reze-engine 0.11.0 → 0.11.2

package/src/shaders/body.ts

@@ -48,43 +48,38 @@ struct LightVP { viewProj: mat4x4f, };
 @group(2) @binding(1) var<uniform> material: MaterialUniforms;
 
 fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
+  // Back-facing to key light: direct contribution is zero anyway, skip 9 texture samples.
+  if (dot(n, -light.lights[0].direction.xyz) <= 0.0) { return 0.0; }
   let biasedPos = worldPos + n * 0.08;
   let lclip = lightVP.viewProj * vec4f(biasedPos, 1.0);
   let ndc = lclip.xyz / max(lclip.w, 1e-6);
   let suv = vec2f(ndc.x * 0.5 + 0.5, 0.5 - ndc.y * 0.5);
   let cmpZ = ndc.z - 0.001;
-  let ts = 1.0 / 4096.0;
-  var vis = 0.0;
-  for (var y = -1; y <= 1; y++) {
-    for (var x = -1; x <= 1; x++) {
-      vis += textureSampleCompare(shadowMap, shadowSampler, suv + vec2f(f32(x), f32(y)) * ts, cmpZ);
-    }
-  }
-  return vis / 9.0;
+  let ts = 1.0 / 2048.0;
+  // 3x3 PCF unrolled — Safari's Metal backend doesn't unroll nested shadow loops reliably.
+  let s00 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, -ts), cmpZ);
+  let s10 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0, -ts), cmpZ);
+  let s20 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, -ts), cmpZ);
+  let s01 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, 0.0), cmpZ);
+  let s11 = textureSampleCompareLevel(shadowMap, shadowSampler, suv, cmpZ);
+  let s21 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, 0.0), cmpZ);
+  let s02 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts,  ts), cmpZ);
+  let s12 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0,  ts), cmpZ);
+  let s22 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts,  ts), cmpZ);
+  return (s00 + s10 + s20 + s01 + s11 + s21 + s02 + s12 + s22) * (1.0 / 9.0);
 }
 
 const PI_B: f32 = 3.141592653589793;
-const F0_BODY: f32 = 0.04;
 const BODY_ROUGHNESS: f32 = 0.3;
 // Dump: 层权重.002 Blend; 运算.007 POWER exponent Value_001; 背景 Color; 运算.004 after invert
 const BODY_RIM2_LAYER_BLEND: f32 = 0.20000000298023224;
 const BODY_RIM2_POW: f32 = 1.4300000667572021;
 const BODY_RIM2_BG: vec3f = vec3f(1.0, 0.4303792119026184, 0.3315804898738861);
 const BODY_WARM_AO_MUL: f32 = 0.30000001192092896;
+const BODY_SPECULAR: f32 = 0.5;
 const BODY_MIX_NPR: f32 = 0.5;
-
-fn ggx_d_body(ndoth: f32, a2: f32) -> f32 {
-  let denom = ndoth * ndoth * (a2 - 1.0) + 1.0;
-  return a2 / (PI_B * denom * denom);
-}
-
-fn smith_g1_body(ndotx: f32, a2: f32) -> f32 {
-  return 2.0 * ndotx / (ndotx + sqrt(a2 + (1.0 - a2) * ndotx * ndotx));
-}
-
-fn fresnel_schlick_body(cosTheta: f32, f0: f32) -> f32 {
-  return f0 + (1.0 - f0) * pow(1.0 - cosTheta, 5.0);
-}
+// EEVEE Light Clamp equivalent — caps firefly specular from noise-bumped NDF aliasing.
+const BODY_SPEC_CLAMP: f32 = 10.0;
 
 // smoothstep-based ramp: t*t*(3-2*t) between two color stops
 fn ramp_ease(f: f32, p0: f32, c0: vec4f, p1: f32, c1: vec4f) -> vec4f {
@@ -114,13 +109,19 @@ fn ramp_ease(f: f32, p0: f32, c0: vec4f, p1: f32, c1: vec4f) -> vec4f {
     skinnedNrm += (mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz) * normal) * w;
   }
   output.position = camera.projection * camera.view * vec4f(skinnedPos.xyz, 1.0);
-  output.normal = normalize(skinnedNrm);
+  // Skip VS normalize — interpolation denormalizes anyway, and FS always does normalize(input.normal).
+  output.normal = skinnedNrm;
   output.uv = uv;
   output.worldPos = skinnedPos.xyz;
   return output;
 }
 
