GOOENGINE: Principled bsdf versioning

Increase max parameters to 38 and version the old and new principled bsdf for eevee legacy
Version vector math length squared function. A hacky solution but will look for a proper fix later

source/blender/draw/engines/gooengine/shaders/infos/eevee_legacy_common_info.hh

@@ -62,8 +62,8 @@ GPU_SHADER_CREATE_INFO(eevee_legacy_lights_lib)
     .sampler(8, ImageType::SHADOW_2D_ARRAY, "shadowCubeTexture")
     .sampler(9, ImageType::SHADOW_2D_ARRAY, "shadowCascadeTexture")
     /* GooEngine: Use slots 15,16 as others are already taken. */
-    .sampler(15, ImageType::UINT_2D_ARRAY, "shadowCubeIDTexture")
-    .sampler(16, ImageType::UINT_2D_ARRAY, "shadowCascadeIDTexture");
+    .sampler(23, ImageType::UINT_2D_ARRAY, "shadowCubeIDTexture")
+    .sampler(24, ImageType::UINT_2D_ARRAY, "shadowCascadeIDTexture");
 
 /* Hair lib. */
 GPU_SHADER_CREATE_INFO(eevee_legacy_hair_lib)

source/blender/gpu/GPU_material.hh

@@ -84,6 +84,13 @@ enum eGPUMaterialFlag {
 
   GPU_MATFLAG_BARYCENTRIC = (1 << 20),
 
+  /* Optimization to only add the branches of the principled shader that are necessary. */
+  GPU_MATFLAG_PRINCIPLED_COAT = (1 << 21),
+  GPU_MATFLAG_PRINCIPLED_METALLIC = (1 << 22),
+  GPU_MATFLAG_PRINCIPLED_DIELECTRIC = (1 << 23),
+  GPU_MATFLAG_PRINCIPLED_GLASS = (1 << 24),
+  GPU_MATFLAG_PRINCIPLED_ANY = (1 << 25),
+
   /* Tells the render engine the material was just compiled or updated. */
   GPU_MATFLAG_UPDATED = (1 << 29),
 
@@ -220,6 +227,10 @@ char *GPU_material_split_sub_function(GPUMaterial *material,
                                       GPUNodeLink **link);
 
 bool GPU_material_sss_profile_create(GPUMaterial *material, float radii[3]);
+GPUUniformBuf *GPU_material_sss_profile_get(GPUMaterial *material,
+                                            int sample_len,
+                                            GPUTexture **tex_profile);
+                                            
 /**
  * High level functions to create and use GPU materials.
  */

source/blender/gpu/intern/gpu_codegen.cc

@@ -397,6 +397,24 @@ void GPUCodegen::generate_resources()
 {
   GPUCodegenCreateInfo &info = *create_info;
 
+  /* Ref. #98190: Defines are optimizations for old compilers.
+   * Might become unnecessary with EEVEE-Next. */
+  if (GPU_material_flag_get(&mat, GPU_MATFLAG_PRINCIPLED_COAT)) {
+    info.define("PRINCIPLED_COAT");
+  }
+  if (GPU_material_flag_get(&mat, GPU_MATFLAG_PRINCIPLED_METALLIC)) {
+    info.define("PRINCIPLED_METALLIC");
+  }
+  if (GPU_material_flag_get(&mat, GPU_MATFLAG_PRINCIPLED_DIELECTRIC)) {
+    info.define("PRINCIPLED_DIELECTRIC");
+  }
+  if (GPU_material_flag_get(&mat, GPU_MATFLAG_PRINCIPLED_GLASS)) {
+    info.define("PRINCIPLED_GLASS");
+  }
+  if (GPU_material_flag_get(&mat, GPU_MATFLAG_PRINCIPLED_ANY)) {
+    info.define("PRINCIPLED_ANY");
+  }
+
   std::stringstream ss;
 
   /* Textures. */
@@ -944,6 +962,7 @@ bool GPU_pass_async_compilation_try_finalize(GPUPass *pass)
 bool GPU_pass_compile(GPUPass *pass, const char *shname)
 {
   BLI_mutex_lock(&pass->shader_creation_mutex);
+
   bool success = true;
   if (pass->async_compilation_handle > 0) {
     /* We're trying to compile this pass synchronously, but there's a pending asynchronous
@@ -955,6 +974,7 @@ bool GPU_pass_compile(GPUPass *pass, const char *shname)
     GPUShader *shader = GPU_shader_create_from_info(info);
     success = GPU_pass_finalize_compilation(pass, shader);
   }
+
   BLI_mutex_unlock(&pass->shader_creation_mutex);
   return success;
 }

source/blender/gpu/intern/gpu_material_library.hh

@@ -12,7 +12,7 @@
 #include "GPU_material.hh"
 
