Cycles: Add support for uchar4 attributes.

* Added support for uchar4 attributes to Cycles' attribute system.
* This is used for Vertex Colors now, which saves some memory (4 unsigned characters, instead of 4 floats).
* GPU Texture Limit on sm_20 and sm_21 decreased from 95 to 94, because we need a new texture for the uchar4 attributes. This is no problem for sm_30 or newer.

Part of my GSoC 2014.

5 files changed, 32 insertions, 5 deletions

diff --git a/intern/cycles/render/attribute.cpp b/intern/cycles/render/attribute.cpp
index 72781bb0f9b..8abf869a775 100644
--- a/intern/cycles/render/attribute.cpp
+++ b/intern/cycles/render/attribute.cpp
@@ -69,6 +69,15 @@ void Attribute::add(const float& f)
 		buffer.push_back(data[i]);
 }
 
+void Attribute::add(const uchar4& f)
+{
+	char *data = (char*)&f;
+	size_t size = sizeof(f);
+
+	for(size_t i = 0; i < size; i++)
+		buffer.push_back(data[i]);
+}
+
 void Attribute::add(const float3& f)
 {
 	char *data = (char*)&f;
@@ -136,6 +145,7 @@ size_t Attribute::element_size(int numverts, int numtris, int numsteps, int numc
 			size = numtris;
 			break;
 		case ATTR_ELEMENT_CORNER:
+		case ATTR_ELEMENT_CORNER_BYTE:
 			size = numtris*3;
 			break;
 		case ATTR_ELEMENT_CURVE:
diff --git a/intern/cycles/render/attribute.h b/intern/cycles/render/attribute.h
index 9fc32db8444..f5227ebde52 100644
--- a/intern/cycles/render/attribute.h
+++ b/intern/cycles/render/attribute.h
@@ -68,6 +68,7 @@ public:
 	float3 *data_float3() { return (float3*)data(); }
 	float4 *data_float4() { return (float4*)data(); }
 	float *data_float() { return (float*)data(); }
+	uchar4 *data_uchar4() { return (uchar4*)data(); }
 	Transform *data_transform() { return (Transform*)data(); }
 	VoxelAttribute *data_voxel()  { return ( VoxelAttribute*)data(); }
 
@@ -80,6 +81,7 @@ public:
 
 	void add(const float& f);
 	void add(const float3& f);
+	void add(const uchar4& f);
 	void add(const Transform& f);
 	void add(const VoxelAttribute& f);
 	void add(const char *data);
diff --git a/intern/cycles/render/image.h b/intern/cycles/render/image.h
index 8abf2a33d74..3eacfa7beed 100644
--- a/intern/cycles/render/image.h
+++ b/intern/cycles/render/image.h
@@ -29,7 +29,7 @@
 CCL_NAMESPACE_BEGIN
 
 /* generic */
-#define TEX_NUM_IMAGES			95
+#define TEX_NUM_IMAGES			94
 #define TEX_IMAGE_BYTE_START	TEX_NUM_FLOAT_IMAGES
 
 /* extended gpu */
diff --git a/intern/cycles/render/mesh.cpp b/intern/cycles/render/mesh.cpp
index b6826f0066b..808db4fe5b2 100644
--- a/intern/cycles/render/mesh.cpp
+++ b/intern/cycles/render/mesh.cpp
@@ -745,7 +745,7 @@ void MeshManager::update_svm_attributes(Device *device, DeviceScene *dscene, Sce
 	device->tex_alloc("__attributes_map", dscene->attributes_map);
 }
 
-static void update_attribute_element_offset(Mesh *mesh, vector<float>& attr_float, vector<float4>& attr_float3,
+static void update_attribute_element_offset(Mesh *mesh, vector<float>& attr_float, vector<float4>& attr_float3, vector<uchar4>& attr_uchar4,
 	Attribute *mattr, TypeDesc& type, int& offset, AttributeElement& element)
 {
 	if(mattr) {
@@ -766,6 +766,15 @@ static void update_attribute_element_offset(Mesh *mesh, vector<float>& attr_floa
 			VoxelAttribute *voxel_data = mattr->data_voxel();
 			offset = voxel_data->slot;
 		}
+		if(mattr->element == ATTR_ELEMENT_CORNER_BYTE) {
+			uchar4 *data = mattr->data_uchar4();
+			offset = attr_uchar4.size();
+
+			attr_uchar4.resize(attr_uchar4.size() + size);
+
+			for(size_t k = 0; k < size; k++)
+				attr_uchar4[offset+k] = data[k];
+		}
 		else if(mattr->type == TypeDesc::TypeFloat) {
 			float *data = mattr->data_float();
 			offset = attr_float.size();
@@ -802,7 +811,7 @@ static void update_attribute_element_offset(Mesh *mesh, vector<float>& attr_floa
 			offset -= mesh->vert_offset;
 		else if(element == ATTR_ELEMENT_FACE)
 			offset -= mesh->tri_offset;
-		else if(element == ATTR_ELEMENT_CORNER)
+		else if(element == ATTR_ELEMENT_CORNER || element == ATTR_ELEMENT_CORNER_BYTE)
 			offset -= 3*mesh->tri_offset;
 		else if(element == ATTR_ELEMENT_CURVE)
 			offset -= mesh->curve_offset;
@@ -843,6 +852,7 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 	 * maps next */
 	vector<float> attr_float;
 	vector<float4> attr_float3;
+	vector<uchar4> attr_uchar4;
 
 	for(size_t i = 0; i < scene->meshes.size(); i++) {
 		Mesh *mesh = scene->meshes[i];
@@ -863,10 +873,10 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 					memcpy(triangle_mattr->data_float3(), &mesh->verts[0], sizeof(float3)*mesh->verts.size());
 			}
 
-			update_attribute_element_offset(mesh, attr_float, attr_float3, triangle_mattr,
+			update_attribute_element_offset(mesh, attr_float, attr_float3, attr_uchar4, triangle_mattr,
 				req.triangle_type, req.triangle_offset, req.triangle_element);
 
-			update_attribute_element_offset(mesh, attr_float, attr_float3, curve_mattr,
+			update_attribute_element_offset(mesh, attr_float, attr_float3, attr_uchar4, curve_mattr,
 				req.curve_type, req.curve_offset, req.curve_element);
 	
 			if(progress.get_cancel()) return;
@@ -892,6 +902,10 @@ void MeshManager::device_update_attributes(Device *device, DeviceScene *dscene,
 		dscene->attributes_float3.copy(&attr_float3[0], attr_float3.size());
 		device->tex_alloc("__attributes_float3", dscene->attributes_float3);
 	}
+	if(attr_uchar4.size()) {
+		dscene->attributes_uchar4.copy(&attr_uchar4[0], attr_uchar4.size());
+		device->tex_alloc("__attributes_uchar4", dscene->attributes_uchar4);
+	}
 }
 
 void MeshManager::device_update_mesh(Device *device, DeviceScene *dscene, Scene *scene, Progress& progress)
diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h
index 777d26ac79e..e5c7444c92d 100644
--- a/intern/cycles/render/scene.h
+++ b/intern/cycles/render/scene.h
@@ -85,6 +85,7 @@ public:
 	device_vector<uint4> attributes_map;
 	device_vector<float> attributes_float;
 	device_vector<float4> attributes_float3;
+	device_vector<uchar4> attributes_uchar4;
 
 	/* lights */
 	device_vector<float4> light_distribution;