-@fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+struct FSOut {
+  @location(0) color: vec4f,
+  @location(1) mask: f32,
+};
+
+@fragment fn fs(input: VertexOutput) -> FSOut {
   let alpha = material.alpha;
   if (alpha < 0.001) { discard; }
 
@@ -137,13 +138,13 @@ fn ramp_ease(f: f32, p0: f32, c0: vec4f, p1: f32, c1: vec4f) -> vec4f {
   let toon = ramp_constant(ndotl_raw, 0.0, vec4f(0,0,0,1), 0.2966, vec4f(1,1,1,1)).r;
 
   // ═══ TOON COLOR: Mix.004 A=HueSat, B=HueSat.001, Fac=ramp.008 (R) ═══
-  let shadow_tint = hue_sat(0.5, 2.0, 0.3499999940395355, 1.0, tex_color);
-  let lit_tint = hue_sat(0.5, 1.5, 1.0, 1.0, tex_color);
+  let shadow_tint = hue_sat_id(2.0, 0.3499999940395355, 1.0, tex_color);
+  let lit_tint = hue_sat_id(1.5, 1.0, 1.0, tex_color);
   let toon_color = mix_blend(toon, shadow_tint, lit_tint);
   let bc = bright_contrast(toon_color, 0.1, 0.2);
 
   // ═══ AO CHAIN: AO → ramp CONSTANT [0→white, 0.5995→black] → Mix.003 ═══
-  let ao = ao_fake(n, v);
+  let ao = 1.0; // ao_fake(n, v) — no SSAO yet; inline 1.0 so the ramp/mix chain folds at compile time.
   let ao_ramp = ramp_constant(ao, 0.0, vec4f(1,1,1,1), 0.5995, vec4f(0,0,0,1)).r;
   let ao_mixed = mix_blend(ao_ramp, bc, vec3f(0.8301780223846436, 0.3345769941806793, 0.27946099638938904));
 
@@ -174,38 +175,48 @@ fn ramp_ease(f: f32, p0: f32, c0: vec4f, p1: f32, c1: vec4f) -> vec4f {
   let npr_stack = add0 + warm_emission;
 
   // ═══ PRINCIPLED BSDF: noise bump, GGX specular, SSS from AO ═══
-  let gen = mapping_point(input.worldPos, vec3f(0.0), vec3f(0.0), vec3f(1.0, 1.0, 1.5));
-  let noise_val = tex_noise(gen, 1.0, 2.0, 0.5, 0.0);
+  // Mapping loc=rot=0 → plain scale multiply, inline.
+  let noise_val = tex_noise_d2(input.worldPos * vec3f(1.0, 1.0, 1.5), 1.0);
   let noise_ramp = ramp_linear(noise_val, 0.0, vec4f(0,0,0,1), 1.0, vec4f(1,1,1,1)).r;
   let bumped_n = bump_lh(0.324644535779953, noise_ramp, n, input.worldPos);
 
   let principled_base = mix_blend(noise_ramp, bc, vec3f(0.6831911206245422, 0.19474034011363983, 0.13732507824897766));
   let p_emission = bc * 0.2;
 
