/intern/cycles/kernel/kernel_shadow.h
+++ b/intern/cycles/kernel/kernel_shadow.h
@@ -45,7 +45,7 @@ ccl_device_forceinline bool shadow_handle_transparent_isect(
 	/* Setup shader data at surface. */
 	shader_setup_from_ray(kg, shadow_sd, isect, ray);
 	/* Attenuation from transparent surface. */
-	if(!(ccl_fetch(shadow_sd, flag) & SD_HAS_ONLY_VOLUME)) {
+	if(!(shadow_sd->flag & SD_HAS_ONLY_VOLUME)) {
 		path_state_modify_bounce(state, true);
 		shader_eval_surface(kg,
 		                    shadow_sd,
@@ -180,7 +180,7 @@ ccl_device bool shadow_blocked_transparent_all_loop(KernelGlobals *kg,
 				return true;
 			}
 			/* Move ray forward. */
-			ray->P = ccl_fetch(shadow_sd, P);
+			ray->P = shadow_sd->P;
 			if(ray->t != FLT_MAX) {
 				ray->D = normalize_len(Pend - ray->P, &ray->t);
 			}
@@ -308,7 +308,7 @@ ccl_device bool shadow_blocked_transparent_stepped_loop(
 				return true;
 			}
 			/* Move ray forward. */
-			ray->P = ray_offset(ccl_fetch(shadow_sd, P), -ccl_fetch(shadow_sd, Ng));
+			ray->P = ray_offset(shadow_sd->P, -shadow_sd->Ng);
 			if(ray->t != FLT_MAX) {
 				ray->D = normalize_len(Pend - ray->P, &ray->t);
 			}
@@ -374,7 +374,7 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg,
 #ifdef __SPLIT_KERNEL__
 	Ray private_ray = *ray_input;
 	Ray *ray = &private_ray;
-	Intersection *isect = &kernel_split_state.isect_shadow[SD_THREAD];
+	Intersection *isect = &kernel_split_state.isect_shadow[ccl_global_id(1) * ccl_global_size(0) + ccl_global_id(0)];
 #else  /* __SPLIT_KERNEL__ */
 	Ray *ray = ray_input;
 	Intersection isect_object;
diff --git a/intern/cycles/kernel/kernel_subsurface.h b/intern/cycles/kernel/kernel_subsurface.h
index 52c05b85aee..a8fa6432542 100644
--- a/intern/cycles/kernel/kernel_subsurface.h
+++ b/intern/cycles/kernel/kernel_subsurface.h
@@ -298,20 +298,20 @@ ccl_device_inline int subsurface_scatter_multi_intersect(
 	for(int hit = 0; hit < num_eval_hits; hit++) {
 		/* Quickly retrieve P and Ng without setting up ShaderData. */
 		float3 hit_P;
-		if(ccl_fetch(sd, type) & PRIMITIVE_TRIANGLE) {
+		if(sd->type & PRIMITIVE_TRIANGLE) {
 			hit_P = triangle_refine_subsurface(kg,
 			                                   sd,
 			                                   &ss_isect->hits[hit],
 			                                   ray);
 		}
 #ifdef __OBJECT_MOTION__
-		else  if(ccl_fetch(sd, type) & PRIMITIVE_MOTION_TRIANGLE) {
+		else  if(sd->type & PRIMITIVE_MOTION_TRIANGLE) {
 			float3 verts[3];
 			motion_triangle_vertices(
 			        kg,
-			        ccl_fetch(sd, object),
+			        sd->object,
 			        kernel_tex_fetch(__prim_index, ss_isect->hits[hit].prim),
-			        ccl_fetch(sd, time),
+			        sd->time,
 			        verts);
 			hit_P = motion_triangle_refine_subsurface(kg,
 			                                          sd,
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 5a80cdd6585..279cd6dedab 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -800,99 +800,77 @@ enum ShaderDataObjectFlag {
 	                   SD_OBJECT_INTERSECTS_VOLUME)
 };
 
-#ifdef __SPLIT_KERNEL__
-#  define SD_THREAD (ccl_global_id(1) * ccl_global_size(0) + ccl_global_id(0))
-#  if !defined(__SPLIT_KERNEL_SOA__)
-     /* ShaderData is stored as an Array-of-Structures */
-#    define ccl_soa_member(type, name) type soa_##name
-#    define ccl_fetch(s, t) (s[SD_THREAD].soa_##t)
-#    define ccl_fetch_array(s, t, index) (&s[SD_THREAD].soa_##t[index])
-#  else
-     /* ShaderData is stored as an Structure-of-Arrays */
-#    define SD_GLOBAL_SIZE (ccl_global_size(0) * ccl_global_size(1))
-#    define SD_FIELD_SIZE(t) sizeof(((struct ShaderData*)0)->t)
-#    define SD_OFFSETOF(t) ((char*)(&((struct ShaderData*)0)->t) - (char*)0)
-#    define ccl_soa_member(type, name) type soa_##name
-#    define ccl_fetch(s, t) (((ShaderData*)((ccl_addr_space char*)s + SD_GLOBAL_SIZE * SD_OFFSETOF(soa_##t) +  SD_FIELD_SIZE(soa_##t) * SD_THREAD - SD_OFFSETOF(soa_##t)))->soa_##t)
-#    define ccl_fetch_array(s, t, index) (&ccl_fetch(s, t)[index])
-#  endif
-#else
-#  define ccl_soa_member(type, name) type name
-#  define ccl_fetch(s, t) (s->t)
-#  define ccl_fetch_array(s, t, index) (&s->t[index])
-#endif
-
 typedef ccl_addr_space struct ShaderData {
 	/* position */
-	ccl_soa_member(float3, P);
+	float3 P;
 	/* smooth normal for shading */
-	ccl_soa_member(float3, N);
+	float3 N;
 	/* true geometric normal */
-	ccl_soa_member(float3, Ng);
+	float3 Ng;
 	/* view/incoming direction */
-	ccl_soa_member(float3, I);
+	float3 I;
 	/* shader id */
-	ccl_soa_member(int, shader);
+	int shader;
 	/* booleans describing shader, see ShaderDataFlag */
-	ccl_soa_member(int, flag);
+	int flag;
 	/* booleans describing object of the shader, see ShaderDataObjectFlag */
-	ccl_soa_member(int, object_flag);
+	int object_flag;
 
 	/* primitive id if there is one, ~0 otherwise */
-	ccl_soa_member(int, prim);
+	int prim;
 
 	/* combined type and curve segment for hair */
-	ccl_soa_member(int, type);
+	int type;
 
 	/* parametric coordinates
 	 * - barycentric weights for triangles */
-	ccl_soa_member(float, u);
-	ccl_soa_member(float, v);
+	float u;
+	float v;
 	/* object id if there is one, ~0 otherwise */
-	ccl_soa_member(int, object);
+	int object;
 
 	/* motion blur sample time */
-	ccl_soa_member(float, time);
+	float time;
 
 	/* length of the ray being shaded */
-	ccl_soa_member(float, ray_length);
+	float ray_length;
 
 #ifdef __RAY_DIFFERENTIALS__
 	/* differential of P. these are orthogonal to Ng, not N */
-	ccl_soa_member(differential3, dP);
+	differential3 dP;
 	/* differential of I */
-	ccl_soa_member(differential3, dI);
+	differential3 dI;
 	/* differential of u, v */
-	ccl_soa_member(differential, du);
-	ccl_soa_member(differential, dv);
+	differential du;
+	differential dv;
 #endif
 #ifdef __DPDU__
 	/* differential of P w.r.t. parametric coordinates. note that dPdu is
 	 * not readily suitable as a tangent for shading on triangles. */
-	ccl_soa_member(float3, dPdu);
-	ccl_soa_member(float3, dPdv);
+	float3 dPdu;
+	float3 dPdv;
 #endif
 
 #ifdef __OBJECT_MOTION__
 	/* object <-> world space transformations, cached to avoid
 	 * re-interpolating them constantly for shading */
-	ccl_soa_member(Transform, ob_tfm);
-	ccl_soa_member(Transform, ob_itfm);
+	Transform ob_tfm;
+	Transform ob_itfm;
 #endif
 
 	/* Closure data, we store a fixed array of closures */
-	ccl_soa_member(struct ShaderClosure, closure[MAX_CLOSURE]);
-	ccl_soa_member(int, num_closure);
-	ccl_soa_member(int, num_closure_extra);
-	ccl_soa_member(float, randb_closure);
-	ccl_soa_member(float3, svm_closure_weight);
+	struct ShaderClosure closure[MAX_CLOSURE];
+	int num_closure;
+	int num_closure_extra;
+	float randb_closure;
+	float3 svm_closure_weight;
 
 	/* LCG state for closures that require additional random numbers. */
-	ccl_soa_member(uint, lcg_state);
+	uint lcg_state;
 
 	/* ray start position, only set for backgrounds */
-	ccl_soa_member(float3, ray_P);
-	ccl_soa_member(differential3, ray_dP);
+	float3 ray_P;
+	differential3 ray_dP;
 
