Cycles: Solve speed regression of classroom scene after principled commit

This way we can skip it from compiling into OpenCL kernels by making
this shader compile-time feature.

8 files changed, 38 insertions, 3 deletions

diff --git a/intern/cycles/device/device.cpp b/intern/cycles/device/device.cpp
index 968af447e29..c024021b4b3 100644
--- a/intern/cycles/device/device.cpp
+++ b/intern/cycles/device/device.cpp
@@ -66,6 +66,8 @@ std::ostream& operator <<(std::ostream &os,
 	   << string_from_bool(requested_features.use_patch_evaluation) << std::endl;
 	os << "Use Transparent Shadows: "
 	   << string_from_bool(requested_features.use_transparent) << std::endl;
+	os << "Use Principled BSDF: "
+	   << string_from_bool(requested_features.use_principled) << std::endl;
 	return os;
 }
 
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index ac06e561795..6051dd8b3eb 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -124,6 +124,9 @@ public:
 	/* Use various shadow tricks, such as shadow catcher. */
 	bool use_shadow_tricks;
 
+	/* Per-uber shader usage flags. */
+	bool use_principled;
+
 	DeviceRequestedFeatures()
 	{
 		/* TODO(sergey): Find more meaningful defaults. */
@@ -141,6 +144,7 @@ public:
 		use_patch_evaluation = false;
 		use_transparent = false;
 		use_shadow_tricks = false;
+		use_principled = false;
 	}
 
 	bool modified(const DeviceRequestedFeatures& requested_features)
@@ -158,7 +162,8 @@ public:
 		         use_integrator_branched == requested_features.use_integrator_branched &&
 		         use_patch_evaluation == requested_features.use_patch_evaluation &&
 		         use_transparent == requested_features.use_transparent &&
-		         use_shadow_tricks == requested_features.use_shadow_tricks);
+		         use_shadow_tricks == requested_features.use_shadow_tricks &&
+		         use_principled == requested_features.use_principled);
 	}
 
