Code refactor: make Transform always affine, dropping last row.

This save a little memory and copying in the kernel by storing only a 4x3
matrix instead of a 4x4 matrix. We already did this in a few places, and
those don't need to be special exceptions anymore now.

7 files changed, 32 insertions, 38 deletions

diff --git a/intern/cycles/render/light.cpp b/intern/cycles/render/light.cpp
index a3f67361f6e..8dec7e4ea64 100644
--- a/intern/cycles/render/light.cpp
+++ b/intern/cycles/render/light.cpp
@@ -785,11 +785,8 @@ void LightManager::device_update_points(Device *,
 		klights[light_index].max_bounces = max_bounces;
 		klights[light_index].random = random;
 
-
-		Transform tfm = light->tfm;
-		Transform itfm = transform_inverse(tfm);
-		memcpy(&klights[light_index].tfm, &tfm, sizeof(float4)*3);
-		memcpy(&klights[light_index].itfm, &itfm, sizeof(float4)*3);
+		klights[light_index].tfm = light->tfm;
+		klights[light_index].itfm = transform_inverse(light->tfm);
 
 		light_index++;
 	}
@@ -825,10 +822,8 @@ void LightManager::device_update_points(Device *,
 		klights[light_index].area.dir[0] = dir.x;
 		klights[light_index].area.dir[1] = dir.y;
 		klights[light_index].area.dir[2] = dir.z;
-		Transform tfm = light->tfm;
-		Transform itfm = transform_inverse(tfm);
-		memcpy(&klights[light_index].tfm, &tfm, sizeof(float4)*3);
-		memcpy(&klights[light_index].itfm, &itfm, sizeof(float4)*3);
+		klights[light_index].tfm = light->tfm;
+		klights[light_index].itfm = transform_inverse(light->tfm);
 
 		light_index++;
 	}
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index 47d24970949..71cbf5c24b6 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -1445,11 +1445,11 @@ static void update_attribute_element_offset(Mesh *mesh,
 			Transform *tfm = mattr->data_transform();
 			offset = attr_float3_offset;
 
-			assert(attr_float3.size() >= offset + size * 4);
-			for(size_t k = 0; k < size*4; k++) {
+			assert(attr_float3.size() >= offset + size * 3);
+			for(size_t k = 0; k < size*3; k++) {
 				attr_float3[offset+k] = (&tfm->x)[k];
 			}
-			attr_float3_offset += size * 4;
+			attr_float3_offset += size * 3;
 		}
 		else {
 			float4 *data = mattr->data_float4();
diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp
index 41c9730e6fb..f117962a2ea 100644
--- a/intern/cycles/render/nodes.cpp
+++ b/intern/cycles/render/nodes.cpp
@@ -117,8 +117,7 @@ Transform TextureMapping::compute_transform()
 		case NORMAL:
 			/* no translation for normals, and inverse transpose */
 			mat = rmat*smat;
-			mat = transform_inverse(mat);
-			mat = transform_transpose(mat);
+			mat = transform_transposed_inverse(mat);
 			break;
 	}
 
@@ -153,7 +152,6 @@ void TextureMapping::compile(SVMCompiler& compiler, int offset_in, int offset_ou
 	compiler.add_node(tfm.x);
 	compiler.add_node(tfm.y);
 	compiler.add_node(tfm.z);
-	compiler.add_node(tfm.w);
 