 #ifdef __OSL__
 	struct KernelGlobals *osl_globals;
diff --git a/intern/cycles/kernel/osl/osl_bssrdf.cpp b/intern/cycles/kernel/osl/osl_bssrdf.cpp
index 3614717e28c..d3a69d39597 100644
--- a/intern/cycles/kernel/osl/osl_bssrdf.cpp
+++ b/intern/cycles/kernel/osl/osl_bssrdf.cpp
@@ -78,7 +78,7 @@ public:
 				bssrdf->albedo = albedo.x;
 				bssrdf->sharpness = sharpness;
 				bssrdf->N = params.N;
-				ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+				sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 			}
 
 			bssrdf = bssrdf_alloc(sd, make_float3(0.0f, weight.y, 0.0f));
@@ -89,7 +89,7 @@ public:
 				bssrdf->albedo = albedo.y;
 				bssrdf->sharpness = sharpness;
 				bssrdf->N = params.N;
-				ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+				sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 			}
 
 			bssrdf = bssrdf_alloc(sd, make_float3(0.0f, 0.0f, weight.z));
@@ -100,7 +100,7 @@ public:
 				bssrdf->albedo = albedo.z;
 				bssrdf->sharpness = sharpness;
 				bssrdf->N = params.N;
-				ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+				sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 			}
 		}
 	}
diff --git a/intern/cycles/kernel/split/kernel_background_buffer_update.h b/intern/cycles/kernel/split/kernel_background_buffer_update.h
index 6d48856eb44..07e5522c830 100644
--- a/intern/cycles/kernel/split/kernel_background_buffer_update.h
+++ b/intern/cycles/kernel/split/kernel_background_buffer_update.h
@@ -156,7 +156,7 @@ ccl_device void kernel_background_buffer_update(KernelGlobals *kg)
 		if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
 #ifdef __BACKGROUND__
 			/* sample background shader */
-			float3 L_background = indirect_background(kg, kernel_split_state.sd_DL_shadow, state, ray);
+			float3 L_background = indirect_background(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, ray);
 			path_radiance_accum_background(L, (*throughput), L_background, state->bounce);
 #endif
 			ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
@@ -210,7 +210,7 @@ ccl_device void kernel_background_buffer_update(KernelGlobals *kg)
 				*throughput = make_float3(1.0f, 1.0f, 1.0f);
 				*L_transparent = 0.0f;
 				path_radiance_init(L, kernel_data.film.use_light_pass);
-				path_state_init(kg, kernel_split_state.sd_DL_shadow, state, rng, sample, ray);
+				path_state_init(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, rng, sample, ray);
 #ifdef __KERNEL_DEBUG__
 				debug_data_init(debug_data);
 #endif
diff --git a/intern/cycles/kernel/split/kernel_data_init.h b/intern/cycles/kernel/split/kernel_data_init.h
index eeeb5294fdc..3251c091468 100644
--- a/intern/cycles/kernel/split/kernel_data_init.h
+++ b/intern/cycles/kernel/split/kernel_data_init.h
@@ -190,7 +190,7 @@ ccl_device void kernel_data_init(
 		kernel_split_state.L_transparent[ray_index] = 0.0f;
 		path_radiance_init(&kernel_split_state.path_radiance[ray_index], kernel_data.film.use_light_pass);
 		path_state_init(kg,
-		                kernel_split_state.sd_DL_shadow,
+		                &kernel_split_state.sd_DL_shadow[ray_index],
 		                &kernel_split_state.path_state[ray_index],
 		                &kernel_split_state.rng[ray_index],
 		                my_sample,
diff --git a/intern/cycles/kernel/split/kernel_direct_lighting.h b/intern/cycles/kernel/split/kernel_direct_lighting.h
index 43b943710a1..5163b8edc04 100644
--- a/intern/cycles/kernel/split/kernel_direct_lighting.h
+++ b/intern/cycles/kernel/split/kernel_direct_lighting.h
@@ -82,12 +82,12 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg)
 
 	if(IS_STATE(kernel_split_state.ray_state, ray_index, RAY_ACTIVE)) {
 		ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
-		ShaderData *sd = kernel_split_state.sd;
+		ShaderData *sd = &kernel_split_state.sd[ray_index];
 
 		/* direct lighting */
 #ifdef __EMISSION__
 		if((kernel_data.integrator.use_direct_light &&
-		    (ccl_fetch(sd, flag) & SD_BSDF_HAS_EVAL)))
+		    (sd->flag & SD_BSDF_HAS_EVAL)))
 		{
 			/* Sample illumination from lights to find path contribution. */
 			ccl_global RNG* rng = &kernel_split_state.rng[ray_index];
@@ -99,19 +99,19 @@ ccl_device void kernel_direct_lighting(KernelGlobals *kg)
 			LightSample ls;
 			if(light_sample(kg,
 			                light_t, light_u, light_v,
-			                ccl_fetch(sd, time),
-			                ccl_fetch(sd, P),
+			                sd->time,
+			                sd->P,
 			                state->bounce,
 			                &ls)) {
 
 				Ray light_ray;
 #ifdef __OBJECT_MOTION__
-				light_ray.time = ccl_fetch(sd, time);
+				light_ray.time = sd->time;
 #endif
 
 				BsdfEval L_light;
 				bool is_lamp;
-				if(direct_emission(kg, sd, kernel_split_state.sd_DL_shadow, &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
+				if(direct_emission(kg, sd, &kernel_split_state.sd_DL_shadow[ray_index], &ls, state, &light_ray, &L_light, &is_lamp, terminate)) {
 					/* Write intermediate data to global memory to access from
 					 * the next kernel.
 					 */
diff --git a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
index 3a7f1629e66..ee9c4280b22 100644
--- a/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
+++ b/intern/cycles/kernel/split/kernel_holdout_emission_blurring_pathtermination_ao.h
@@ -122,7 +122,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 	float3 throughput;
 
 	ccl_global char *ray_state = kernel_split_state.ray_state;
-	ShaderData *sd = kernel_split_state.sd;
+	ShaderData *sd = &kernel_split_state.sd[ray_index];
 	ccl_global float *per_sample_output_buffers = kernel_split_state.per_sample_output_buffers;
 
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
@@ -145,13 +145,13 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 
 		/* holdout */
 #ifdef __HOLDOUT__
-		if(((ccl_fetch(sd, flag) & SD_HOLDOUT) ||
-		    (ccl_fetch(sd, object_flag) & SD_OBJECT_HOLDOUT_MASK)) &&
+		if(((sd->flag & SD_HOLDOUT) ||
+		    (sd->object_flag & SD_OBJECT_HOLDOUT_MASK)) &&
 		   (state->flag & PATH_RAY_CAMERA))
 		{
 			if(kernel_data.background.transparent) {
 				float3 holdout_weight;
-				if(ccl_fetch(sd, object_flag) & SD_OBJECT_HOLDOUT_MASK) {
+				if(sd->object_flag & SD_OBJECT_HOLDOUT_MASK) {
 					holdout_weight = make_float3(1.0f, 1.0f, 1.0f);
 				}
 				else {
@@ -160,7 +160,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 				/* any throughput is ok, should all be identical here */
 				kernel_split_state.L_transparent[ray_index] += average(holdout_weight*throughput);
 			}
-			if(ccl_fetch(sd, object_flag) & SD_OBJECT_HOLDOUT_MASK) {
+			if(sd->object_flag & SD_OBJECT_HOLDOUT_MASK) {
 				ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
 				enqueue_flag = 1;
 			}
@@ -191,7 +191,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 
 #ifdef __EMISSION__
 		/* emission */
-		if(ccl_fetch(sd, flag) & SD_EMISSION) {
+		if(sd->flag & SD_EMISSION) {
 			/* TODO(sergey): is isect.t wrong here for transparent surfaces? */
 			float3 emission = indirect_primitive_emission(
 			        kg,
@@ -232,7 +232,7 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 	if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
 		/* ambient occlusion */
 		if(kernel_data.integrator.use_ambient_occlusion ||
-		   (ccl_fetch(sd, flag) & SD_AO))
+		   (sd->flag & SD_AO))
 		{
 			/* todo: solve correlation */
 			float bsdf_u, bsdf_v;
@@ -247,15 +247,15 @@ ccl_device void kernel_holdout_emission_blurring_pathtermination_ao(KernelGlobal
 			float ao_pdf;
 			sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
-			if(dot(ccl_fetch(sd, Ng), ao_D) > 0.0f && ao_pdf != 0.0f) {
+			if(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f) {
 				Ray _ray;
-				_ray.P = ray_offset(ccl_fetch(sd, P), ccl_fetch(sd, Ng));
+				_ray.P = ray_offset(sd->P, sd->Ng);
 				_ray.D = ao_D;
 				_ray.t = kernel_data.background.ao_distance;
 #ifdef __OBJECT_MOTION__
-				_ray.time = ccl_fetch(sd, time);
+				_ray.time = sd->time;
 #endif
-				_ray.dP = ccl_fetch(sd, dP);
+				_ray.dP = sd->dP;
 				_ray.dD = differential3_zero();
 				kernel_split_state.ao_light_ray[ray_index] = _ray;
 
diff --git a/intern/cycles/kernel/split/kernel_lamp_emission.h b/intern/cycles/kernel/split/kernel_lamp_emission.h
index a9820af3001..261625da31d 100644
--- a/intern/cycles/kernel/split/kernel_lamp_emission.h
+++ b/intern/cycles/kernel/split/kernel_lamp_emission.h
@@ -85,7 +85,7 @@ ccl_device void kernel_lamp_emission(KernelGlobals *kg)
 			/* intersect with lamp */
 			float3 emission;
 
-			if(indirect_lamp_emission(kg, kernel_split_state.sd_DL_shadow, state, &light_ray, &emission)) {
+			if(indirect_lamp_emission(kg, &kernel_split_state.sd_DL_shadow[ray_index], state, &light_ray, &emission)) {
 				path_radiance_accum_emission(L, throughput, emission, state->bounce);
 			}
 		}
diff --git a/intern/cycles/kernel/split/kernel_next_iteration_setup.h b/intern/cycles/kernel/split/kernel_next_iteration_setup.h
index 0ff8286e59b..a6f26278116 100644
--- a/intern/cycles/kernel/split/kernel_next_iteration_setup.h
+++ b/intern/cycles/kernel/split/kernel_next_iteration_setup.h
@@ -161,7 +161,7 @@ ccl_device void kernel_next_iteration_setup(KernelGlobals *kg)
 		L = &kernel_split_state.path_radiance[ray_index];
 