-  let ao2 = ao_fake(n, v);
-  let sss = ramp_linear(ao2, 0.003, vec4f(0,0,0,1), 1.0, vec4f(0.0786, 0.0786, 0.0786, 1.0)).r;
-
-  let p_ndotl = max(dot(bumped_n, l), 0.0);
-  let p_ndotv = max(dot(bumped_n, v), 0.001);
-  let h = normalize(l + v);
-  let p_ndoth = max(dot(bumped_n, h), 0.0);
-  let p_vdoth = max(dot(v, h), 0.0);
-  let a2 = BODY_ROUGHNESS * BODY_ROUGHNESS;
-  let D = ggx_d_body(p_ndoth, a2);
-  let G = smith_g1_body(p_ndotl, a2) * smith_g1_body(p_ndotv, a2);
-  let F = fresnel_schlick_body(p_vdoth, F0_BODY);
-  let spec = (D * G * F) / max(4.0 * p_ndotl * p_ndotv, 0.001) * ltc_brdf_scale(p_ndotv, BODY_ROUGHNESS);
-  let kd = (1.0 - F) * principled_base / PI_B;
-  let direct = (kd + spec) * sun * p_ndotl * shadow;
+  // Reuse 'ao' (ao_fake(n, v) above) — identical inputs, avoid a second procedural AO pass.
+  let sss = ramp_linear(ao, 0.003, vec4f(0,0,0,1), 1.0, vec4f(0.0786, 0.0786, 0.0786, 1.0)).r;
+
+  // 原理化BSDF (EEVEE port): metallic=0, specular=0.5, roughness=0.3, specular_tint=0.
+  let NL = max(dot(bumped_n, l), 0.0);
+  let NV = max(dot(bumped_n, v), 1e-4);
+
+  // f0/f90 per gpu_shader_material_principled.glsl — specular_tint=0 → dielectric_f0_color=white.
+  let f0 = vec3f(0.08 * BODY_SPECULAR);
+  let f90 = mix(f0, vec3f(1.0), sqrt(BODY_SPECULAR));
+  let brdf_lut = brdf_lut_sample(NV, BODY_ROUGHNESS);
+  let reflection_color = F_brdf_multi_scatter(f0, f90, brdf_lut.xy);
+
+  // Direct glossy — bsdf_ggx already includes NL; no F applied here (tinted after accum).
+  // ltc_brdf_scale: EEVEE direct path uses LTC; split-sum LUT path is rescaled to match.
+  let spec_direct_raw = bsdf_ggx(bumped_n, l, v, NL, NV, BODY_ROUGHNESS) * sun * shadow * ltc_brdf_scale_from_lut(brdf_lut);
+  let spec_direct = min(spec_direct_raw, vec3f(BODY_SPEC_CLAMP));
+  // Indirect glossy — flat world probe (solid color). Phase 2 adds cubemap.
+  let spec_indirect = light.ambientColor.xyz;
+  let spec_radiance = (spec_direct + spec_indirect) * reflection_color;
+
   // Indirect diffuse = base_color × L_w per Blender closure_eval_surface_lib.glsl line 302;
   // probe_evaluate_world_diff returns radiance (SH-projected, not cosine-convolved).
-  let ambient = principled_base * light.ambientColor.xyz;
-  let principled = ambient + direct + p_emission + vec3f(sss);
+  // No (1-F) factor per EEVEE — it doesn't energy-conserve spec<->diffuse.
+  let diffuse_radiance = principled_base * (sun * NL * shadow / PI_B + light.ambientColor.xyz);
+  let principled = diffuse_radiance + spec_radiance + p_emission + vec3f(sss);
 
   // 混合着色器.001: Shader=相加着色器.001, Shader_001=原理化BSDF
   let final_color = mix(npr_stack, principled, BODY_MIX_NPR);
 
-  return vec4f(final_color, alpha);
+  var out: FSOut;
+  out.color = vec4f(final_color, alpha);
+  out.mask = 1.0;
+  return out;
 }
 
 `

package/src/shaders/cloth_rough.ts

@@ -49,19 +49,25 @@ struct LightVP { viewProj: mat4x4f, };
 @group(2) @binding(1) var<uniform> material: MaterialUniforms;
 
 fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
+  // Back-facing to key light: direct contribution is zero anyway, skip 9 texture samples.
+  if (dot(n, -light.lights[0].direction.xyz) <= 0.0) { return 0.0; }
   let biasedPos = worldPos + n * 0.08;
   let lclip = lightVP.viewProj * vec4f(biasedPos, 1.0);
   let ndc = lclip.xyz / max(lclip.w, 1e-6);
   let suv = vec2f(ndc.x * 0.5 + 0.5, 0.5 - ndc.y * 0.5);
   let cmpZ = ndc.z - 0.001;
-  let ts = 1.0 / 4096.0;
-  var vis = 0.0;
-  for (var y = -1; y <= 1; y++) {
-    for (var x = -1; x <= 1; x++) {
-      vis += textureSampleCompare(shadowMap, shadowSampler, suv + vec2f(f32(x), f32(y)) * ts, cmpZ);
-    }
-  }
-  return vis / 9.0;
+  let ts = 1.0 / 2048.0;
+  // 3x3 PCF unrolled — Safari's Metal backend doesn't unroll nested shadow loops reliably.
+  let s00 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, -ts), cmpZ);
+  let s10 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0, -ts), cmpZ);
+  let s20 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, -ts), cmpZ);
+  let s01 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, 0.0), cmpZ);
+  let s11 = textureSampleCompareLevel(shadowMap, shadowSampler, suv, cmpZ);
+  let s21 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, 0.0), cmpZ);
+  let s02 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts,  ts), cmpZ);
+  let s12 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0,  ts), cmpZ);
+  let s22 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts,  ts), cmpZ);
+  return (s00 + s10 + s20 + s01 + s11 + s21 + s02 + s12 + s22) * (1.0 / 9.0);
 }
 
