Cycles: Calculate face normal on the fly.

Instead of pre-calculation and storage, we now calculate the face normal during render.
This gives a small slowdown (~1%) but decreases memory usage, which is especially important for GPUs,
where you have limited VRAM.

Part of my GSoC 2014.

8 files changed, 37 insertions, 37 deletions

diff --git a/intern/cycles/kernel/geom/geom_bvh_shadow.h b/intern/cycles/kernel/geom/geom_bvh_shadow.h
index 48876da049e..1f6e4942fab 100644
--- a/intern/cycles/kernel/geom/geom_bvh_shadow.h
+++ b/intern/cycles/kernel/geom/geom_bvh_shadow.h
@@ -252,8 +252,7 @@ ccl_device bool BVH_FUNCTION_NAME
 							if(kernel_tex_fetch(__prim_type, isect_array->prim) & PRIMITIVE_ALL_TRIANGLE)
 #endif
 							{
-								float4 Ns = kernel_tex_fetch(__tri_normal, prim);
-								shader = __float_as_int(Ns.w);
+								shader =  __float_as_int(kernel_tex_fetch(__tri_shader, prim));
 							}
 #ifdef __HAIR__
 							else {
diff --git a/intern/cycles/kernel/geom/geom_motion_triangle.h b/intern/cycles/kernel/geom/geom_motion_triangle.h
index 73fcf1adda6..5ab0b731bdd 100644
--- a/intern/cycles/kernel/geom/geom_motion_triangle.h
+++ b/intern/cycles/kernel/geom/geom_motion_triangle.h
@@ -233,8 +233,7 @@ ccl_device_inline float3 motion_triangle_refine_subsurface(KernelGlobals *kg, Sh
 ccl_device_noinline void motion_triangle_shader_setup(KernelGlobals *kg, ShaderData *sd, const Intersection *isect, const Ray *ray, bool subsurface)
 {
 	/* get shader */
-	float4 Ns = kernel_tex_fetch(__tri_normal, sd->prim);
-	sd->shader = __float_as_int(Ns.w);
+	sd->shader =  __float_as_int(kernel_tex_fetch(__tri_shader, sd->prim));
 
 	/* get motion info */
 	int numsteps, numverts;
diff --git a/intern/cycles/kernel/geom/geom_triangle.h b/intern/cycles/kernel/geom/geom_triangle.h
index 355e36fef0c..4ee5d5aa6d5 100644
--- a/intern/cycles/kernel/geom/geom_triangle.h
+++ b/intern/cycles/kernel/geom/geom_triangle.h
@@ -116,6 +116,20 @@ ccl_device_inline float3 triangle_refine_subsurface(KernelGlobals *kg, ShaderDat
 #endif
 }
 
+/* normal on triangle  */
+ccl_device_inline float3 triangle_normal(KernelGlobals *kg, int prim)
+{
+	/* load triangle vertices */
+	float3 tri_vindex = float4_to_float3(kernel_tex_fetch(__tri_vindex, prim));
+
+	float3 v0 = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.x)));
+	float3 v1 = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.y)));
+	float3 v2 = float4_to_float3(kernel_tex_fetch(__tri_verts, __float_as_int(tri_vindex.z)));
+	
+	/* return normal */
+	return normalize(cross(v1 - v0, v2 - v0));
+}
+
 /* point and normal on triangle  */
 ccl_device_inline void triangle_point_normal(KernelGlobals *kg, int prim, float u, float v, float3 *P, float3 *Ng, int *shader)
 {
@@ -130,9 +144,11 @@ ccl_device_inline void triangle_point_normal(KernelGlobals *kg, int prim, float
 	float t = 1.0f - u - v;
 	*P = (u*v0 + v*v1 + t*v2);
 
-	float4 Nm = kernel_tex_fetch(__tri_normal, prim);
-	*Ng = make_float3(Nm.x, Nm.y, Nm.z);
-	*shader = __float_as_int(Nm.w);
+	/* compute normal */
+	*Ng = normalize(cross(v1 - v0, v2 - v0));
+
+	/* shader`*/
+	*shader = __float_as_int(kernel_tex_fetch(__tri_shader, prim));
 }
 