 #define MAX_FUNCTION_NAME 64
-#define MAX_PARAMETER 36
+#define MAX_PARAMETER 38
 
 struct GSet;
 

source/blender/gpu/shaders/material/gpu_shader_material_principled.glsl

@@ -58,7 +58,11 @@ void node_bsdf_principled(vec4 base_color,
                           float emission_strength,
                           float thin_film_thickness,
                           float thin_film_ior,
+                          const float do_diffuse,
+                          const float do_coat,
+                          const float do_refraction,
                           const float do_multiscatter,
+                          const float do_sss,
                           out Closure result)
 {
   /* Match cycles. */
@@ -91,6 +95,198 @@ void node_bsdf_principled(vec4 base_color,
   vec3 V = coordinate_incoming(g_data.P);
   float NV = dot(N, V);
 
+#ifdef GPU_SHADER_EEVEE_LEGACY_DEFINES
+    ClosureTransparency transparency_data;
+    transparency_data.weight = weight;
+    transparency_data.transmittance = vec3(1.0 - alpha);
+    transparency_data.holdout = 0.0;
+    weight *= alpha;
+
+    /* First layer: Sheen */
+    vec3 sheen_data_color = vec3(0.0);
+    if (sheen_weight > 0.0) {
+      /* TODO: Maybe sheen_weight should be specular. */
+      vec3 sheen_color = sheen_weight * sheen_tint.rgb * principled_sheen(NV, sheen_roughness);
+      sheen_data_color = weight * sheen_color;
+      /* Attenuate lower layers */
+      weight *= max((1.0 - math_reduce_max(sheen_color)), 0.0);
+    }
+
+    /* Second layer: Coat */
+    ClosureReflection coat_data;
+    coat_data.N = CN;
+    coat_data.roughness = coat_roughness;
+    coat_data.color = vec3(1.0);
+
+    if (coat_weight > 0.0) {
+      float coat_NV = dot(coat_data.N, V);
+      float reflectance = bsdf_lut(coat_NV, coat_data.roughness, coat_ior, false).x;
+      coat_data.weight = weight * coat_weight * reflectance;
+      /* Attenuate lower layers */
+      weight *= max((1.0 - reflectance * coat_weight), 0.0);
+
+      if (!all(equal(coat_tint.rgb, vec3(1.0)))) {
+        float coat_neta = 1.0 / coat_ior;
+        float NT = sqrt_fast(1.0 - coat_neta * coat_neta * (1 - NV * NV));
+        /* Tint lower layers. */
+        coat_tint.rgb = mix(vec3(1.0), pow(coat_tint.rgb, vec3(1.0 / NT)), saturate(coat_weight));
+      }
+    }
+    else {
+      coat_tint.rgb = vec3(1.0);
+      coat_data.weight = 0.0;
+    }
+
+    /* Attenuated by sheen and coat. */
+    ClosureEmission emission_data;
+    emission_data.weight = weight;
+    emission_data.emission = coat_tint.rgb * emission.rgb * emission_strength;
+
+    /* Metallic component */
+    ClosureReflection reflection_data;
+    reflection_data.N = N;
+    reflection_data.roughness = roughness;
+    vec3 reflection_tint = specular_tint.rgb;
+    if (metallic > 0.0) {
+      vec3 F0 = clamped_base_color.rgb;
+      vec3 F82 = min(reflection_tint, vec3(1.0));
+      vec3 metallic_brdf;
+      brdf_f82_tint_lut(F0, F82, NV, roughness, do_multiscatter != 0.0, metallic_brdf);
+      reflection_data.color = weight * metallic * metallic_brdf;
+      /* Attenuate lower layers */
+      weight *= max((1.0 - metallic), 0.0);
+    }
+    else {
+      reflection_data.color = vec3(0.0);
+    }
+
+    /* Transmission component */
+    ClosureRefraction refraction_data;
+    refraction_data.N = N;
+    refraction_data.roughness = roughness;
+    refraction_data.ior = ior;
+    if (transmission_weight > 0.0) {
+      vec3 F0 = vec3(F0_from_ior(ior)) * reflection_tint;
+      vec3 F90 = vec3(1.0);
+      vec3 reflectance, transmittance;
+      bsdf_lut(F0,
+              F90,
+              clamped_base_color.rgb,
+              NV,
+              roughness,
+              ior,