 const PI_CR: f32 = 3.141592653589793;
@@ -73,6 +79,8 @@ const CLOTH_R_EMIT_STR: f32 = 18.200000762939453;
 const CLOTH_R_MIX_SHADER_FAC: f32 = 0.8999999761581421;
 const CLOTH_R_NOISE_SCALE: f32 = 17.7;
 const CLOTH_R_BUMP_STR: f32 = 1.0;
+// EEVEE Light Clamp equivalent — caps firefly specular from noise-bumped NDF aliasing.
+const CLOTH_R_SPEC_CLAMP: f32 = 10.0;
 
 @vertex fn vs(
   @location(0) position: vec3f,
@@ -95,13 +103,19 @@ const CLOTH_R_BUMP_STR: f32 = 1.0;
     skinnedNrm += (mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz) * normal) * w;
   }
   output.position = camera.projection * camera.view * vec4f(skinnedPos.xyz, 1.0);
-  output.normal = normalize(skinnedNrm);
+  // Skip VS normalize — interpolation denormalizes anyway, and FS always does normalize(input.normal).
+  output.normal = skinnedNrm;
   output.uv = uv;
   output.worldPos = skinnedPos.xyz;
   return output;
 }
 
-@fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+struct FSOut {
+  @location(0) color: vec4f,
+  @location(1) mask: f32,
+};
+
+@fragment fn fs(input: VertexOutput) -> FSOut {
   let n = normalize(input.normal);
   let v = normalize(camera.viewPos - input.worldPos);
   let l = -light.lights[0].direction.xyz;
@@ -121,7 +135,7 @@ const CLOTH_R_BUMP_STR: f32 = 1.0;
   let mix04_fac = math_multiply(toon_r, CLOTH_R_MIX04_MUL);
 
   // 混合.004: A=色相/饱和度/明度.002(Hue=0.5 Sat=1.0 Val=0.2), B=纹理
-  let dark_tex = hue_sat(0.5, 1.0, 0.19999998807907104, 1.0, tex_rgb);
+  let dark_tex = hue_sat_id(1.0, 0.19999998807907104, 1.0, tex_rgb);
   let mix04 = mix_blend(mix04_fac, dark_tex, tex_rgb);
 
   // 倒角.001.Z → 混合.003: A=混合.004, B=色相/饱和度/明度.002
@@ -129,34 +143,35 @@ const CLOTH_R_BUMP_STR: f32 = 1.0;
   let mix03 = mix_blend(bevel_z, mix04, dark_tex);
 
   // 环境光遮蔽 → 颜色渐变.001 LINEAR → 混合.001 (white/black) → 混合.002 OVERLAY Fac
-  let ao = ao_fake(n, v);
+  let ao = 1.0; // ao_fake(n, v) — no SSAO yet; inline 1.0 so the ramp/mix chain folds at compile time.
   let ao_ramp_c = ramp_linear(ao, 0.0, vec4f(1,1,1,1), 0.8808, vec4f(0,0,0,1));
   let mix01_fac = ao_ramp_c.r;
   let mix01_rgb = mix(vec3f(1.0), vec3f(0.0), mix01_fac);
 
   // 混合.002 OVERLAY: Fac=混合.001, A=混合.003, B=色相/饱和度/明度.004
-  let hue004 = hue_sat(0.5, 0.800000011920929, 2.0, 1.0, mix03);
+  let hue004 = hue_sat_id(0.800000011920929, 2.0, 1.0, mix03);
   let overlay_fac = mix01_rgb.r;
   let npr_rgb = mix_overlay(overlay_fac, mix03, hue004);
   let npr_emission = npr_rgb * CLOTH_R_EMIT_STR;
 
   // 噪波→渐变→凹凸 (LIVE in M_Rough_Cloth — unlike Smooth Cloth): Strength=1.0, noise Scale=17.7.
-  let noise_uv = mapping_point(input.worldPos, vec3f(0.0), vec3f(0.0), vec3f(1.0));
-  let noise_val = tex_noise(noise_uv, CLOTH_R_NOISE_SCALE, 2.0, 0.5, 0.0);
+  // mapping scale=(1,1,1), loc=rot=0 → identity; use worldPos directly.
+  let noise_val = tex_noise_d2(input.worldPos, CLOTH_R_NOISE_SCALE);
   let noise_ramp = ramp_linear(noise_val, 0.0, vec4f(0,0,0,1), 1.0, vec4f(1,1,1,1)).r;
   let bumped_n = bump_lh(CLOTH_R_BUMP_STR, noise_ramp, n, input.worldPos);
 