 		/* Compute direct lighting and next bounce. */
-		if(!kernel_path_surface_bounce(kg, rng, kernel_split_state.sd, throughput, state, L, ray)) {
+		if(!kernel_path_surface_bounce(kg, rng, &kernel_split_state.sd[ray_index], throughput, state, L, ray)) {
 			ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
 			enqueue_flag = 1;
 		}
diff --git a/intern/cycles/kernel/split/kernel_shader_eval.h b/intern/cycles/kernel/split/kernel_shader_eval.h
index 8328f60af39..35ee19ddf1b 100644
--- a/intern/cycles/kernel/split/kernel_shader_eval.h
+++ b/intern/cycles/kernel/split/kernel_shader_eval.h
@@ -82,11 +82,11 @@ ccl_device void kernel_shader_eval(KernelGlobals *kg)
 		Ray ray = kernel_split_state.ray[ray_index];
 
 		shader_setup_from_ray(kg,
-		                      kernel_split_state.sd,
+		                      &kernel_split_state.sd[ray_index],
 		                      isect,
 		                      &ray);
 		float rbsdf = path_state_rng_1D_for_decision(kg, rng, state, PRNG_BSDF);
-		shader_eval_surface(kg, kernel_split_state.sd, rng, state, rbsdf, state->flag, SHADER_CONTEXT_MAIN);
+		shader_eval_surface(kg, &kernel_split_state.sd[ray_index], rng, state, rbsdf, state->flag, SHADER_CONTEXT_MAIN);
 	}
 }
 
diff --git a/intern/cycles/kernel/split/kernel_shadow_blocked.h b/intern/cycles/kernel/split/kernel_shadow_blocked.h
index f862c37ea64..d532c7cf55b 100644
--- a/intern/cycles/kernel/split/kernel_shadow_blocked.h
+++ b/intern/cycles/kernel/split/kernel_shadow_blocked.h
@@ -94,7 +94,7 @@ ccl_device void kernel_shadow_blocked(KernelGlobals *kg)
 
 		float3 shadow;
 		update_path_radiance = !(shadow_blocked(kg,
-		                                        kernel_split_state.sd_DL_shadow,
+		                                        &kernel_split_state.sd_DL_shadow[thread_index],
 		                                        state,
 		                                        light_ray_global,
 		                                        &shadow));
diff --git a/intern/cycles/kernel/svm/svm.h b/intern/cycles/kernel/svm/svm.h
index 88ec7fe6fcc..57ec9f94a3d 100644
--- a/intern/cycles/kernel/svm/svm.h
+++ b/intern/cycles/kernel/svm/svm.h
@@ -192,7 +192,7 @@ CCL_NAMESPACE_BEGIN
 ccl_device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ccl_addr_space PathState *state, ShaderType type, int path_flag)
 {
 	float stack[SVM_STACK_SIZE];
-	int offset = ccl_fetch(sd, shader) & SHADER_MASK;
+	int offset = sd->shader & SHADER_MASK;
 
 	while(1) {
 		uint4 node = read_node(kg, &offset);
diff --git a/intern/cycles/kernel/svm/svm_attribute.h b/intern/cycles/kernel/svm/svm_attribute.h
index 0e55c99ae97..229a3f20421 100644
--- a/intern/cycles/kernel/svm/svm_attribute.h
+++ b/intern/cycles/kernel/svm/svm_attribute.h
@@ -27,7 +27,7 @@ ccl_device AttributeDescriptor svm_node_attr_init(KernelGlobals *kg, ShaderData
 
 	AttributeDescriptor desc;
 
-	if(ccl_fetch(sd, object) != OBJECT_NONE) {
+	if(sd->object != OBJECT_NONE) {
 		desc = find_attribute(kg, sd, node.y);
 		if(desc.offset == ATTR_STD_NOT_FOUND) {
 			desc = attribute_not_found();
diff --git a/intern/cycles/kernel/svm/svm_bump.h b/intern/cycles/kernel/svm/svm_bump.h
index 04a8c7b64e5..610d9af9e1f 100644
--- a/intern/cycles/kernel/svm/svm_bump.h
+++ b/intern/cycles/kernel/svm/svm_bump.h
@@ -21,9 +21,9 @@ CCL_NAMESPACE_BEGIN
 ccl_device void svm_node_enter_bump_eval(KernelGlobals *kg, ShaderData *sd, float *stack, uint offset)
 {
 	/* save state */
-	stack_store_float3(stack, offset+0, ccl_fetch(sd, P));
-	stack_store_float3(stack, offset+3, ccl_fetch(sd, dP).dx);
-	stack_store_float3(stack, offset+6, ccl_fetch(sd, dP).dy);
+	stack_store_float3(stack, offset+0, sd->P);
+	stack_store_float3(stack, offset+3, sd->dP.dx);
+	stack_store_float3(stack, offset+6, sd->dP.dy);
 
 	/* set state as if undisplaced */
 	const AttributeDescriptor desc = find_attribute(kg, sd, ATTR_STD_POSITION_UNDISPLACED);
@@ -36,18 +36,18 @@ ccl_device void svm_node_enter_bump_eval(KernelGlobals *kg, ShaderData *sd, floa
 		object_dir_transform(kg, sd, &dPdx);
 		object_dir_transform(kg, sd, &dPdy);
 
-		ccl_fetch(sd, P) = P;
-		ccl_fetch(sd, dP).dx = dPdx;
-		ccl_fetch(sd, dP).dy = dPdy;
+		sd->P = P;
+		sd->dP.dx = dPdx;
+		sd->dP.dy = dPdy;
 	}
 }
 
 ccl_device void svm_node_leave_bump_eval(KernelGlobals *kg, ShaderData *sd, float *stack, uint offset)
 {
 	/* restore state */
-	ccl_fetch(sd, P) = stack_load_float3(stack, offset+0);
-	ccl_fetch(sd, dP).dx = stack_load_float3(stack, offset+3);
-	ccl_fetch(sd, dP).dy = stack_load_float3(stack, offset+6);
+	sd->P = stack_load_float3(stack, offset+0);
+	sd->dP.dx = stack_load_float3(stack, offset+3);
+	sd->dP.dy = stack_load_float3(stack, offset+6);
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_camera.h b/intern/cycles/kernel/svm/svm_camera.h
index 00678a49d70..90249dfd978 100644
--- a/intern/cycles/kernel/svm/svm_camera.h
+++ b/intern/cycles/kernel/svm/svm_camera.h
@@ -23,7 +23,7 @@ ccl_device void svm_node_camera(KernelGlobals *kg, ShaderData *sd, float *stack,
 	float3 vector;
 
 	Transform tfm = kernel_data.cam.worldtocamera;
-	vector = transform_point(&tfm, ccl_fetch(sd, P));
+	vector = transform_point(&tfm, sd->P);
 	zdepth = vector.z;
 	distance = len(vector);
 
diff --git a/intern/cycles/kernel/svm/svm_closure.h b/intern/cycles/kernel/svm/svm_closure.h
index 017d697f9f8..1885e1af851 100644
--- a/intern/cycles/kernel/svm/svm_closure.h
+++ b/intern/cycles/kernel/svm/svm_closure.h
@@ -25,13 +25,13 @@ ccl_device void svm_node_glass_setup(ShaderData *sd, MicrofacetBsdf *bsdf, int t
 			bsdf->alpha_y = 0.0f;
 			bsdf->alpha_x = 0.0f;
 			bsdf->ior = eta;
-			ccl_fetch(sd, flag) |= bsdf_refraction_setup(bsdf);
+			sd->flag |= bsdf_refraction_setup(bsdf);
 		}
 		else {
 			bsdf->alpha_y = 0.0f;
 			bsdf->alpha_x = 0.0f;
 			bsdf->ior = 0.0f;
-			ccl_fetch(sd, flag) |= bsdf_reflection_setup(bsdf);
+			sd->flag |= bsdf_reflection_setup(bsdf);
 		}
 	}
 	else if(type == CLOSURE_BSDF_MICROFACET_BECKMANN_GLASS_ID) {
@@ -40,9 +40,9 @@ ccl_device void svm_node_glass_setup(ShaderData *sd, MicrofacetBsdf *bsdf, int t
 		bsdf->ior = eta;
 
 		if(refract)
-			ccl_fetch(sd, flag) |= bsdf_microfacet_beckmann_refraction_setup(bsdf);
+			sd->flag |= bsdf_microfacet_beckmann_refraction_setup(bsdf);
 		else
-			ccl_fetch(sd, flag) |= bsdf_microfacet_beckmann_setup(bsdf);
+			sd->flag |= bsdf_microfacet_beckmann_setup(bsdf);
 	}
 	else {
 		bsdf->alpha_x = roughness;
@@ -50,9 +50,9 @@ ccl_device void svm_node_glass_setup(ShaderData *sd, MicrofacetBsdf *bsdf, int t
 		bsdf->ior = eta;
 
 		if(refract)
-			ccl_fetch(sd, flag) |= bsdf_microfacet_ggx_refraction_setup(bsdf);
+			sd->flag |= bsdf_microfacet_ggx_refraction_setup(bsdf);
 		else
-			ccl_fetch(sd, flag) |= bsdf_microfacet_ggx_setup(bsdf);
+			sd->flag |= bsdf_microfacet_ggx_setup(bsdf);
 	}
 }
 
@@ -70,14 +70,14 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 	if(mix_weight == 0.0f)
 		return;
 