 	if(use_minmax) {
 		compiler.add_node(NODE_MIN_MAX, offset_out, offset_out);
@@ -193,9 +191,7 @@ void TextureMapping::compile_end(SVMCompiler& compiler, ShaderInput *vector_in,
 void TextureMapping::compile(OSLCompiler &compiler)
 {
 	if(!skip()) {
-		Transform tfm = transform_transpose(compute_transform());
-
-		compiler.parameter("mapping", tfm);
+		compiler.parameter("mapping", compute_transform());
 		compiler.parameter("use_mapping", 1);
 	}
 }
@@ -1434,7 +1430,6 @@ void PointDensityTextureNode::compile(SVMCompiler& compiler)
 				compiler.add_node(tfm.x);
 				compiler.add_node(tfm.y);
 				compiler.add_node(tfm.z);
-				compiler.add_node(tfm.w);
 			}
 		}
 		else {
@@ -1478,7 +1473,7 @@ void PointDensityTextureNode::compile(OSLCompiler& compiler)
 			compiler.parameter("filename", string_printf("@%d", slot).c_str());
 		}
 		if(space == NODE_TEX_VOXEL_SPACE_WORLD) {
-			compiler.parameter("mapping", transform_transpose(tfm));
+			compiler.parameter("mapping", tfm);
 			compiler.parameter("use_mapping", 1);
 		}
 		compiler.parameter(this, "interpolation");
@@ -1558,8 +1553,7 @@ void MappingNode::compile(SVMCompiler& compiler)
 
 void MappingNode::compile(OSLCompiler& compiler)
 {
-	Transform tfm = transform_transpose(tex_mapping.compute_transform());
-	compiler.parameter("Matrix", tfm);
+	compiler.parameter("Matrix", tex_mapping.compute_transform());
 	compiler.parameter_point("mapping_min", tex_mapping.min);
 	compiler.parameter_point("mapping_max", tex_mapping.max);
 	compiler.parameter("use_minmax", tex_mapping.use_minmax);
@@ -3220,7 +3214,6 @@ void TextureCoordinateNode::compile(SVMCompiler& compiler)
 			compiler.add_node(ob_itfm.x);
 			compiler.add_node(ob_itfm.y);
 			compiler.add_node(ob_itfm.z);
-			compiler.add_node(ob_itfm.w);
 		}
 	}
 
@@ -3259,7 +3252,7 @@ void TextureCoordinateNode::compile(OSLCompiler& compiler)
 	if(compiler.output_type() == SHADER_TYPE_VOLUME)
 		compiler.parameter("is_volume", true);
 	compiler.parameter(this, "use_transform");
-	Transform ob_itfm = transform_transpose(transform_inverse(ob_tfm));
+	Transform ob_itfm = transform_transposed_inverse(ob_tfm);
 	compiler.parameter("object_itfm", ob_itfm);
 
 	compiler.parameter(this, "from_dupli");
diff --git a/intern/cycles/render/object.cpp b/intern/cycles/render/object.cpp
index f632f7c4a53..92c7e9cf1b8 100644
--- a/intern/cycles/render/object.cpp
+++ b/intern/cycles/render/object.cpp
@@ -132,7 +132,7 @@ void Object::apply_transform(bool apply_to_motion)
 		/* store matrix to transform later. when accessing these as attributes we
 		 * do not want the transform to be applied for consistency between static
 		 * and dynamic BVH, so we do it on packing. */
-		mesh->transform_normal = transform_transpose(transform_inverse(tfm));
+		mesh->transform_normal = transform_transposed_inverse(tfm);
 
 		/* apply to mesh vertices */
 		for(size_t i = 0; i < mesh->verts.size(); i++)
@@ -290,7 +290,7 @@ void ObjectManager::device_update_object_transform(UpdateObjectTransformState *s
                                                    int object_index)
 {
 	KernelObject& kobject = state->objects[object_index];
-	float4 *objects_vector = state->objects_vector;
+	Transform *objects_vector = state->objects_vector;
 
 	Mesh *mesh = ob->mesh;
 	uint flag = 0;
@@ -357,11 +357,8 @@ void ObjectManager::device_update_object_transform(UpdateObjectTransformState *s
 		}
 	}
 
-	/* OBJECT_TRANSFORM */
-	memcpy(&kobject.tfm.pre, &tfm, sizeof(float4)*3);
-	/* OBJECT_INVERSE_TRANSFORM */
-	memcpy(&kobject.tfm.mid, &itfm, sizeof(float4)*3);
-	/* OBJECT_PROPERTIES */
+	memcpy(&kobject.tfm.pre, &tfm, sizeof(tfm));
+	memcpy(&kobject.tfm.mid, &itfm, sizeof(itfm));
 	kobject.surface_area = surface_area;
 	kobject.pass_id = pass_id;
 	kobject.random_number = random_number;
@@ -395,8 +392,8 @@ void ObjectManager::device_update_object_transform(UpdateObjectTransformState *s
 			mtfm.post = mtfm.post * itfm;
 		}
 