   // 原理化BSDF (EEVEE port): metallic=0, specular=0.8, roughness=0.8187, specular_tint=0.
-  let principled_base = hue_sat(0.5, 1.0, 0.800000011920929, 1.0, tex_rgb);
+  let principled_base = hue_sat_id(1.0, 0.800000011920929, 1.0, tex_rgb);
   let NL = max(dot(bumped_n, l), 0.0);
   let NV = max(dot(bumped_n, v), 1e-4);
 
   let f0 = vec3f(0.08 * CLOTH_R_SPECULAR);
   let f90 = mix(f0, vec3f(1.0), sqrt(CLOTH_R_SPECULAR));
-  let split_sum = brdf_lut_baked(NV, CLOTH_R_ROUGHNESS);
-  let reflection_color = F_brdf_multi_scatter(f0, f90, split_sum);
+  let brdf_lut = brdf_lut_sample(NV, CLOTH_R_ROUGHNESS);
+  let reflection_color = F_brdf_multi_scatter(f0, f90, brdf_lut.xy);
 
-  let spec_direct = bsdf_ggx(bumped_n, l, v, CLOTH_R_ROUGHNESS) * sun * shadow * ltc_brdf_scale(NV, CLOTH_R_ROUGHNESS);
+  let spec_direct_raw = bsdf_ggx(bumped_n, l, v, NL, NV, CLOTH_R_ROUGHNESS) * sun * shadow * ltc_brdf_scale_from_lut(brdf_lut);
+  let spec_direct = min(spec_direct_raw, vec3f(CLOTH_R_SPEC_CLAMP));
   let spec_indirect = amb;
   let spec_radiance = (spec_direct + spec_indirect) * reflection_color;
 
@@ -168,7 +183,10 @@ const CLOTH_R_BUMP_STR: f32 = 1.0;
   // 混合着色器.001 Fac=0.9: Shader=自发光.005, Shader_001=原理化BSDF
   let final_color = mix(npr_emission, principled, CLOTH_R_MIX_SHADER_FAC);
 
-  return vec4f(final_color, out_alpha);
+  var out: FSOut;
+  out.color = vec4f(final_color, out_alpha);
+  out.mask = 1.0;
+  return out;
 }
 
 `

package/src/shaders/cloth_smooth.ts

@@ -48,19 +48,25 @@ struct LightVP { viewProj: mat4x4f, };
 @group(2) @binding(1) var<uniform> material: MaterialUniforms;
 
 fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
+  // Back-facing to key light: direct contribution is zero anyway, skip 9 texture samples.
+  if (dot(n, -light.lights[0].direction.xyz) <= 0.0) { return 0.0; }
   let biasedPos = worldPos + n * 0.08;
   let lclip = lightVP.viewProj * vec4f(biasedPos, 1.0);
   let ndc = lclip.xyz / max(lclip.w, 1e-6);
   let suv = vec2f(ndc.x * 0.5 + 0.5, 0.5 - ndc.y * 0.5);
   let cmpZ = ndc.z - 0.001;
-  let ts = 1.0 / 4096.0;
-  var vis = 0.0;
-  for (var y = -1; y <= 1; y++) {
-    for (var x = -1; x <= 1; x++) {
-      vis += textureSampleCompare(shadowMap, shadowSampler, suv + vec2f(f32(x), f32(y)) * ts, cmpZ);
-    }
-  }
-  return vis / 9.0;
+  let ts = 1.0 / 2048.0;
+  // 3x3 PCF unrolled — Safari's Metal backend doesn't unroll nested shadow loops reliably.
+  let s00 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, -ts), cmpZ);
+  let s10 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0, -ts), cmpZ);
+  let s20 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, -ts), cmpZ);
+  let s01 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, 0.0), cmpZ);
+  let s11 = textureSampleCompareLevel(shadowMap, shadowSampler, suv, cmpZ);
+  let s21 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, 0.0), cmpZ);
+  let s02 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts,  ts), cmpZ);
+  let s12 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0,  ts), cmpZ);
+  let s22 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts,  ts), cmpZ);
+  return (s00 + s10 + s20 + s01 + s11 + s21 + s02 + s12 + s22) * (1.0 / 9.0);
 }
 
