Octahedral Normal/Tangent Compression

Implementation of Octahedral normal compression into Godot 4.0

4 files changed, 54 insertions, 42 deletions

diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl
index e9515c7670..6b4e4a5a16 100644
--- a/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl
+++ b/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl
@@ -15,11 +15,11 @@ layout(location = 0) in vec3 vertex_attrib;
 //only for pure render depth when normal is not used
 
 #ifdef NORMAL_USED
-layout(location = 1) in vec3 normal_attrib;
+layout(location = 1) in vec2 normal_attrib;
 #endif
 
 #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED)
-layout(location = 2) in vec4 tangent_attrib;
+layout(location = 2) in vec2 tangent_attrib;
 #endif
 
 #if defined(COLOR_USED)
@@ -58,6 +58,13 @@ layout(location = 10) in uvec4 bone_attrib;
 layout(location = 11) in vec4 weight_attrib;
 #endif
 
+vec3 oct_to_vec3(vec2 e) {
+	vec3 v = vec3(e.xy, 1.0 - abs(e.x) - abs(e.y));
+	float t = max(-v.z, 0.0);
+	v.xy += t * -sign(v.xy);
+	return v;
+}
+
 /* Varyings */
 
 layout(location = 0) out vec3 vertex_interp;
@@ -231,12 +238,13 @@ void vertex_shader(in uint instance_index, in bool is_multimesh, in SceneData sc
 
 	vec3 vertex = vertex_attrib;
 #ifdef NORMAL_USED
-	vec3 normal = normal_attrib * 2.0 - 1.0;
+	vec3 normal = oct_to_vec3(normal_attrib * 2.0 - 1.0);
 #endif
 
 #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED)
-	vec3 tangent = tangent_attrib.xyz * 2.0 - 1.0;
-	float binormalf = tangent_attrib.a * 2.0 - 1.0;
+	vec2 signed_tangent_attrib = tangent_attrib * 2.0 - 1.0;
+	vec3 tangent = oct_to_vec3(vec2(signed_tangent_attrib.x, abs(signed_tangent_attrib.y) * 2.0 - 1.0));
+	float binormalf = sign(signed_tangent_attrib.y);
 	vec3 binormal = normalize(cross(normal, tangent) * binormalf);
 #endif
 
diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl
index 6548793bee..0960533917 100644
--- a/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl
+++ b/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl
@@ -16,11 +16,11 @@ layout(location = 0) in vec3 vertex_attrib;
 //only for pure render depth when normal is not used
 
 #ifdef NORMAL_USED
-layout(location = 1) in vec3 normal_attrib;
+layout(location = 1) in vec2 normal_attrib;
 #endif
 
 #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED)
-layout(location = 2) in vec4 tangent_attrib;
+layout(location = 2) in vec2 tangent_attrib;
 #endif
 
 #if defined(COLOR_USED)
@@ -59,6 +59,13 @@ layout(location = 10) in uvec4 bone_attrib;
 layout(location = 11) in vec4 weight_attrib;
 #endif
 
+vec3 oct_to_vec3(vec2 e) {
+	vec3 v = vec3(e.xy, 1.0 - abs(e.x) - abs(e.y));
+	float t = max(-v.z, 0.0);
+	v.xy += t * -sign(v.xy);
+	return v;
+}
+
 /* Varyings */
 
 layout(location = 0) highp out vec3 vertex_interp;
@@ -229,12 +236,13 @@ void main() {
 
 	vec3 vertex = vertex_attrib;
 #ifdef NORMAL_USED
-	vec3 normal = normal_attrib * 2.0 - 1.0;
+	vec3 normal = oct_to_vec3(normal_attrib * 2.0 - 1.0);
 #endif
 
 #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED)
-	vec3 tangent = tangent_attrib.xyz * 2.0 - 1.0;
-	float binormalf = tangent_attrib.a * 2.0 - 1.0;
+	vec3 signed_tangent_attrib = tangent_attrib * 2.0 - 1.0;
+	vec3 tangent = oct_to_vec3(vec2(signed_tangent_attrib.x, abs(signed_tangent_attrib.y) * 2.0 - 1.0));
+	float binormalf = sign(signed_tangent_attrib.y);
 	vec3 binormal = normalize(cross(normal, tangent) * binormalf);
 #endif
 
diff --git a/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp b/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
index dc3f35f942..71cfb32828 100644
--- a/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
+++ b/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
@@ -1057,10 +1057,9 @@ void MeshStorage::_mesh_surface_generate_version_for_input_mask(Mesh::Surface::V
 				} break;
 				case RS::ARRAY_NORMAL: {
 					vd.offset = stride;
+					vd.format = RD::DATA_FORMAT_R16G16_UNORM;
+					stride += sizeof(uint16_t) * 2;
 