 /* Triangle vertex locations */
diff --git a/intern/cycles/kernel/kernel_shader.h b/intern/cycles/kernel/kernel_shader.h
index c6a7f9f1570..0ea21a85fcb 100644
--- a/intern/cycles/kernel/kernel_shader.h
+++ b/intern/cycles/kernel/kernel_shader.h
@@ -86,9 +86,8 @@ ccl_device void shader_setup_from_ray(KernelGlobals *kg, ShaderData *sd,
 #endif
 	if(sd->type & PRIMITIVE_TRIANGLE) {
 		/* static triangle */
-		float4 Ns = kernel_tex_fetch(__tri_normal, sd->prim);
-		float3 Ng = make_float3(Ns.x, Ns.y, Ns.z);
-		sd->shader = __float_as_int(Ns.w);
+		float3 Ng = triangle_normal(kg, sd->prim);
+		sd->shader =  __float_as_int(kernel_tex_fetch(__tri_shader, sd->prim));
 
 		/* vectors */
 		sd->P = triangle_refine(kg, sd, isect, ray);
@@ -166,9 +165,8 @@ ccl_device_inline void shader_setup_from_subsurface(KernelGlobals *kg, ShaderDat
 
 	/* fetch triangle data */
 	if(sd->type == PRIMITIVE_TRIANGLE) {
-		float4 Ns = kernel_tex_fetch(__tri_normal, sd->prim);
-		float3 Ng = make_float3(Ns.x, Ns.y, Ns.z);
-		sd->shader = __float_as_int(Ns.w);
+		float3 Ng = triangle_normal(kg, sd->prim);
+		sd->shader =  __float_as_int(kernel_tex_fetch(__tri_shader, sd->prim));
 
 		/* static triangle */
 		sd->P = triangle_refine_subsurface(kg, sd, isect, ray);
@@ -1028,8 +1026,7 @@ ccl_device bool shader_transparent_shadow(KernelGlobals *kg, Intersection *isect
 #ifdef __HAIR__
 	if(kernel_tex_fetch(__prim_type, isect->prim) & PRIMITIVE_ALL_TRIANGLE) {
 #endif
-		float4 Ns = kernel_tex_fetch(__tri_normal, prim);
-		shader = __float_as_int(Ns.w);
+		shader = __float_as_int(kernel_tex_fetch(__tri_shader, prim));
 #ifdef __HAIR__
 	}
 	else {
diff --git a/intern/cycles/kernel/kernel_textures.h b/intern/cycles/kernel/kernel_textures.h
index b07075c6c95..dfef8ed9c66 100644
--- a/intern/cycles/kernel/kernel_textures.h
+++ b/intern/cycles/kernel/kernel_textures.h
@@ -36,7 +36,7 @@ KERNEL_TEX(float4, texture_float4, __objects)
 KERNEL_TEX(float4, texture_float4, __objects_vector)
 
 /* triangles */
-KERNEL_TEX(float4, texture_float4, __tri_normal)
+KERNEL_TEX(float, texture_float, __tri_shader)
 KERNEL_TEX(float4, texture_float4, __tri_vnormal)
 KERNEL_TEX(float4, texture_float4, __tri_vindex)
 KERNEL_TEX(float4, texture_float4, __tri_verts)
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 9c5ddd55010..b6826f0066b 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -377,12 +377,10 @@ void Mesh::add_vertex_normals()
 	}
 }
 
-void Mesh::pack_normals(Scene *scene, float4 *normal, float4 *vnormal)
+void Mesh::pack_normals(Scene *scene, float *tri_shader, float4 *vnormal)
 {
-	Attribute *attr_fN = attributes.find(ATTR_STD_FACE_NORMAL);
 	Attribute *attr_vN = attributes.find(ATTR_STD_VERTEX_NORMAL);
 
-	float3 *fN = attr_fN->data_float3();
 	float3 *vN = attr_vN->data_float3();
 	int shader_id = 0;
 	uint last_shader = -1;
@@ -394,24 +392,15 @@ void Mesh::pack_normals(Scene *scene, float4 *normal, float4 *vnormal)
 	bool do_transform = transform_applied;
 	Transform ntfm = transform_normal;
 
+	/* save shader */
 	for(size_t i = 0; i < triangles_size; i++) {
-		float3 fNi = fN[i];
-
-		if(do_transform)
-			fNi = normalize(transform_direction(&ntfm, fNi));
-
-		normal[i].x = fNi.x;
-		normal[i].y = fNi.y;
-		normal[i].z = fNi.z;
-
-		/* stuff shader id in here too */
 		if(shader_ptr[i] != last_shader || last_smooth != smooth[i]) {
 			last_shader = shader_ptr[i];
 			last_smooth = smooth[i];
 			shader_id = scene->shader_manager->get_shader_id(last_shader, this, last_smooth);
 		}
 