 const PI_C: f32 = 3.141592653589793;
@@ -92,13 +98,19 @@ const NPR_MIX_SHADER_FAC: f32 = 0.8999999761581421;
     skinnedNrm += (mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz) * normal) * w;
   }
   output.position = camera.projection * camera.view * vec4f(skinnedPos.xyz, 1.0);
-  output.normal = normalize(skinnedNrm);
+  // Skip VS normalize — interpolation denormalizes anyway, and FS always does normalize(input.normal).
+  output.normal = skinnedNrm;
   output.uv = uv;
   output.worldPos = skinnedPos.xyz;
   return output;
 }
 
-@fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+struct FSOut {
+  @location(0) color: vec4f,
+  @location(1) mask: f32,
+};
+
+@fragment fn fs(input: VertexOutput) -> FSOut {
   let n = normalize(input.normal);
   let v = normalize(camera.viewPos - input.worldPos);
   let l = -light.lights[0].direction.xyz;
@@ -119,7 +131,7 @@ const NPR_MIX_SHADER_FAC: f32 = 0.8999999761581421;
   let mix04_fac = math_multiply(toon_r, CLOTH_MIX04_MUL);
 
   // 混合.004: A=色相/饱和度/明度.002, B=纹理
-  let dark_tex = hue_sat(0.5, 1.0, 0.19999998807907104, 1.0, tex_rgb);
+  let dark_tex = hue_sat_id(1.0, 0.19999998807907104, 1.0, tex_rgb);
   let mix04 = mix_blend(mix04_fac, dark_tex, tex_rgb);
 
   // 倒角.001→Z → 混合.003 Factor; A=混合.004, B=色相/饱和度/明度.002
@@ -127,32 +139,32 @@ const NPR_MIX_SHADER_FAC: f32 = 0.8999999761581421;
   let mix03 = mix_blend(bevel_z, mix04, dark_tex);
 
   // 环境光遮蔽 → 颜色渐变.001 LINEAR → 混合.001 (白/黑) → 混合.002 OVERLAY Fac
-  let ao = ao_fake(n, v);
+  let ao = 1.0; // ao_fake(n, v) — no SSAO yet; inline 1.0 so the ramp/mix chain folds at compile time.
   let ao_ramp_c = ramp_linear(ao, 0.0, vec4f(1,1,1,1), 0.8808, vec4f(0,0,0,1));
   let mix01_fac = ao_ramp_c.r;
   let mix01_rgb = mix(vec3f(1.0), vec3f(0.0), mix01_fac);
 
   // 混合.002 OVERLAY: Fac=混合.001, A=混合.003, B=色相/饱和度/明度.004
-  let hue004 = hue_sat(0.5, 0.800000011920929, 2.0, 1.0, mix03);
+  let hue004 = hue_sat_id(0.800000011920929, 2.0, 1.0, mix03);
   let overlay_fac = mix01_rgb.r;
   let npr_rgb = mix_overlay(overlay_fac, mix03, hue004);
   let npr_emission = npr_rgb * NPR_EMIT_STR;
 
   // 原理化BSDF (EEVEE port): metallic=0, specular=0.8, roughness=0.5, specular_tint=0.
   // Bump subtree is dead in the Blender graph (噪波→凹凸 not linked to Principled.Normal).
-  let principled_base = hue_sat(0.5, 1.0, 0.800000011920929, 1.0, tex_rgb);
+  let principled_base = hue_sat_id(1.0, 0.800000011920929, 1.0, tex_rgb);
   let NL = max(dot(n, l), 0.0);
   let NV = max(dot(n, v), 1e-4);
 
   // f0/f90 per gpu_shader_material_principled.glsl — specular_tint=0 → dielectric_f0_color=white.
   let f0 = vec3f(0.08 * CLOTH_SPECULAR);
   let f90 = mix(f0, vec3f(1.0), sqrt(CLOTH_SPECULAR));
-  let split_sum = brdf_lut_baked(NV, CLOTH_ROUGHNESS);
-  let reflection_color = F_brdf_multi_scatter(f0, f90, split_sum);
+  let brdf_lut = brdf_lut_sample(NV, CLOTH_ROUGHNESS);
+  let reflection_color = F_brdf_multi_scatter(f0, f90, brdf_lut.xy);
 