-					vd.format = RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32;
-
-					stride += sizeof(uint32_t);
 					if (mis) {
 						buffer = mis->vertex_buffer;
 					} else {
@@ -1069,9 +1068,9 @@ void MeshStorage::_mesh_surface_generate_version_for_input_mask(Mesh::Surface::V
 				} break;
 				case RS::ARRAY_TANGENT: {
 					vd.offset = stride;
+					vd.format = RD::DATA_FORMAT_R16G16_UNORM;
+					stride += sizeof(uint16_t) * 2;
 
-					vd.format = RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32;
-					stride += sizeof(uint32_t);
 					if (mis) {
 						buffer = mis->vertex_buffer;
 					} else {
diff --git a/servers/rendering_server.cpp b/servers/rendering_server.cpp
index a54d4f0384..f79c4ac54c 100644
--- a/servers/rendering_server.cpp
+++ b/servers/rendering_server.cpp
@@ -398,16 +398,14 @@ Error RenderingServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint
 
 				const Vector3 *src = array.ptr();
 				for (int i = 0; i < p_vertex_array_len; i++) {
-					Vector3 n = src[i] * Vector3(0.5, 0.5, 0.5) + Vector3(0.5, 0.5, 0.5);
-
-					uint32_t value = 0;
-					value |= CLAMP(int(n.x * 1023.0), 0, 1023);
-					value |= CLAMP(int(n.y * 1023.0), 0, 1023) << 10;
-					value |= CLAMP(int(n.z * 1023.0), 0, 1023) << 20;
+					Vector2 res = src[i].octahedron_encode();
+					int16_t vector[2] = {
+						(int16_t)CLAMP(res.x * 65535, 0, 65535),
+						(int16_t)CLAMP(res.y * 65535, 0, 65535),
+					};
 
-					memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], &value, 4);
+					memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], vector, 4);
 				}
-
 			} break;
 
 			case RS::ARRAY_TANGENT: {
@@ -416,33 +414,32 @@ Error RenderingServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint
 				if (type == Variant::PACKED_FLOAT32_ARRAY) {
 					Vector<float> array = p_arrays[ai];
 					ERR_FAIL_COND_V(array.size() != p_vertex_array_len * 4, ERR_INVALID_PARAMETER);
-					const float *src = array.ptr();
+					const float *src_ptr = array.ptr();
 
 					for (int i = 0; i < p_vertex_array_len; i++) {
-						uint32_t value = 0;
-						value |= CLAMP(int((src[i * 4 + 0] * 0.5 + 0.5) * 1023.0), 0, 1023);
-						value |= CLAMP(int((src[i * 4 + 1] * 0.5 + 0.5) * 1023.0), 0, 1023) << 10;
-						value |= CLAMP(int((src[i * 4 + 2] * 0.5 + 0.5) * 1023.0), 0, 1023) << 20;
-						if (src[i * 4 + 3] > 0) {
-							value |= 3UL << 30;
-						}
-
-						memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], &value, 4);
+						const Vector3 src(src_ptr[i * 4 + 0], src_ptr[i * 4 + 1], src_ptr[i * 4 + 2]);
+						Vector2 res = src.octahedron_tangent_encode(src_ptr[i * 4 + 3]);
+						int16_t vector[2] = {
+							(int16_t)CLAMP(res.x * 65535, 0, 65535),
+							(int16_t)CLAMP(res.y * 65535, 0, 65535),
+						};
+
+						memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], vector, 4);
 					}
 				} else { // PACKED_FLOAT64_ARRAY
 					Vector<double> array = p_arrays[ai];
 					ERR_FAIL_COND_V(array.size() != p_vertex_array_len * 4, ERR_INVALID_PARAMETER);
-					const double *src = array.ptr();
+					const double *src_ptr = array.ptr();
 
 					for (int i = 0; i < p_vertex_array_len; i++) {
-						uint32_t value = 0;
-						value |= CLAMP(int((src[i * 4 + 0] * 0.5 + 0.5) * 1023.0), 0, 1023);
-						value |= CLAMP(int((src[i * 4 + 1] * 0.5 + 0.5) * 1023.0), 0, 1023) << 10;
-						value |= CLAMP(int((src[i * 4 + 2] * 0.5 + 0.5) * 1023.0), 0, 1023) << 20;
-						if (src[i * 4 + 3] > 0) {
-							value |= 3UL << 30;
-						}
-						memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], &value, 4);
+						const Vector3 src(src_ptr[i * 4 + 0], src_ptr[i * 4 + 1], src_ptr[i * 4 + 2]);
+						Vector2 res = src.octahedron_tangent_encode(src_ptr[i * 4 + 3]);
+						int16_t vector[2] = {
+							(int16_t)CLAMP(res.x * 65535, 0, 65535),
+							(int16_t)CLAMP(res.y * 65535, 0, 65535),
+						};
+
+						memcpy(&vw[p_offsets[ai] + i * p_vertex_stride], vector, 4);
 					}
 				}
 			} break;