Merge pull request #60309 from The-O-King/oct

5 files changed, 106 insertions, 54 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/shaders/skeleton.glsl b/servers/rendering/renderer_rd/shaders/skeleton.glsl
index a893a66c94..75bea9300b 100644
--- a/servers/rendering/renderer_rd/shaders/skeleton.glsl
+++ b/servers/rendering/renderer_rd/shaders/skeleton.glsl
@@ -54,14 +54,54 @@ layout(push_constant, std430) uniform Params {
 }
 params;
 
-vec4 decode_abgr_2_10_10_10(uint base) {
-	uvec4 abgr_2_10_10_10 = (uvec4(base) >> uvec4(0, 10, 20, 30)) & uvec4(0x3FF, 0x3FF, 0x3FF, 0x3);
-	return vec4(abgr_2_10_10_10) / vec4(1023.0, 1023.0, 1023.0, 3.0) * 2.0 - 1.0;
+vec2 uint_to_vec2(uint base) {
+	uvec2 decode = (uvec2(base) >> uvec2(0, 16)) & uvec2(0xFFFF, 0xFFFF);
+	return vec2(decode) / vec2(65535.0, 65535.0) * 2.0 - 1.0;
 }
 
-uint encode_abgr_2_10_10_10(vec4 base) {
-	uvec4 abgr_2_10_10_10 = uvec4(clamp(ivec4((base * 0.5 + 0.5) * vec4(1023.0, 1023.0, 1023.0, 3.0)), ivec4(0), ivec4(0x3FF, 0x3FF, 0x3FF, 0x3))) << uvec4(0, 10, 20, 30);
-	return abgr_2_10_10_10.x | abgr_2_10_10_10.y | abgr_2_10_10_10.z | abgr_2_10_10_10.w;
+vec3 oct_to_vec3(vec2 oct) {
+	vec3 v = vec3(oct.xy, 1.0 - abs(oct.x) - abs(oct.y));
+	float t = max(-v.z, 0.0);
+	v.xy += t * -sign(v.xy);
+	return v;
+}
+
+vec3 decode_uint_oct_to_norm(uint base) {
+	return oct_to_vec3(uint_to_vec2(base));
+}
+
+vec4 decode_uint_oct_to_tang(uint base) {
+	vec2 oct_sign_encoded = uint_to_vec2(base);
+	// Binormal sign encoded in y component
+	vec2 oct = vec2(oct_sign_encoded.x, abs(oct_sign_encoded.y) * 2.0 - 1.0);
+	return vec4(oct_to_vec3(oct), sign(oct_sign_encoded.y));
+}
+
+vec2 signNotZero(vec2 v) {
+	return mix(vec2(-1.0), vec2(1.0), greaterThanEqual(v.xy, vec2(0.0)));
+}
+
+uint vec2_to_uint(vec2 base) {
+	uvec2 enc = uvec2(clamp(ivec2(base * vec2(65535, 65535)), ivec2(0), ivec2(0xFFFF, 0xFFFF))) << uvec2(0, 16);
+	return enc.x | enc.y;
+}
+
+vec2 vec3_to_oct(vec3 e) {
+	e /= abs(e.x) + abs(e.y) + abs(e.z);
+	vec2 oct = e.z >= 0.0f ? e.xy : (vec2(1.0f) - abs(e.yx)) * signNotZero(e.xy);
+	return oct * 0.5f + 0.5f;
+}
+
+uint encode_norm_to_uint_oct(vec3 base) {
+	return vec2_to_uint(vec3_to_oct(base));
+}
+
+uint encode_tang_to_uint_oct(vec4 base) {
+	vec2 oct = vec3_to_oct(base.xyz);
+	// Encode binormal sign in y component
+	oct.y = oct.y * 0.5f + 0.5f;
+	oct.y = base.w >= 0.0f ? oct.y : 1 - oct.y;
+	return vec2_to_uint(oct);
 }
 
 void main() {
@@ -131,12 +171,12 @@ void main() {
 	src_offset += 3;
 
 	if (params.has_normal) {
-		normal = decode_abgr_2_10_10_10(src_vertices.data[src_offset]).rgb;
+		normal = decode_uint_oct_to_norm(src_vertices.data[src_offset]);
 		src_offset++;
 	}
 
 	if (params.has_tangent) {
-		tangent = decode_abgr_2_10_10_10(src_vertices.data[src_offset]);
+		tangent = decode_uint_oct_to_tang(src_vertices.data[src_offset]);
 	}
 
 	if (params.has_blend_shape) {
@@ -155,12 +195,12 @@ void main() {
 				base_offset += 3;
 
 				if (params.has_normal) {
-					blend_normal += decode_abgr_2_10_10_10(src_blend_shapes.data[base_offset]).rgb * w;
+					blend_normal += decode_uint_oct_to_norm(src_blend_shapes.data[base_offset]) * w;
 					base_offset++;
 				}
 
 				if (params.has_tangent) {
-					blend_tangent += decode_abgr_2_10_10_10(src_blend_shapes.data[base_offset]).rgb * w;
+					blend_tangent += decode_uint_oct_to_tang(src_blend_shapes.data[base_offset]).rgb * w;
 				}
 
 				blend_total += w;
@@ -234,12 +274,12 @@ void main() {
 	dst_offset += 3;
 
 	if (params.has_normal) {
-		dst_vertices.data[dst_offset] = encode_abgr_2_10_10_10(vec4(normal, 0.0));
+		dst_vertices.data[dst_offset] = encode_norm_to_uint_oct(normal);
 		dst_offset++;
 	}
 
 	if (params.has_tangent) {
-		dst_vertices.data[dst_offset] = encode_abgr_2_10_10_10(tangent);
+		dst_vertices.data[dst_offset] = encode_tang_to_uint_oct(tangent);
 	}
 
 #endif
diff --git a/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp b/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
index 1516e5e4fd..49d7198ec2 100644
--- a/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
+++ b/servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
@@ -1073,10 +1073,9 @@ 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 {
@@ -1085,9 +1084,9 @@ 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 dcdc56d948..bbe78236b5 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;