   // Direct glossy — bsdf_ggx already includes NL; no F applied here (tinted after accum).
   // ltc_brdf_scale: EEVEE direct path uses LTC; split-sum LUT path is rescaled to match.
-  let spec_direct = bsdf_ggx(n, l, v, CLOTH_ROUGHNESS) * sun * shadow * ltc_brdf_scale(NV, CLOTH_ROUGHNESS);
+  let spec_direct = bsdf_ggx(n, l, v, NL, NV, CLOTH_ROUGHNESS) * sun * shadow * ltc_brdf_scale_from_lut(brdf_lut);
   // Indirect glossy — flat world probe (solid color). Phase 2 adds cubemap.
   let spec_indirect = amb;
   let spec_radiance = (spec_direct + spec_indirect) * reflection_color;
@@ -167,7 +179,10 @@ const NPR_MIX_SHADER_FAC: f32 = 0.8999999761581421;
   // 混合着色器.001: Shader=自发光.005, Shader_001=原理化BSDF, Fac=0.9
   let final_color = mix(npr_emission, principled, NPR_MIX_SHADER_FAC);
 
-  return vec4f(final_color, out_alpha);
+  var out: FSOut;
+  out.color = vec4f(final_color, out_alpha);
+  out.mask = 1.0;
+  return out;
 }
 
 `

package/src/shaders/default.ts

@@ -2,10 +2,14 @@
 // Metallic=0, Specular=0.5 (F0=0.04), Roughness=0.5.
 // Tone mapping via LUT sampled from Blender's OCIO pipeline (exposure -0.3 baked in).
 
+import { NODES_WGSL } from "./nodes"
+
 export const DEFAULT_SHADER_WGSL = /* wgsl */ `
 
+${NODES_WGSL}
+
 const PI: f32 = 3.141592653589793;
-const F0_DIELECTRIC: f32 = 0.04;
+const DEFAULT_SPECULAR: f32 = 0.5;
 const ROUGHNESS: f32 = 0.5;
 
 struct CameraUniforms {
@@ -52,21 +56,6 @@ struct LightVP { viewProj: mat4x4f, };
 @group(2) @binding(0) var diffuseTexture: texture_2d<f32>;
 @group(2) @binding(1) var<uniform> material: MaterialUniforms;
 
-// ─── GGX specular helpers ───────────────────────────────────────────
-
-fn ggx_d(ndoth: f32, a2: f32) -> f32 {
-  let denom = ndoth * ndoth * (a2 - 1.0) + 1.0;
-  return a2 / (PI * denom * denom);
-}
-
-fn smith_g1(ndotx: f32, a2: f32) -> f32 {
-  return 2.0 * ndotx / (ndotx + sqrt(a2 + (1.0 - a2) * ndotx * ndotx));
-}
-
-fn fresnel_schlick(cosTheta: f32, f0: f32) -> f32 {
-  return f0 + (1.0 - f0) * pow(1.0 - cosTheta, 5.0);
-}
-
 // ─── Filmic tone mapping (LUT extracted from Blender 3.6 OCIO) ─────
 // View transform = Filmic, Look = Medium High Contrast, Exposure = -0.3.
 // 14 samples at integer log2 stops from -10 to +3 (inclusive).
@@ -91,19 +80,25 @@ fn tonemap(hdr: vec3f) -> vec3f {
 // ─── Shadow sampling (3×3 PCF) ──────────────────────────────────────
 
 fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
+  // Back-facing to key light: direct contribution is zero anyway, skip 9 texture samples.
+  if (dot(n, -light.lights[0].direction.xyz) <= 0.0) { return 0.0; }
   let biasedPos = worldPos + n * 0.08;
   let lclip = lightVP.viewProj * vec4f(biasedPos, 1.0);
   let ndc = lclip.xyz / max(lclip.w, 1e-6);
   let suv = vec2f(ndc.x * 0.5 + 0.5, 0.5 - ndc.y * 0.5);
   let cmpZ = ndc.z - 0.001;
-  let ts = 1.0 / 4096.0;
-  var vis = 0.0;
-  for (var y = -1; y <= 1; y++) {
-    for (var x = -1; x <= 1; x++) {
-      vis += textureSampleCompare(shadowMap, shadowSampler, suv + vec2f(f32(x), f32(y)) * ts, cmpZ);
-    }
-  }
-  return vis / 9.0;
+  let ts = 1.0 / 2048.0;
+  // 3x3 PCF unrolled — Safari's Metal backend doesn't unroll nested shadow loops reliably.
+  let s00 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, -ts), cmpZ);
+  let s10 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0, -ts), cmpZ);
+  let s20 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, -ts), cmpZ);
+  let s01 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts, 0.0), cmpZ);
+  let s11 = textureSampleCompareLevel(shadowMap, shadowSampler, suv, cmpZ);
+  let s21 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts, 0.0), cmpZ);
+  let s02 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(-ts,  ts), cmpZ);
+  let s12 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f(0.0,  ts), cmpZ);
+  let s22 = textureSampleCompareLevel(shadowMap, shadowSampler, suv + vec2f( ts,  ts), cmpZ);
+  return (s00 + s10 + s20 + s01 + s11 + s21 + s02 + s12 + s22) * (1.0 / 9.0);
 }
 