-	float3 N = stack_valid(data_node.x)? stack_load_float3(stack, data_node.x): ccl_fetch(sd, N);
+	float3 N = stack_valid(data_node.x)? stack_load_float3(stack, data_node.x): sd->N;
 
 	float param1 = (stack_valid(param1_offset))? stack_load_float(stack, param1_offset): __uint_as_float(node.z);
 	float param2 = (stack_valid(param2_offset))? stack_load_float(stack, param2_offset): __uint_as_float(node.w);
 
 	switch(type) {
 		case CLOSURE_BSDF_DIFFUSE_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			OrenNayarBsdf *bsdf = (OrenNayarBsdf*)bsdf_alloc(sd, sizeof(OrenNayarBsdf), weight);
 
 			if(bsdf) {
@@ -86,31 +86,31 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				float roughness = param1;
 
 				if(roughness == 0.0f) {
-					ccl_fetch(sd, flag) |= bsdf_diffuse_setup((DiffuseBsdf*)bsdf);
+					sd->flag |= bsdf_diffuse_setup((DiffuseBsdf*)bsdf);
 				}
 				else {
 					bsdf->roughness = roughness;
-					ccl_fetch(sd, flag) |= bsdf_oren_nayar_setup(bsdf);
+					sd->flag |= bsdf_oren_nayar_setup(bsdf);
 				}
 			}
 			break;
 		}
 		case CLOSURE_BSDF_TRANSLUCENT_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			DiffuseBsdf *bsdf = (DiffuseBsdf*)bsdf_alloc(sd, sizeof(DiffuseBsdf), weight);
 
 			if(bsdf) {
 				bsdf->N = N;
-				ccl_fetch(sd, flag) |= bsdf_translucent_setup(bsdf);
+				sd->flag |= bsdf_translucent_setup(bsdf);
 			}
 			break;
 		}
 		case CLOSURE_BSDF_TRANSPARENT_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			ShaderClosure *bsdf = bsdf_alloc(sd, sizeof(ShaderClosure), weight);
 
 			if(bsdf) {
-				ccl_fetch(sd, flag) |= bsdf_transparent_setup(bsdf);
+				sd->flag |= bsdf_transparent_setup(bsdf);
 			}
 			break;
 		}
@@ -123,7 +123,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 			if(!kernel_data.integrator.caustics_reflective && (path_flag & PATH_RAY_DIFFUSE))
 				break;
 #endif
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			MicrofacetBsdf *bsdf = (MicrofacetBsdf*)bsdf_alloc(sd, sizeof(MicrofacetBsdf), weight);
 
 			if(bsdf) {
@@ -135,21 +135,21 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 
 				/* setup bsdf */
 				if(type == CLOSURE_BSDF_REFLECTION_ID)
-					ccl_fetch(sd, flag) |= bsdf_reflection_setup(bsdf);
+					sd->flag |= bsdf_reflection_setup(bsdf);
 				else if(type == CLOSURE_BSDF_MICROFACET_BECKMANN_ID)
-					ccl_fetch(sd, flag) |= bsdf_microfacet_beckmann_setup(bsdf);
+					sd->flag |= bsdf_microfacet_beckmann_setup(bsdf);
 				else if(type == CLOSURE_BSDF_MICROFACET_GGX_ID)
-					ccl_fetch(sd, flag) |= bsdf_microfacet_ggx_setup(bsdf);
+					sd->flag |= bsdf_microfacet_ggx_setup(bsdf);
 				else if(type == CLOSURE_BSDF_MICROFACET_MULTI_GGX_ID) {
 					kernel_assert(stack_valid(data_node.z));
 					bsdf->extra = (MicrofacetExtra*)closure_alloc_extra(sd, sizeof(MicrofacetExtra));
 					if(bsdf->extra) {
 						bsdf->extra->color = stack_load_float3(stack, data_node.z);
-						ccl_fetch(sd, flag) |= bsdf_microfacet_multi_ggx_setup(bsdf);
+						sd->flag |= bsdf_microfacet_multi_ggx_setup(bsdf);
 					}
 				}
 				else
-					ccl_fetch(sd, flag) |= bsdf_ashikhmin_shirley_setup(bsdf);
+					sd->flag |= bsdf_ashikhmin_shirley_setup(bsdf);
 			}
 
 			break;
@@ -161,7 +161,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 			if(!kernel_data.integrator.caustics_refractive && (path_flag & PATH_RAY_DIFFUSE))
 				break;
 #endif
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			MicrofacetBsdf *bsdf = (MicrofacetBsdf*)bsdf_alloc(sd, sizeof(MicrofacetBsdf), weight);
 
 			if(bsdf) {
@@ -169,7 +169,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				bsdf->extra = NULL;
 
 				float eta = fmaxf(param2, 1e-5f);
-				eta = (ccl_fetch(sd, flag) & SD_BACKFACING)? 1.0f/eta: eta;
+				eta = (sd->flag & SD_BACKFACING)? 1.0f/eta: eta;
 
 				/* setup bsdf */
 				if(type == CLOSURE_BSDF_REFRACTION_ID) {
@@ -177,7 +177,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					bsdf->alpha_y = 0.0f;
 					bsdf->ior = eta;
 
-					ccl_fetch(sd, flag) |= bsdf_refraction_setup(bsdf);
+					sd->flag |= bsdf_refraction_setup(bsdf);
 				}
 				else {
 					bsdf->alpha_x = param1;
@@ -185,9 +185,9 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					bsdf->ior = eta;
 
 					if(type == CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID)
-						ccl_fetch(sd, flag) |= bsdf_microfacet_beckmann_refraction_setup(bsdf);
+						sd->flag |= bsdf_microfacet_beckmann_refraction_setup(bsdf);
 					else
-						ccl_fetch(sd, flag) |= bsdf_microfacet_ggx_refraction_setup(bsdf);
+						sd->flag |= bsdf_microfacet_ggx_refraction_setup(bsdf);
 				}
 			}
 
@@ -203,14 +203,14 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				break;
 			}
 #endif
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 
 			/* index of refraction */
 			float eta = fmaxf(param2, 1e-5f);
-			eta = (ccl_fetch(sd, flag) & SD_BACKFACING)? 1.0f/eta: eta;
+			eta = (sd->flag & SD_BACKFACING)? 1.0f/eta: eta;
 
 			/* fresnel */
-			float cosNO = dot(N, ccl_fetch(sd, I));
+			float cosNO = dot(N, sd->I);
 			float fresnel = fresnel_dielectric_cos(cosNO, eta);
 			float roughness = param1;
 
@@ -249,7 +249,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 			if(!kernel_data.integrator.caustics_reflective && !kernel_data.integrator.caustics_refractive && (path_flag & PATH_RAY_DIFFUSE))
 				break;
 #endif
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			MicrofacetBsdf *bsdf = (MicrofacetBsdf*)bsdf_alloc(sd, sizeof(MicrofacetBsdf), weight);
 			MicrofacetExtra *extra = (MicrofacetExtra*)closure_alloc_extra(sd, sizeof(MicrofacetExtra));
 
@@ -261,13 +261,13 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				bsdf->alpha_x = param1;
 				bsdf->alpha_y = param1;
 				float eta = fmaxf(param2, 1e-5f);
-				bsdf->ior = (ccl_fetch(sd, flag) & SD_BACKFACING)? 1.0f/eta: eta;
+				bsdf->ior = (sd->flag & SD_BACKFACING)? 1.0f/eta: eta;
 
 				kernel_assert(stack_valid(data_node.z));
 				bsdf->extra->color = stack_load_float3(stack, data_node.z);
 
 				/* setup bsdf */
-				ccl_fetch(sd, flag) |= bsdf_microfacet_multi_ggx_glass_setup(bsdf);
+				sd->flag |= bsdf_microfacet_multi_ggx_glass_setup(bsdf);
 			}
 
 			break;
@@ -280,7 +280,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 			if(!kernel_data.integrator.caustics_reflective && (path_flag & PATH_RAY_DIFFUSE))
 				break;
 #endif
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			MicrofacetBsdf *bsdf = (MicrofacetBsdf*)bsdf_alloc(sd, sizeof(MicrofacetBsdf), weight);
 
 			if(bsdf) {
@@ -310,33 +310,33 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				bsdf->ior = 0.0f;
 
 				if(type == CLOSURE_BSDF_MICROFACET_BECKMANN_ANISO_ID) {
-					ccl_fetch(sd, flag) |= bsdf_microfacet_beckmann_aniso_setup(bsdf);
+					sd->flag |= bsdf_microfacet_beckmann_aniso_setup(bsdf);
 				}
 				else if(type == CLOSURE_BSDF_MICROFACET_GGX_ANISO_ID) {
-					ccl_fetch(sd, flag) |= bsdf_microfacet_ggx_aniso_setup(bsdf);
+					sd->flag |= bsdf_microfacet_ggx_aniso_setup(bsdf);
 				}
 				else if(type == CLOSURE_BSDF_MICROFACET_MULTI_GGX_ANISO_ID) {
 					kernel_assert(stack_valid(data_node.w));
 					bsdf->extra = (MicrofacetExtra*)closure_alloc_extra(sd, sizeof(MicrofacetExtra));
 					if(bsdf->extra) {
 						bsdf->extra->color = stack_load_float3(stack, data_node.w);
-						ccl_fetch(sd, flag) |= bsdf_microfacet_multi_ggx_aniso_setup(bsdf);
+						sd->flag |= bsdf_microfacet_multi_ggx_aniso_setup(bsdf);
 					}
 				}
 				else
-					ccl_fetch(sd, flag) |= bsdf_ashikhmin_shirley_aniso_setup(bsdf);
+					sd->flag |= bsdf_ashikhmin_shirley_aniso_setup(bsdf);
 			}
 			break;
 		}
 		case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			VelvetBsdf *bsdf = (VelvetBsdf*)bsdf_alloc(sd, sizeof(VelvetBsdf), weight);
 
 			if(bsdf) {
 				bsdf->N = N;
 
 				bsdf->sigma = saturate(param1);
-				ccl_fetch(sd, flag) |= bsdf_ashikhmin_velvet_setup(bsdf);
+				sd->flag |= bsdf_ashikhmin_velvet_setup(bsdf);
 			}
 			break;
 		}
@@ -346,7 +346,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				break;
 #endif
 		case CLOSURE_BSDF_DIFFUSE_TOON_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			ToonBsdf *bsdf = (ToonBsdf*)bsdf_alloc(sd, sizeof(ToonBsdf), weight);
 
 			if(bsdf) {
@@ -355,18 +355,18 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 				bsdf->smooth = param2;
 				