-		normal[i].w = __int_as_float(shader_id);
+		tri_shader[i] = __int_as_float(shader_id);
 	}
 
 	size_t verts_size = verts.size();
@@ -932,13 +921,13 @@ void MeshManager::device_update_mesh(Device *device, DeviceScene *dscene, Scene
 		/* normals */
 		progress.set_status("Updating Mesh", "Computing normals");
 
-		float4 *normal = dscene->tri_normal.resize(tri_size);
+		float *tri_shader = dscene->tri_shader.resize(tri_size);
 		float4 *vnormal = dscene->tri_vnormal.resize(vert_size);
 		float4 *tri_verts = dscene->tri_verts.resize(vert_size);
 		float4 *tri_vindex = dscene->tri_vindex.resize(tri_size);
 
 		foreach(Mesh *mesh, scene->meshes) {
-			mesh->pack_normals(scene, &normal[mesh->tri_offset], &vnormal[mesh->vert_offset]);
+			mesh->pack_normals(scene, &tri_shader[mesh->tri_offset], &vnormal[mesh->vert_offset]);
 			mesh->pack_verts(&tri_verts[mesh->vert_offset], &tri_vindex[mesh->tri_offset], mesh->vert_offset);
 
 			if(progress.get_cancel()) return;
@@ -947,7 +936,7 @@ void MeshManager::device_update_mesh(Device *device, DeviceScene *dscene, Scene
 		/* vertex coordinates */
 		progress.set_status("Updating Mesh", "Copying Mesh to device");
 
-		device->tex_alloc("__tri_normal", dscene->tri_normal);
+		device->tex_alloc("__tri_shader", dscene->tri_shader);
 		device->tex_alloc("__tri_vnormal", dscene->tri_vnormal);
 		device->tex_alloc("__tri_verts", dscene->tri_verts);
 		device->tex_alloc("__tri_vindex", dscene->tri_vindex);
@@ -1119,7 +1108,7 @@ void MeshManager::device_free(Device *device, DeviceScene *dscene)
 	device->tex_free(dscene->prim_visibility);
 	device->tex_free(dscene->prim_index);
 	device->tex_free(dscene->prim_object);
-	device->tex_free(dscene->tri_normal);
+	device->tex_free(dscene->tri_shader);
 	device->tex_free(dscene->tri_vnormal);
 	device->tex_free(dscene->tri_vindex);
 	device->tex_free(dscene->tri_verts);
@@ -1136,7 +1125,7 @@ void MeshManager::device_free(Device *device, DeviceScene *dscene)
 	dscene->prim_visibility.clear();
 	dscene->prim_index.clear();
 	dscene->prim_object.clear();
-	dscene->tri_normal.clear();
+	dscene->tri_shader.clear();
 	dscene->tri_vnormal.clear();
 	dscene->tri_vindex.clear();
 	dscene->tri_verts.clear();
diff --git a/intern/cycles/render/mesh.h b/intern/cycles/render/mesh.h
index 247e3dd555e..5ee774bacc1 100644
--- a/intern/cycles/render/mesh.h
+++ b/intern/cycles/render/mesh.h
@@ -120,7 +120,7 @@ public:
 	void add_face_normals();
 	void add_vertex_normals();
 
-	void pack_normals(Scene *scene, float4 *normal, float4 *vnormal);
+	void pack_normals(Scene *scene, float *shader, float4 *vnormal);
 	void pack_verts(float4 *tri_verts, float4 *tri_vindex, size_t vert_offset);
 	void pack_curves(Scene *scene, float4 *curve_key_co, float4 *curve_data, size_t curvekey_offset);
 	void compute_bvh(SceneParams *params, Progress *progress, int n, int total);
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index d657e48e478..777d26ac79e 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -69,7 +69,7 @@ public:
 	device_vector<uint> prim_object;
 
 	/* mesh */
-	device_vector<float4> tri_normal;
+	device_vector<float> tri_shader;
 	device_vector<float4> tri_vnormal;
 	device_vector<float4> tri_vindex;
 	device_vector<float4> tri_verts;