 // ─── Vertex / Fragment ──────────────────────────────────────────────
@@ -129,13 +124,19 @@ fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
     skinnedNrm += (mat3x3f(m[0].xyz, m[1].xyz, m[2].xyz) * normal) * w;
   }
   output.position = camera.projection * camera.view * vec4f(skinnedPos.xyz, 1.0);
-  output.normal = normalize(skinnedNrm);
+  // Skip VS normalize — interpolation denormalizes anyway, and FS always does normalize(input.normal).
+  output.normal = skinnedNrm;
   output.uv = uv;
   output.worldPos = skinnedPos.xyz;
   return output;
 }
 
-@fragment fn fs(input: VertexOutput) -> @location(0) vec4f {
+struct FSOut {
+  @location(0) color: vec4f,
+  @location(1) mask: f32,
+};
+
+@fragment fn fs(input: VertexOutput) -> FSOut {
   let alpha = material.alpha;
   if (alpha < 0.001) { discard; }
 
@@ -144,26 +145,29 @@ fn sampleShadow(worldPos: vec3f, n: vec3f) -> f32 {
   let albedo = textureSample(diffuseTexture, diffuseSampler, input.uv).rgb;
 
   let l = -light.lights[0].direction.xyz;
-  let sunColor = light.lights[0].color.xyz * light.lights[0].color.w;
-  let h = normalize(l + v);
+  let sun = light.lights[0].color.xyz * light.lights[0].color.w;
+  let amb = light.ambientColor.xyz;
+  let shadow = sampleShadow(input.worldPos, n);
 
-  let ndotl = max(dot(n, l), 0.0);
-  let ndotv = max(dot(n, v), 0.001);
-  let ndoth = max(dot(n, h), 0.0);
-  let vdoth = max(dot(v, h), 0.0);
+  // 原理化BSDF (EEVEE port): metallic=0, specular=0.5, roughness=0.5, specular_tint=0.
+  let NL = max(dot(n, l), 0.0);
+  let NV = max(dot(n, v), 1e-4);
 
-  let a2 = ROUGHNESS * ROUGHNESS;
-  let D = ggx_d(ndoth, a2);
-  let G = smith_g1(ndotl, a2) * smith_g1(ndotv, a2);
-  let F = fresnel_schlick(vdoth, F0_DIELECTRIC);
-  let spec = (D * G * F) / max(4.0 * ndotl * ndotv, 0.001);
+  let f0 = vec3f(0.08 * DEFAULT_SPECULAR);
+  let f90 = mix(f0, vec3f(1.0), sqrt(DEFAULT_SPECULAR));
+  let brdf_lut = brdf_lut_sample(NV, ROUGHNESS);
+  let reflection_color = F_brdf_multi_scatter(f0, f90, brdf_lut.xy);
 
-  let shadow = sampleShadow(input.worldPos, n);
-  let kd = (1.0 - F) * albedo / PI;
-  let direct = (kd + spec) * sunColor * ndotl * shadow;
-  let ambient = albedo * light.ambientColor.xyz;
+  let spec_direct = bsdf_ggx(n, l, v, NL, NV, ROUGHNESS) * sun * shadow * ltc_brdf_scale_from_lut(brdf_lut);
+  let spec_indirect = amb;
+  let spec_radiance = (spec_direct + spec_indirect) * reflection_color;
+
+  let diffuse_radiance = albedo * (sun * NL * shadow / PI + amb);
 
-  return vec4f(ambient + direct, alpha);
+  var out: FSOut;
+  out.color = vec4f(diffuse_radiance + spec_radiance, alpha);
+  out.mask = 1.0;
+  return out;
 }
 
 `