 				if(type == CLOSURE_BSDF_DIFFUSE_TOON_ID)
-					ccl_fetch(sd, flag) |= bsdf_diffuse_toon_setup(bsdf);
+					sd->flag |= bsdf_diffuse_toon_setup(bsdf);
 				else
-					ccl_fetch(sd, flag) |= bsdf_glossy_toon_setup(bsdf);
+					sd->flag |= bsdf_glossy_toon_setup(bsdf);
 			}
 			break;
 		}
 #ifdef __HAIR__
 		case CLOSURE_BSDF_HAIR_REFLECTION_ID:
 		case CLOSURE_BSDF_HAIR_TRANSMISSION_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			
-			if(ccl_fetch(sd, flag) & SD_BACKFACING && ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) {
+			if(sd->flag & SD_BACKFACING && sd->type & PRIMITIVE_ALL_CURVE) {
 				ShaderClosure *bsdf = bsdf_alloc(sd, sizeof(ShaderClosure), weight);
 
 				if(bsdf) {
@@ -376,7 +376,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					 * better figure out a way to skip backfaces from rays
 					 * spawned by transmission from the front */
 					bsdf->weight = make_float3(1.0f, 1.0f, 1.0f);
-					ccl_fetch(sd, flag) |= bsdf_transparent_setup(bsdf);
+					sd->flag |= bsdf_transparent_setup(bsdf);
 				}
 			}
 			else {
@@ -390,18 +390,18 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					if(stack_valid(data_node.y)) {
 						bsdf->T = normalize(stack_load_float3(stack, data_node.y));
 					}
-					else if(!(ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE)) {
-						bsdf->T = normalize(ccl_fetch(sd, dPdv));
+					else if(!(sd->type & PRIMITIVE_ALL_CURVE)) {
+						bsdf->T = normalize(sd->dPdv);
 						bsdf->offset = 0.0f;
 					}
 					else
-						bsdf->T = normalize(ccl_fetch(sd, dPdu));
+						bsdf->T = normalize(sd->dPdu);
 
 					if(type == CLOSURE_BSDF_HAIR_REFLECTION_ID) {
-						ccl_fetch(sd, flag) |= bsdf_hair_reflection_setup(bsdf);
+						sd->flag |= bsdf_hair_reflection_setup(bsdf);
 					}
 					else {
-						ccl_fetch(sd, flag) |= bsdf_hair_transmission_setup(bsdf);
+						sd->flag |= bsdf_hair_transmission_setup(bsdf);
 					}
 				}
 			}
@@ -414,8 +414,8 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 		case CLOSURE_BSSRDF_CUBIC_ID:
 		case CLOSURE_BSSRDF_GAUSSIAN_ID:
 		case CLOSURE_BSSRDF_BURLEY_ID: {
-			float3 albedo = ccl_fetch(sd, svm_closure_weight);
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight;
+			float3 albedo = sd->svm_closure_weight;
+			float3 weight = sd->svm_closure_weight * mix_weight;
 			float sample_weight = fabsf(average(weight));
 			
 			/* disable in case of diffuse ancestor, can't see it well then and
@@ -441,7 +441,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					bssrdf->albedo = albedo.x;
 					bssrdf->sharpness = sharpness;
 					bssrdf->N = N;
-					ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+					sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 				}
 
 				bssrdf = bssrdf_alloc(sd, make_float3(0.0f, weight.y, 0.0f));
@@ -452,7 +452,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					bssrdf->albedo = albedo.y;
 					bssrdf->sharpness = sharpness;
 					bssrdf->N = N;
-					ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+					sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 				}
 
 				bssrdf = bssrdf_alloc(sd, make_float3(0.0f, 0.0f, weight.z));
@@ -463,7 +463,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 					bssrdf->albedo = albedo.z;
 					bssrdf->sharpness = sharpness;
 					bssrdf->N = N;
-					ccl_fetch(sd, flag) |= bssrdf_setup(bssrdf, (ClosureType)type);
+					sd->flag |= bssrdf_setup(bssrdf, (ClosureType)type);
 				}
 			}
 
@@ -493,21 +493,21 @@ ccl_device void svm_node_closure_volume(KernelGlobals *kg, ShaderData *sd, float
 
 	switch(type) {
 		case CLOSURE_VOLUME_ABSORPTION_ID: {
-			float3 weight = (make_float3(1.0f, 1.0f, 1.0f) - ccl_fetch(sd, svm_closure_weight)) * mix_weight * density;
+			float3 weight = (make_float3(1.0f, 1.0f, 1.0f) - sd->svm_closure_weight) * mix_weight * density;
 			ShaderClosure *sc = closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_NONE_ID, weight);
 
 			if(sc) {
-				ccl_fetch(sd, flag) |= volume_absorption_setup(sc);
+				sd->flag |= volume_absorption_setup(sc);
 			}
 			break;
 		}
 		case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID: {
-			float3 weight = ccl_fetch(sd, svm_closure_weight) * mix_weight * density;
+			float3 weight = sd->svm_closure_weight * mix_weight * density;
 			HenyeyGreensteinVolume *volume = (HenyeyGreensteinVolume*)bsdf_alloc(sd, sizeof(HenyeyGreensteinVolume), weight);
 
 			if(volume) {
 				volume->g = param2; /* g */
-				ccl_fetch(sd, flag) |= volume_henyey_greenstein_setup(volume);
+				sd->flag |= volume_henyey_greenstein_setup(volume);
 			}
 			break;
 		}
@@ -527,12 +527,12 @@ ccl_device void svm_node_closure_emission(ShaderData *sd, float *stack, uint4 no
 		if(mix_weight == 0.0f)
 			return;
 
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_EMISSION_ID, ccl_fetch(sd, svm_closure_weight) * mix_weight);
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_EMISSION_ID, sd->svm_closure_weight * mix_weight);
 	}
 	else
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_EMISSION_ID, ccl_fetch(sd, svm_closure_weight));
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_EMISSION_ID, sd->svm_closure_weight);
 
-	ccl_fetch(sd, flag) |= SD_EMISSION;
+	sd->flag |= SD_EMISSION;
 }
 
 ccl_device void svm_node_closure_background(ShaderData *sd, float *stack, uint4 node)
@@ -545,10 +545,10 @@ ccl_device void svm_node_closure_background(ShaderData *sd, float *stack, uint4
 		if(mix_weight == 0.0f)
 			return;
 
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_BACKGROUND_ID, ccl_fetch(sd, svm_closure_weight) * mix_weight);
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_BACKGROUND_ID, sd->svm_closure_weight * mix_weight);
 	}
 	else
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_BACKGROUND_ID, ccl_fetch(sd, svm_closure_weight));
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_BACKGROUND_ID, sd->svm_closure_weight);
 }
 
 ccl_device void svm_node_closure_holdout(ShaderData *sd, float *stack, uint4 node)
@@ -561,12 +561,12 @@ ccl_device void svm_node_closure_holdout(ShaderData *sd, float *stack, uint4 nod
 		if(mix_weight == 0.0f)
 			return;
 
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_HOLDOUT_ID, ccl_fetch(sd, svm_closure_weight) * mix_weight);
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_HOLDOUT_ID, sd->svm_closure_weight * mix_weight);
 	}
 	else
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_HOLDOUT_ID, ccl_fetch(sd, svm_closure_weight));
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_HOLDOUT_ID, sd->svm_closure_weight);
 
-	ccl_fetch(sd, flag) |= SD_HOLDOUT;
+	sd->flag |= SD_HOLDOUT;
 }
 
 ccl_device void svm_node_closure_ambient_occlusion(ShaderData *sd, float *stack, uint4 node)
@@ -579,19 +579,19 @@ ccl_device void svm_node_closure_ambient_occlusion(ShaderData *sd, float *stack,
 		if(mix_weight == 0.0f)
 			return;
 
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_AMBIENT_OCCLUSION_ID, ccl_fetch(sd, svm_closure_weight) * mix_weight);
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_AMBIENT_OCCLUSION_ID, sd->svm_closure_weight * mix_weight);
 	}
 	else
-		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_AMBIENT_OCCLUSION_ID, ccl_fetch(sd, svm_closure_weight));
+		closure_alloc(sd, sizeof(ShaderClosure), CLOSURE_AMBIENT_OCCLUSION_ID, sd->svm_closure_weight);
 
-	ccl_fetch(sd, flag) |= SD_AO;
+	sd->flag |= SD_AO;
 }
 
 /* Closure Nodes */
 
 ccl_device_inline void svm_node_closure_store_weight(ShaderData *sd, float3 weight)
 {
-	ccl_fetch(sd, svm_closure_weight) = weight;
+	sd->svm_closure_weight = weight;
 }
 
 ccl_device void svm_node_closure_set_weight(ShaderData *sd, uint r, uint g, uint b)
@@ -641,7 +641,7 @@ ccl_device void svm_node_mix_closure(ShaderData *sd, float *stack, uint4 node)
 ccl_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);
-	ccl_fetch(sd, N) = normal;
+	sd->N = normal;
 	stack_store_float3(stack, out_normal, normal);
 }
 
diff --git a/intern/cycles/kernel/svm/svm_displace.h b/intern/cycles/kernel/svm/svm_displace.h
index 890ab41aaaa..c94fa130af7 100644
--- a/intern/cycles/kernel/svm/svm_displace.h
+++ b/intern/cycles/kernel/svm/svm_displace.h
@@ -25,10 +25,10 @@ ccl_device void svm_node_set_bump(KernelGlobals *kg, ShaderData *sd, float *stac
 	uint normal_offset, distance_offset, invert, use_object_space;
 	decode_node_uchar4(node.y, &normal_offset, &distance_offset, &invert, &use_object_space);
 
-	float3 normal_in = stack_valid(normal_offset)? stack_load_float3(stack, normal_offset): ccl_fetch(sd, N);
+	float3 normal_in = stack_valid(normal_offset)? stack_load_float3(stack, normal_offset): sd->N;
 
-	float3 dPdx = ccl_fetch(sd, dP).dx;
-	float3 dPdy = ccl_fetch(sd, dP).dy;
+	float3 dPdx = sd->dP.dx;
+	float3 dPdy = sd->dP.dy;
 
 	if(use_object_space) {
 		object_inverse_normal_transform(kg, sd, &normal_in);
@@ -80,14 +80,14 @@ ccl_device void svm_node_set_displacement(KernelGlobals *kg, ShaderData *sd, flo
 {
 	float d = stack_load_float(stack, fac_offset);
 