 	/* Convert the requested features structure to a build options,
@@ -205,6 +210,9 @@ public:
 		if(!use_shadow_tricks) {
 			build_options += " -D__NO_SHADOW_TRICKS__";
 		}
+		if(!use_principled) {
+			build_options += " -D__NO_PRINCIPLED__";
+		}
 		return build_options;
 	}
 };
diff --git a/intern/cycles/kernel/closure/bsdf.h b/intern/cycles/kernel/closure/bsdf.h
index 0302fa9b43e..a6bba8bf74d 100644
--- a/intern/cycles/kernel/closure/bsdf.h
+++ b/intern/cycles/kernel/closure/bsdf.h
@@ -137,6 +137,7 @@ ccl_device_forceinline int bsdf_sample(KernelGlobals *kg,
 			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;
+#ifdef __PRINCIPLED__
 		case CLOSURE_BSDF_PRINCIPLED_DIFFUSE_ID:
 		case CLOSURE_BSDF_BSSRDF_PRINCIPLED_ID:
 			label = bsdf_principled_diffuse_sample(sc, sd->Ng, sd->I, sd->dI.dx, sd->dI.dy, randu, randv,
@@ -146,6 +147,7 @@ ccl_device_forceinline int bsdf_sample(KernelGlobals *kg,
 			label = bsdf_principled_sheen_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  /* __PRINCIPLED__ */
 #endif
 #ifdef __VOLUME__
 		case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
@@ -243,6 +245,7 @@ float3 bsdf_eval(KernelGlobals *kg,
 			case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
 				eval = bsdf_hair_transmission_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
+#ifdef __PRINCIPLED__
 			case CLOSURE_BSDF_PRINCIPLED_DIFFUSE_ID:
 			case CLOSURE_BSDF_BSSRDF_PRINCIPLED_ID:
 				eval = bsdf_principled_diffuse_eval_reflect(sc, sd->I, omega_in, pdf);
@@ -250,6 +253,7 @@ float3 bsdf_eval(KernelGlobals *kg,
 			case CLOSURE_BSDF_PRINCIPLED_SHEEN_ID:
 				eval = bsdf_principled_sheen_eval_reflect(sc, sd->I, omega_in, pdf);
 				break;
+#endif  /* __PRINCIPLED__ */
 #endif
 #ifdef __VOLUME__
 			case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
@@ -323,6 +327,7 @@ float3 bsdf_eval(KernelGlobals *kg,
 			case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
 				eval = bsdf_hair_transmission_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
+#ifdef __PRINCIPLED__
 			case CLOSURE_BSDF_PRINCIPLED_DIFFUSE_ID:
 			case CLOSURE_BSDF_BSSRDF_PRINCIPLED_ID:
 				eval = bsdf_principled_diffuse_eval_transmit(sc, sd->I, omega_in, pdf);
@@ -330,6 +335,7 @@ float3 bsdf_eval(KernelGlobals *kg,
 			case CLOSURE_BSDF_PRINCIPLED_SHEEN_ID:
 				eval = bsdf_principled_sheen_eval_transmit(sc, sd->I, omega_in, pdf);
 				break;
+#endif  /* __PRINCIPLED__ */
 #endif
 #ifdef __VOLUME__
 			case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
diff --git a/intern/cycles/kernel/closure/bssrdf.h b/intern/cycles/kernel/closure/bssrdf.h
index 8363cef53c8..f9236a6e52c 100644
--- a/intern/cycles/kernel/closure/bssrdf.h
+++ b/intern/cycles/kernel/closure/bssrdf.h
@@ -362,6 +362,7 @@ ccl_device int bssrdf_setup(Bssrdf *bssrdf, ClosureType type)
 	if(bssrdf->radius < BSSRDF_MIN_RADIUS) {
 		/* revert to diffuse BSDF if radius too small */
 		int flag;
+#ifdef __PRINCIPLED__
 		if(type == CLOSURE_BSSRDF_PRINCIPLED_ID) {
 			float roughness = bssrdf->roughness;
 			float3 N = bssrdf->N;
@@ -377,7 +378,9 @@ ccl_device int bssrdf_setup(Bssrdf *bssrdf, ClosureType type)
 			flag = bsdf_principled_diffuse_setup(bsdf);
 			bsdf->type = CLOSURE_BSDF_BSSRDF_PRINCIPLED_ID;
 		}
-		else {
+		else
+#endif  /* __PRINCIPLED__ */
+		{
 			DiffuseBsdf *bsdf = (DiffuseBsdf*)bssrdf;
 			bsdf->N = bssrdf->N;
 			flag = bsdf_diffuse_setup(bsdf);
@@ -391,7 +394,9 @@ ccl_device int bssrdf_setup(Bssrdf *bssrdf, ClosureType type)
 		bssrdf->sharpness = saturate(bssrdf->sharpness);
 		bssrdf->type = type;
 
-		if(type == CLOSURE_BSSRDF_BURLEY_ID || type == CLOSURE_BSSRDF_PRINCIPLED_ID) {
+		if(type == CLOSURE_BSSRDF_BURLEY_ID ||
+		   type == CLOSURE_BSSRDF_PRINCIPLED_ID)
+		{
 			bssrdf_burley_setup(bssrdf);
 		}
 
diff --git a/intern/cycles/kernel/kernel_subsurface.h b/intern/cycles/kernel/kernel_subsurface.h
index fb4a054dcd0..f58eef7a368 100644
--- a/intern/cycles/kernel/kernel_subsurface.h
+++ b/intern/cycles/kernel/kernel_subsurface.h
@@ -149,6 +149,7 @@ ccl_device void subsurface_scatter_setup_diffuse_bsdf(ShaderData *sd, ShaderClos
 
 	if(hit) {
 		Bssrdf *bssrdf = (Bssrdf *)sc;
+#ifdef __PRINCIPLED__
 		if(bssrdf->type == CLOSURE_BSSRDF_PRINCIPLED_ID) {
 			PrincipledDiffuseBsdf *bsdf = (PrincipledDiffuseBsdf*)bsdf_alloc(sd, sizeof(PrincipledDiffuseBsdf), weight);
 
@@ -164,6 +165,7 @@ ccl_device void subsurface_scatter_setup_diffuse_bsdf(ShaderData *sd, ShaderClos
 		}
 		else if(CLOSURE_IS_BSDF_BSSRDF(bssrdf->type) ||
 		        CLOSURE_IS_BSSRDF(bssrdf->type))
+#endif  /* __PRINCIPLED__ */
 		{
 			DiffuseBsdf *bsdf = (DiffuseBsdf*)bsdf_alloc(sd, sizeof(DiffuseBsdf), weight);
 