-		memcpy(&objects_vector[object_index*OBJECT_VECTOR_SIZE+0], &mtfm.pre, sizeof(float4)*3);
-		memcpy(&objects_vector[object_index*OBJECT_VECTOR_SIZE+3], &mtfm.post, sizeof(float4)*3);
+		objects_vector[object_index*OBJECT_VECTOR_SIZE+0] = mtfm.pre;
+		objects_vector[object_index*OBJECT_VECTOR_SIZE+1] = mtfm.post;
 	}
 	else if(state->need_motion == Scene::MOTION_BLUR) {
 		if(ob->use_motion) {
diff --git a/intern/cycles/render/object.h b/intern/cycles/render/object.h
index 54046d7069f..59c1d916595 100644
--- a/intern/cycles/render/object.h
+++ b/intern/cycles/render/object.h
@@ -135,7 +135,7 @@ protected:
 		/* Packed object arrays. Those will be filled in. */
 		uint *object_flag;
 		KernelObject *objects;
-		float4 *objects_vector;
+		Transform *objects_vector;
 
 		/* Flags which will be synchronized to Integrator. */
 		bool have_motion;
diff --git a/intern/cycles/render/osl.cpp b/intern/cycles/render/osl.cpp
index 9e931280691..f1a22350060 100644
--- a/intern/cycles/render/osl.cpp
+++ b/intern/cycles/render/osl.cpp
@@ -34,6 +34,7 @@
 #include "util/util_md5.h"
 #include "util/util_path.h"
 #include "util/util_progress.h"
+#include "util/util_projection.h"
 
 #endif
 
@@ -832,7 +833,9 @@ void OSLCompiler::parameter(ShaderNode* node, const char *name)
 		case SocketType::TRANSFORM:
 		{
 			Transform value = node->get_transform(socket);
-			ss->Parameter(uname, TypeDesc::TypeMatrix, &value);
+			ProjectionTransform projection(value);
+			projection = projection_transpose(projection);
+			ss->Parameter(uname, TypeDesc::TypeMatrix, &projection);
 			break;
 		}
 		case SocketType::BOOLEAN_ARRAY:
@@ -900,7 +903,11 @@ void OSLCompiler::parameter(ShaderNode* node, const char *name)
 		case SocketType::TRANSFORM_ARRAY:
 		{
 			const array<Transform>& value = node->get_transform_array(socket);
-			ss->Parameter(uname, array_typedesc(TypeDesc::TypeMatrix, value.size()), value.data());
+			array<ProjectionTransform> fvalue(value.size());
+			for(size_t i = 0; i < value.size(); i++) {
+				fvalue[i] = projection_transpose(ProjectionTransform(value[i]));
+			}
+			ss->Parameter(uname, array_typedesc(TypeDesc::TypeMatrix, fvalue.size()), fvalue.data());
 			break;
 		}
 		case SocketType::CLOSURE:
@@ -967,7 +974,9 @@ void OSLCompiler::parameter(const char *name, ustring s)
 void OSLCompiler::parameter(const char *name, const Transform& tfm)
 {
 	OSL::ShadingSystem *ss = (OSL::ShadingSystem*)shadingsys;
-	ss->Parameter(name, TypeDesc::TypeMatrix, (float*)&tfm);
+	ProjectionTransform projection(tfm);
+	projection = projection_transpose(projection);
+	ss->Parameter(name, TypeDesc::TypeMatrix, (float*)&projection);
 }
 
 void OSLCompiler::parameter_array(const char *name, const float f[], int arraylen)
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index 316ffeb1092..cd5c9c1dadc 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -87,7 +87,7 @@ public:
 
 	/* objects */
 	device_vector<KernelObject> objects;
-	device_vector<float4> objects_vector;
+	device_vector<Transform> objects_vector;
 
 	/* attributes */
 	device_vector<uint4> attributes_map;