-	float3 dP = ccl_fetch(sd, N);
+	float3 dP = sd->N;
 	object_inverse_normal_transform(kg, sd, &dP);
 
 	dP *= d*0.1f; /* todo: get rid of this factor */
 
 	object_dir_transform(kg, sd, &dP);
 
-	ccl_fetch(sd, P) += dP;
+	sd->P += dP;
 }
 
 CCL_NAMESPACE_END
diff --git a/intern/cycles/kernel/svm/svm_fresnel.h b/intern/cycles/kernel/svm/svm_fresnel.h
index 23c97d80cb0..3703ec55015 100644
--- a/intern/cycles/kernel/svm/svm_fresnel.h
+++ b/intern/cycles/kernel/svm/svm_fresnel.h
@@ -23,12 +23,12 @@ ccl_device void svm_node_fresnel(ShaderData *sd, float *stack, uint ior_offset,
 	uint normal_offset, out_offset;
 	decode_node_uchar4(node, &normal_offset, &out_offset, NULL, NULL);
 	float eta = (stack_valid(ior_offset))? stack_load_float(stack, ior_offset): __uint_as_float(ior_value);
-	float3 normal_in = stack_valid(normal_offset)? stack_load_float3(stack, normal_offset): ccl_fetch(sd, N);
+	float3 normal_in = stack_valid(normal_offset)? stack_load_float3(stack, normal_offset): sd->N;
 	
 	eta = fmaxf(eta, 1e-5f);
-	eta = (ccl_fetch(sd, flag) & SD_BACKFACING)? 1.0f/eta: eta;
+	eta = (sd->flag & SD_BACKFACING)? 1.0f/eta: eta;
 
-	float f = fresnel_dielectric_cos(dot(ccl_fetch(sd, I), normal_in), eta);
+	float f = fresnel_dielectric_cos(dot(sd->I, normal_in), eta);
 
 	stack_store_float(stack, out_offset, f);
 }
@@ -44,18 +44,18 @@ ccl_device void svm_node_layer_weight(ShaderData *sd, float *stack, uint4 node)
 	decode_node_uchar4(node.w, &type, &normal_offset, &out_offset, NULL);
 
 	float blend = (stack_valid(blend_offset))? stack_load_float(stack, blend_offset): __uint_as_float(blend_value);
-	float3 normal_in = (stack_valid(normal_offset))? stack_load_float3(stack, normal_offset): ccl_fetch(sd, N);
+	float3 normal_in = (stack_valid(normal_offset))? stack_load_float3(stack, normal_offset): sd->N;
 
 	float f;
 
 	if(type == NODE_LAYER_WEIGHT_FRESNEL) {
 		float eta = fmaxf(1.0f - blend, 1e-5f);
-		eta = (ccl_fetch(sd, flag) & SD_BACKFACING)? eta: 1.0f/eta;
+		eta = (sd->flag & SD_BACKFACING)? eta: 1.0f/eta;
 
-		f = fresnel_dielectric_cos(dot(ccl_fetch(sd, I), normal_in), eta);
+		f = fresnel_dielectric_cos(dot(sd->I, normal_in), eta);
 	}
 	else {
-		f = fabsf(dot(ccl_fetch(sd, I), normal_in));
+		f = fabsf(dot(sd->I, normal_in));
 
 		if(blend != 0.5f) {
 			blend = clamp(blend, 0.0f, 1.0f-1e-5f);
diff --git a/intern/cycles/kernel/svm/svm_geometry.h b/intern/cycles/kernel/svm/svm_geometry.h
index 7d512f7ff4d..4a09d9f6653 100644
--- a/intern/cycles/kernel/svm/svm_geometry.h
+++ b/intern/cycles/kernel/svm/svm_geometry.h
@@ -27,15 +27,15 @@ ccl_device_inline void svm_node_geometry(KernelGlobals *kg,
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = ccl_fetch(sd, P); break;
-		case NODE_GEOM_N: data = ccl_fetch(sd, N); break;
+		case NODE_GEOM_P: data = sd->P; break;
+		case NODE_GEOM_N: data = sd->N; break;
 #ifdef __DPDU__
 		case NODE_GEOM_T: data = primitive_tangent(kg, sd); break;
 #endif
-		case NODE_GEOM_I: data = ccl_fetch(sd, I); break;
-		case NODE_GEOM_Ng: data = ccl_fetch(sd, Ng); break;
+		case NODE_GEOM_I: data = sd->I; break;
+		case NODE_GEOM_Ng: data = sd->Ng; break;
 #ifdef __UV__
-		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u), ccl_fetch(sd, v), 0.0f); break;
+		case NODE_GEOM_uv: data = make_float3(sd->u, sd->v, 0.0f); break;
 #endif
 	}
 
@@ -48,8 +48,8 @@ ccl_device void svm_node_geometry_bump_dx(KernelGlobals *kg, ShaderData *sd, flo
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx; break;
-		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u) + ccl_fetch(sd, du).dx, ccl_fetch(sd, v) + ccl_fetch(sd, dv).dx, 0.0f); break;
+		case NODE_GEOM_P: data = sd->P + sd->dP.dx; break;
+		case NODE_GEOM_uv: data = make_float3(sd->u + sd->du.dx, sd->v + sd->dv.dx, 0.0f); break;
 		default: svm_node_geometry(kg, sd, stack, type, out_offset); return;
 	}
 
@@ -65,8 +65,8 @@ ccl_device void svm_node_geometry_bump_dy(KernelGlobals *kg, ShaderData *sd, flo
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy; break;
-		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u) + ccl_fetch(sd, du).dy, ccl_fetch(sd, v) + ccl_fetch(sd, dv).dy, 0.0f); break;
+		case NODE_GEOM_P: data = sd->P + sd->dP.dy; break;
+		case NODE_GEOM_uv: data = make_float3(sd->u + sd->du.dy, sd->v + sd->dv.dy, 0.0f); break;
 		default: svm_node_geometry(kg, sd, stack, type, out_offset); return;
 	}
 
@@ -87,9 +87,9 @@ ccl_device void svm_node_object_info(KernelGlobals *kg, ShaderData *sd, float *s
 			stack_store_float3(stack, out_offset, object_location(kg, sd));
 			return;
 		}
-		case NODE_INFO_OB_INDEX: data = object_pass_id(kg, ccl_fetch(sd, object)); break;
+		case NODE_INFO_OB_INDEX: data = object_pass_id(kg, sd->object); break;
 		case NODE_INFO_MAT_INDEX: data = shader_pass_id(kg, sd); break;
-		case NODE_INFO_OB_RANDOM: data = object_random_number(kg, ccl_fetch(sd, object)); break;
+		case NODE_INFO_OB_RANDOM: data = object_random_number(kg, sd->object); break;
 		default: data = 0.0f; break;
 	}
 
@@ -106,44 +106,44 @@ ccl_device void svm_node_particle_info(KernelGlobals *kg,
 {
 	switch(type) {
 		case NODE_INFO_PAR_INDEX: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float(stack, out_offset, particle_index(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_AGE: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float(stack, out_offset, particle_age(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_LIFETIME: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float(stack, out_offset, particle_lifetime(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_LOCATION: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float3(stack, out_offset, particle_location(kg, particle_id));
 			break;
 		}
 #if 0	/* XXX float4 currently not supported in SVM stack */
 		case NODE_INFO_PAR_ROTATION: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float4(stack, out_offset, particle_rotation(kg, particle_id));
 			break;
 		}
 #endif
 		case NODE_INFO_PAR_SIZE: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float(stack, out_offset, particle_size(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_VELOCITY: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float3(stack, out_offset, particle_velocity(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_ANGULAR_VELOCITY: {
-			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
+			int particle_id = object_particle_id(kg, sd->object);
 			stack_store_float3(stack, out_offset, particle_angular_velocity(kg, particle_id));
 			break;
 		}
@@ -165,7 +165,7 @@ ccl_device void svm_node_hair_info(KernelGlobals *kg,
 
 	switch(type) {
 		case NODE_INFO_CURVE_IS_STRAND: {
-			data = (ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) != 0;
+			data = (sd->type & PRIMITIVE_ALL_CURVE) != 0;
 			stack_store_float(stack, out_offset, data);
 			break;
 		}
@@ -177,7 +177,7 @@ ccl_device void svm_node_hair_info(KernelGlobals *kg,
 			break;
 		}
 		/*case NODE_INFO_CURVE_FADE: {
-			data = ccl_fetch(sd, curve_transparency);
+			data = sd->curve_transparency;
 			stack_store_float(stack, out_offset, data);
 			break;
 		}*/
diff --git a/intern/cycles/kernel/svm/svm_image.h b/intern/cycles/kernel/svm/svm_image.h
index 0d6efb47223..76acc9253a1 100644
--- a/intern/cycles/kernel/svm/svm_image.h
+++ b/intern/cycles/kernel/svm/svm_image.h
@@ -237,9 +237,9 @@ ccl_device void svm_node_tex_image(KernelGlobals *kg, ShaderData *sd, float *sta
 ccl_device void svm_node_tex_image_box(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node)
 {
 	/* get object space normal */
-	float3 N = ccl_fetch(sd, N);
+	float3 N = sd->N;
 
-	N = ccl_fetch(sd, N);
+	N = sd->N;
 	object_inverse_normal_transform(kg, sd, &N);
 