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 9d5cad3f3e1..a744c1d8277 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -78,6 +78,7 @@ CCL_NAMESPACE_BEGIN
 #    define __OSL__
 #  endif
 #  define __SUBSURFACE__
+#  define __PRINCIPLED__
 #  define __CMJ__
 #  define __VOLUME__
 #  define __VOLUME_SCATTER__
@@ -94,6 +95,7 @@ CCL_NAMESPACE_BEGIN
 #  define __VOLUME__
 #  define __VOLUME_SCATTER__
 #  define __SUBSURFACE__
+#  define __PRINCIPLED__
 #  define __SHADOW_RECORD_ALL__
 #  ifndef __SPLIT_KERNEL__
 #    define __BRANCHED_PATH__
@@ -109,6 +111,7 @@ CCL_NAMESPACE_BEGIN
 #    define __KERNEL_SHADING__
 #    define __KERNEL_ADV_SHADING__
 #    define __SUBSURFACE__
+#    define __PRINCIPLED__
 #    define __VOLUME__
 #    define __VOLUME_SCATTER__
 #    define __SHADOW_RECORD_ALL__
@@ -134,6 +137,7 @@ CCL_NAMESPACE_BEGIN
 #    define __KERNEL_SHADING__
 #    define __KERNEL_ADV_SHADING__
 #    define __SUBSURFACE__
+#    define __PRINCIPLED__
 #    define __VOLUME__
 #    define __VOLUME_SCATTER__
 #    define __SHADOW_RECORD_ALL__
@@ -222,6 +226,9 @@ CCL_NAMESPACE_BEGIN
 #ifdef __NO_SHADOW_TRICKS__
 #  undef __SHADOW_TRICKS__
 #endif
+#ifdef __NO_PRINCIPLED__
+#  undef __PRINCIPLED__
+#endif
 
 /* Random Numbers */
 
diff --git a/intern/cycles/kernel/svm/svm_closure.h b/intern/cycles/kernel/svm/svm_closure.h
index 9a3689a94f4..407f8e784c0 100644
--- a/intern/cycles/kernel/svm/svm_closure.h
+++ b/intern/cycles/kernel/svm/svm_closure.h
@@ -76,6 +76,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 	float param2 = (stack_valid(param2_offset))? stack_load_float(stack, param2_offset): __uint_as_float(node.w);
 
 	switch(type) {
+#ifdef __PRINCIPLED__
 		case CLOSURE_BSDF_PRINCIPLED_ID: {
 			uint specular_offset, roughness_offset, specular_tint_offset, anisotropic_offset, sheen_offset,
 				sheen_tint_offset, clearcoat_offset, clearcoat_gloss_offset, eta_offset, transparency_offset,
@@ -408,6 +409,7 @@ ccl_device void svm_node_closure_bsdf(KernelGlobals *kg, ShaderData *sd, float *
 
 			break;
 		}
+#endif  /* __PRINCIPLED__ */
 		case CLOSURE_BSDF_DIFFUSE_ID: {
 			float3 weight = sd->svm_closure_weight * mix_weight;
 			OrenNayarBsdf *bsdf = (OrenNayarBsdf*)bsdf_alloc(sd, sizeof(OrenNayarBsdf), weight);
diff --git a/intern/cycles/render/shader.cpp b/intern/cycles/render/shader.cpp
index 23eee1916bd..a7d42a4b4a0 100644
--- a/intern/cycles/render/shader.cpp
+++ b/intern/cycles/render/shader.cpp
@@ -569,6 +569,9 @@ void ShaderManager::get_requested_graph_features(ShaderGraph *graph,
 			if(CLOSURE_IS_VOLUME(bsdf_node->closure)) {
 				requested_features->nodes_features |= NODE_FEATURE_VOLUME;
 			}
+			else if(CLOSURE_IS_PRINCIPLED(bsdf_node->closure)) {
+				requested_features->use_principled = true;
+			}
 		}
 		if(node->has_surface_bssrdf()) {
 			requested_features->use_subsurface = true;