 	/* project from direction vector to barycentric coordinates in triangles */
diff --git a/intern/cycles/kernel/svm/svm_light_path.h b/intern/cycles/kernel/svm/svm_light_path.h
index 04f6f623f18..1492e358608 100644
--- a/intern/cycles/kernel/svm/svm_light_path.h
+++ b/intern/cycles/kernel/svm/svm_light_path.h
@@ -31,8 +31,8 @@ ccl_device void svm_node_light_path(ShaderData *sd, ccl_addr_space PathState *st
 		case NODE_LP_reflection: info = (path_flag & PATH_RAY_REFLECT)? 1.0f: 0.0f; break;
 		case NODE_LP_transmission: info = (path_flag & PATH_RAY_TRANSMIT)? 1.0f: 0.0f; break;
 		case NODE_LP_volume_scatter: info = (path_flag & PATH_RAY_VOLUME_SCATTER)? 1.0f: 0.0f; break;
-		case NODE_LP_backfacing: info = (ccl_fetch(sd, flag) & SD_BACKFACING)? 1.0f: 0.0f; break;
-		case NODE_LP_ray_length: info = ccl_fetch(sd, ray_length); break;
+		case NODE_LP_backfacing: info = (sd->flag & SD_BACKFACING)? 1.0f: 0.0f; break;
+		case NODE_LP_ray_length: info = sd->ray_length; break;
 		case NODE_LP_ray_depth: info = (float)state->bounce; break;
 		case NODE_LP_ray_diffuse: info = (float)state->diffuse_bounce; break;
 		case NODE_LP_ray_glossy: info = (float)state->glossy_bounce; break;
@@ -56,14 +56,14 @@ ccl_device void svm_node_light_falloff(ShaderData *sd, float *stack, uint4 node)
 
 	switch(type) {
 		case NODE_LIGHT_FALLOFF_QUADRATIC: break;
-		case NODE_LIGHT_FALLOFF_LINEAR: strength *= ccl_fetch(sd, ray_length); break;
-		case NODE_LIGHT_FALLOFF_CONSTANT: strength *= ccl_fetch(sd, ray_length)*ccl_fetch(sd, ray_length); break;
+		case NODE_LIGHT_FALLOFF_LINEAR: strength *= sd->ray_length; break;
+		case NODE_LIGHT_FALLOFF_CONSTANT: strength *= sd->ray_length*sd->ray_length; break;
 	}
 
 	float smooth = stack_load_float(stack, smooth_offset);
 
 	if(smooth > 0.0f) {
-		float squared = ccl_fetch(sd, ray_length)*ccl_fetch(sd, ray_length);
+		float squared = sd->ray_length*sd->ray_length;
 		/* Distant lamps set the ray length to FLT_MAX, which causes squared to overflow. */
 		if(isfinite(squared)) {
 			strength *= squared/(smooth + squared);
diff --git a/intern/cycles/kernel/svm/svm_tex_coord.h b/intern/cycles/kernel/svm/svm_tex_coord.h
index c0b01262212..c94327401f5 100644
--- a/intern/cycles/kernel/svm/svm_tex_coord.h
+++ b/intern/cycles/kernel/svm/svm_tex_coord.h
@@ -31,9 +31,9 @@ ccl_device void svm_node_tex_coord(KernelGlobals *kg,
 
 	switch(type) {
 		case NODE_TEXCO_OBJECT: {
-			data = ccl_fetch(sd, P);
+			data = sd->P;
 			if(node.w == 0) {
-				if(ccl_fetch(sd, object) != OBJECT_NONE) {
+				if(sd->object != OBJECT_NONE) {
 					object_inverse_position_transform(kg, sd, &data);
 				}
 			}
@@ -48,47 +48,47 @@ ccl_device void svm_node_tex_coord(KernelGlobals *kg,
 			break;
 		}
 		case NODE_TEXCO_NORMAL: {
-			data = ccl_fetch(sd, N);
+			data = sd->N;
 			object_inverse_normal_transform(kg, sd, &data);
 			break;
 		}
 		case NODE_TEXCO_CAMERA: {
 			Transform tfm = kernel_data.cam.worldtocamera;
 
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = transform_point(&tfm, ccl_fetch(sd, P));
+			if(sd->object != OBJECT_NONE)
+				data = transform_point(&tfm, sd->P);
 			else
-				data = transform_point(&tfm, ccl_fetch(sd, P) + camera_position(kg));
+				data = transform_point(&tfm, sd->P + camera_position(kg));
 			break;
 		}
 		case NODE_TEXCO_WINDOW: {
-			if((path_flag & PATH_RAY_CAMERA) && ccl_fetch(sd, object) == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, ray_P));
+			if((path_flag & PATH_RAY_CAMERA) && sd->object == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
+				data = camera_world_to_ndc(kg, sd, sd->ray_P);
 			else
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, P));
+				data = camera_world_to_ndc(kg, sd, sd->P);
 			data.z = 0.0f;
 			break;
 		}
 		case NODE_TEXCO_REFLECTION: {
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = 2.0f*dot(ccl_fetch(sd, N), ccl_fetch(sd, I))*ccl_fetch(sd, N) - ccl_fetch(sd, I);
+			if(sd->object != OBJECT_NONE)
+				data = 2.0f*dot(sd->N, sd->I)*sd->N - sd->I;
 			else
-				data = ccl_fetch(sd, I);
+				data = sd->I;
 			break;
 		}
 		case NODE_TEXCO_DUPLI_GENERATED: {
-			data = object_dupli_generated(kg, ccl_fetch(sd, object));
+			data = object_dupli_generated(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_DUPLI_UV: {
-			data = object_dupli_uv(kg, ccl_fetch(sd, object));
+			data = object_dupli_uv(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_VOLUME_GENERATED: {
-			data = ccl_fetch(sd, P);
+			data = sd->P;
 
 #ifdef __VOLUME__
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
+			if(sd->object != OBJECT_NONE)
 				data = volume_normalized_position(kg, sd, data);
 #endif
 			break;
@@ -112,9 +112,9 @@ ccl_device void svm_node_tex_coord_bump_dx(KernelGlobals *kg,
 
 	switch(type) {
 		case NODE_TEXCO_OBJECT: {
-			data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx;
+			data = sd->P + sd->dP.dx;
 			if(node.w == 0) {
-				if(ccl_fetch(sd, object) != OBJECT_NONE) {
+				if(sd->object != OBJECT_NONE) {
 					object_inverse_position_transform(kg, sd, &data);
 				}
 			}
@@ -129,47 +129,47 @@ ccl_device void svm_node_tex_coord_bump_dx(KernelGlobals *kg,
 			break;
 		}
 		case NODE_TEXCO_NORMAL: {
-			data = ccl_fetch(sd, N);
+			data = sd->N;
 			object_inverse_normal_transform(kg, sd, &data);
 			break;
 		}
 		case NODE_TEXCO_CAMERA: {
 			Transform tfm = kernel_data.cam.worldtocamera;
 
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = transform_point(&tfm, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx);
+			if(sd->object != OBJECT_NONE)
+				data = transform_point(&tfm, sd->P + sd->dP.dx);
 			else
-				data = transform_point(&tfm, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx + camera_position(kg));
+				data = transform_point(&tfm, sd->P + sd->dP.dx + camera_position(kg));
 			break;
 		}
 		case NODE_TEXCO_WINDOW: {
-			if((path_flag & PATH_RAY_CAMERA) && ccl_fetch(sd, object) == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, ray_P) + ccl_fetch(sd, ray_dP).dx);
+			if((path_flag & PATH_RAY_CAMERA) && sd->object == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
+				data = camera_world_to_ndc(kg, sd, sd->ray_P + sd->ray_dP.dx);
 			else
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx);
+				data = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dx);
 			data.z = 0.0f;
 			break;
 		}
 		case NODE_TEXCO_REFLECTION: {
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = 2.0f*dot(ccl_fetch(sd, N), ccl_fetch(sd, I))*ccl_fetch(sd, N) - ccl_fetch(sd, I);
+			if(sd->object != OBJECT_NONE)
+				data = 2.0f*dot(sd->N, sd->I)*sd->N - sd->I;
 			else
-				data = ccl_fetch(sd, I);
+				data = sd->I;
 			break;
 		}
 		case NODE_TEXCO_DUPLI_GENERATED: {
-			data = object_dupli_generated(kg, ccl_fetch(sd, object));
+			data = object_dupli_generated(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_DUPLI_UV: {
-			data = object_dupli_uv(kg, ccl_fetch(sd, object));
+			data = object_dupli_uv(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_VOLUME_GENERATED: {
-			data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx;
+			data = sd->P + sd->dP.dx;
 
 #ifdef __VOLUME__
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
+			if(sd->object != OBJECT_NONE)
 				data = volume_normalized_position(kg, sd, data);
 #endif
 			break;
@@ -196,9 +196,9 @@ ccl_device void svm_node_tex_coord_bump_dy(KernelGlobals *kg,
 
 	switch(type) {
 		case NODE_TEXCO_OBJECT: {
-			data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy;
+			data = sd->P + sd->dP.dy;
 			if(node.w == 0) {
-				if(ccl_fetch(sd, object) != OBJECT_NONE) {
+				if(sd->object != OBJECT_NONE) {
 					object_inverse_position_transform(kg, sd, &data);
 				}
 			}
@@ -213,47 +213,47 @@ ccl_device void svm_node_tex_coord_bump_dy(KernelGlobals *kg,
 			break;
 		}
 		case NODE_TEXCO_NORMAL: {
-			data = ccl_fetch(sd, N);
+			data = sd->N;
 			object_inverse_normal_transform(kg, sd, &data);
 			break;
 		}
 		case NODE_TEXCO_CAMERA: {
 			Transform tfm = kernel_data.cam.worldtocamera;
 
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = transform_point(&tfm, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy);
+			if(sd->object != OBJECT_NONE)
+				data = transform_point(&tfm, sd->P + sd->dP.dy);
 			else
-				data = transform_point(&tfm, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy + camera_position(kg));
+				data = transform_point(&tfm, sd->P + sd->dP.dy + camera_position(kg));
 			break;
 		}
 		case NODE_TEXCO_WINDOW: {
-			if((path_flag & PATH_RAY_CAMERA) && ccl_fetch(sd, object) == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, ray_P) + ccl_fetch(sd, ray_dP).dy);
+			if((path_flag & PATH_RAY_CAMERA) && sd->object == OBJECT_NONE && kernel_data.cam.type == CAMERA_ORTHOGRAPHIC)
+				data = camera_world_to_ndc(kg, sd, sd->ray_P + sd->ray_dP.dy);
 			else
-				data = camera_world_to_ndc(kg, sd, ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy);
+				data = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dy);
 			data.z = 0.0f;
 			break;
 		}
 		case NODE_TEXCO_REFLECTION: {
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
-				data = 2.0f*dot(ccl_fetch(sd, N), ccl_fetch(sd, I))*ccl_fetch(sd, N) - ccl_fetch(sd, I);
+			if(sd->object != OBJECT_NONE)
+				data = 2.0f*dot(sd->N, sd->I)*sd->N - sd->I;
 			else
-				data = ccl_fetch(sd, I);
+				data = sd->I;
 			break;
 		}
 		case NODE_TEXCO_DUPLI_GENERATED: {
-			data = object_dupli_generated(kg, ccl_fetch(sd, object));
+			data = object_dupli_generated(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_DUPLI_UV: {
-			data = object_dupli_uv(kg, ccl_fetch(sd, object));
+			data = object_dupli_uv(kg, sd->object);
 			break;
 		}
 		case NODE_TEXCO_VOLUME_GENERATED: {
-			data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy;
+			data = sd->P + sd->dP.dy;
 
 #ifdef __VOLUME__
-			if(ccl_fetch(sd, object) != OBJECT_NONE)
+			if(sd->object != OBJECT_NONE)
 				data = volume_normalized_position(kg, sd, data);
 #endif
 			break;
@@ -274,12 +274,12 @@ ccl_device void svm_node_normal_map(KernelGlobals *kg, ShaderData *sd, float *st
 	float3 color = stack_load_float3(stack, color_offset);
 	color = 2.0f*make_float3(color.x - 0.5f, color.y - 0.5f, color.z - 0.5f);
 
-	bool is_backfacing = (ccl_fetch(sd, flag) & SD_BACKFACING) != 0;
+	bool is_backfacing = (sd->flag & SD_BACKFACING) != 0;
 	float3 N;
 
 	if(space == NODE_NORMAL_MAP_TANGENT) {
 		/* tangent space */
-		if(ccl_fetch(sd, object) == OBJECT_NONE) {
+		if(sd->object == OBJECT_NONE) {
 			stack_store_float3(stack, normal_offset, make_float3(0.0f, 0.0f, 0.0f));
 			return;
 		}
@@ -299,11 +299,11 @@ ccl_device void svm_node_normal_map(KernelGlobals *kg, ShaderData *sd, float *st
 		float sign = primitive_attribute_float(kg, sd, attr_sign, NULL, NULL);
 		float3 normal;
 
-		if(ccl_fetch(sd, shader) & SHADER_SMOOTH_NORMAL) {
+		if(sd->shader & SHADER_SMOOTH_NORMAL) {
 			normal = primitive_attribute_float3(kg, sd, attr_normal, NULL, NULL);
 		}
 		else {
-			normal = ccl_fetch(sd, Ng);
+			normal = sd->Ng;
 
 			/* the normal is already inverted, which is too soon for the math here */
 			if(is_backfacing) {
@@ -345,11 +345,11 @@ ccl_device void svm_node_normal_map(KernelGlobals *kg, ShaderData *sd, float *st
 
 	if(strength != 1.0f) {
 		strength = max(strength, 0.0f);
-		N = safe_normalize(ccl_fetch(sd, N) + (N - ccl_fetch(sd, N))*strength);
+		N = safe_normalize(sd->N + (N - sd->N)*strength);
 	}
 
 	if(is_zero(N)) {
-		N = ccl_fetch(sd, N);
+		N = sd->N;
 	}
 
 	stack_store_float3(stack, normal_offset, N);
@@ -377,7 +377,7 @@ ccl_device void svm_node_tangent(KernelGlobals *kg, ShaderData *sd, float *stack
 		float3 generated;
 
 		if(desc.offset == ATTR_STD_NOT_FOUND)
-			generated = ccl_fetch(sd, P);
+			generated = sd->P;
 		else
 			generated = primitive_attribute_float3(kg, sd, desc, NULL, NULL);
 
@@ -390,7 +390,7 @@ ccl_device void svm_node_tangent(KernelGlobals *kg, ShaderData *sd, float *stack
 	}
 
 	object_normal_transform(kg, sd, &tangent);
-	tangent = cross(ccl_fetch(sd, N), normalize(cross(tangent, ccl_fetch(sd, N))));
+	tangent = cross(sd->N, normalize(cross(tangent, sd->N)));
 	stack_store_float3(stack, tangent_offset, tangent);
 }
 
diff --git a/intern/cycles/kernel/svm/svm_vector_transform.h b/intern/cycles/kernel/svm/svm_vector_transform.h
index 4c32130d06d..4e92f27acdb 100644
--- a/intern/cycles/kernel/svm/svm_vector_transform.h
+++ b/intern/cycles/kernel/svm/svm_vector_transform.h
@@ -33,7 +33,7 @@ ccl_device void svm_node_vector_transform(KernelGlobals *kg, ShaderData *sd, flo
 	NodeVectorTransformConvertSpace to = (NodeVectorTransformConvertSpace)ito;
 	
 	Transform tfm;
-	bool is_object = (ccl_fetch(sd, object) != OBJECT_NONE);
+	bool is_object = (sd->object != OBJECT_NONE);
 	bool is_direction = (type == NODE_VECTOR_TRANSFORM_TYPE_VECTOR || type == NODE_VECTOR_TRANSFORM_TYPE_NORMAL);
 	
 	/* From world */
diff --git a/intern/cycles/kernel/svm/svm_wireframe.h b/intern/cycles/kernel/svm/svm_wireframe.h
index 87e40791333..3c6353c8001 100644
--- a/intern/cycles/kernel/svm/svm_wireframe.h
+++ b/intern/cycles/kernel/svm/svm_wireframe.h
@@ -41,9 +41,9 @@ ccl_device_inline float wireframe(KernelGlobals *kg,
                                   float3 *P)
 {
 #ifdef __HAIR__
-	if(ccl_fetch(sd, prim) != PRIM_NONE && ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE)
+	if(sd->prim != PRIM_NONE && sd->type & PRIMITIVE_ALL_TRIANGLE)
 #else
-	if(ccl_fetch(sd, prim) != PRIM_NONE)
+	if(sd->prim != PRIM_NONE)
 #endif
 	{
 		float3 Co[3];
@@ -52,12 +52,12 @@ ccl_device_inline float wireframe(KernelGlobals *kg,
 		/* Triangles */
 		int np = 3;
 
-		if(ccl_fetch(sd, type) & PRIMITIVE_TRIANGLE)
-			triangle_vertices(kg, ccl_fetch(sd, prim), Co);
+		if(sd->type & PRIMITIVE_TRIANGLE)
+			triangle_vertices(kg, sd->prim, Co);
 		else
-			motion_triangle_vertices(kg, ccl_fetch(sd, object), ccl_fetch(sd, prim), ccl_fetch(sd, time), Co);
+			motion_triangle_vertices(kg, sd->object, sd->prim, sd->time, Co);
 
-		if(!(ccl_fetch(sd, object_flag) & SD_OBJECT_TRANSFORM_APPLIED)) {
+		if(!(sd->object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
 			object_position_transform(kg, sd, &Co[0]);
 			object_position_transform(kg, sd, &Co[1]);
 			object_position_transform(kg, sd, &Co[2]);
@@ -66,8 +66,8 @@ ccl_device_inline float wireframe(KernelGlobals *kg,
 		if(pixel_size) {
 			// Project the derivatives of P to the viewing plane defined
 			// by I so we have a measure of how big is a pixel at this point
-			float pixelwidth_x = len(ccl_fetch(sd, dP).dx - dot(ccl_fetch(sd, dP).dx, ccl_fetch(sd, I)) * ccl_fetch(sd, I));
-			float pixelwidth_y = len(ccl_fetch(sd, dP).dy - dot(ccl_fetch(sd, dP).dy, ccl_fetch(sd, I)) * ccl_fetch(sd, I));
+			float pixelwidth_x = len(sd->dP.dx - dot(sd->dP.dx, sd->I) * sd->I);
+			float pixelwidth_y = len(sd->dP.dy - dot(sd->dP.dy, sd->I) * sd->I);
 			// Take the average of both axis' length
 			pixelwidth = (pixelwidth_x + pixelwidth_y) * 0.5f;
 		}
@@ -113,20 +113,20 @@ ccl_device void svm_node_wireframe(KernelGlobals *kg,
 	 * With OpenCL 2.0 it's possible to avoid this change, but for until
 	 * then we'll be living with such an exception.
 	 */
-	float3 P = ccl_fetch(sd, P);
+	float3 P = sd->P;
 	float f = wireframe(kg, sd, size, pixel_size, &P);
 #else
-	float f = wireframe(kg, sd, size, pixel_size, &ccl_fetch(sd, P));
+	float f = wireframe(kg, sd, size, pixel_size, &sd->P);
 #endif
 
 	/* TODO(sergey): Think of faster way to calculate derivatives. */
 	if(bump_offset == NODE_BUMP_OFFSET_DX) {
-		float3 Px = ccl_fetch(sd, P) - ccl_fetch(sd, dP).dx;
-		f += (f - wireframe(kg, sd, size, pixel_size, &Px)) / len(ccl_fetch(sd, dP).dx);
+		float3 Px = sd->P - sd->dP.dx;
+		f += (f - wireframe(kg, sd, size, pixel_size, &Px)) / len(sd->dP.dx);
 	}
 	else if(bump_offset == NODE_BUMP_OFFSET_DY) {
-		float3 Py = ccl_fetch(sd, P) - ccl_fetch(sd, dP).dy;
-		f += (f - wireframe(kg, sd, size, pixel_size, &Py)) / len(ccl_fetch(sd, dP).dy);
+		float3 Py = sd->P - sd->dP.dy;
+		f += (f - wireframe(kg, sd, size, pixel_size, &Py)) / len(sd->dP.dy);
 	}
 
 	if(stack_valid(out_fac))