+              do_multiscatter != 0.0,
+              reflectance,
+              transmittance);
+
+      reflection_data.color += weight * transmission_weight * reflectance;
+
+      refraction_data.weight = weight * transmission_weight;
+      refraction_data.color = transmittance * coat_tint.rgb;
+      /* Attenuate lower layers */
+      weight *= max((1.0 - transmission_weight), 0.0);
+    }
+    else {
+      refraction_data.weight = 0.0;
+      refraction_data.color = vec3(0.0);
+    }
+
+    /* Specular component */
+    if (true) {
+      float eta = ior;
+      float f0 = F0_from_ior(eta);
+      if (specular_ior_level != 0.5) {
+        f0 *= 2.0 * specular_ior_level;
+        eta = ior_from_F0(f0);
+        if (ior < 1.0) {
+          eta = 1.0 / eta;
+        }
+      }
+
+      vec3 F0 = vec3(f0) * reflection_tint;
+      F0 = clamp(F0, vec3(0.0), vec3(1.0));
+      vec3 F90 = vec3(1.0);
+      vec3 reflectance, unused;
+      bsdf_lut(F0, F90, vec3(0.0), NV, roughness, eta, do_multiscatter != 0.0, reflectance, unused);
+
+      reflection_data.color += weight * reflectance;
+      /* Adjust the weight of picking the closure. */
+      reflection_data.color *= coat_tint.rgb;
+      reflection_data.weight = math_average(reflection_data.color);
+      reflection_data.color *= safe_rcp(reflection_data.weight);
+
+      /* Attenuate lower layers */
+      weight *= max((1.0 - math_reduce_max(reflectance)), 0.0);
+    }
+
+    /* EEVEE Legacy evaluates the subsurface as a diffuse closure.
+    * So this has no performance penalty. However, using a separate closure for subsurface
+    * (just like for EEVEE-Next) would induce a huge performance hit. */
+    ClosureSubsurface diffuse_data;
+    /* Flag subsurface as disabled by default. */
+    diffuse_data.sss_radius.b = -1.0;
+    diffuse_data.N = N;
+
+    subsurface_radius = max(subsurface_radius * subsurface_scale, vec3(0.0));
+
+    /* Subsurface component */
+    if (subsurface_weight > 0.0) {
+      /* For Legacy EEVEE, Subsurface is just part of the diffuse. Just evaluate the mixed color. */
+      /* Subsurface Scattering materials behave unpredictably with values greater than 1.0 in
+      * Cycles. So it's clamped there and we clamp here for consistency with Cycles. */
+      base_color = mix(base_color, clamped_base_color, subsurface_weight);
+      if (do_sss != 0.0) {
+        diffuse_data.sss_radius = subsurface_weight * subsurface_radius;
+      }
+    }
+
+    /* Diffuse component */
+    if (true) {
+      diffuse_data.color = weight * base_color.rgb * coat_tint.rgb;
+      /* Add energy of the sheen layer until we have proper sheen BSDF. */
+      diffuse_data.color += sheen_data_color;
+
+      diffuse_data.weight = math_average(diffuse_data.color);
+      diffuse_data.color *= safe_rcp(diffuse_data.weight);
+    }
+
+    /* Ref. #98190: Defines are optimizations for old compilers.
+    * Might become unnecessary with EEVEE-Next. */
+    if (do_diffuse == 0.0 && do_refraction == 0.0 && do_coat != 0.0) {
+  #ifdef PRINCIPLED_COAT
+      /* Metallic & Coat case. */
+      result = closure_eval(reflection_data, coat_data);
+  #endif
+    }
+    else if (do_diffuse == 0.0 && do_refraction == 0.0 && do_coat == 0.0) {
+  #ifdef PRINCIPLED_METALLIC
+      /* Metallic case. */
+      result = closure_eval(reflection_data);
+  #endif
+    }
+    else if (do_diffuse != 0.0 && do_refraction == 0.0 && do_coat == 0.0) {
+  #ifdef PRINCIPLED_DIELECTRIC
+      /* Dielectric case. */
+      result = closure_eval(diffuse_data, reflection_data);
+  #endif
+    }
+    else if (do_diffuse == 0.0 && do_refraction != 0.0 && do_coat == 0.0) {
+  #ifdef PRINCIPLED_GLASS
+      /* Glass case. */
+      result = closure_eval(reflection_data, refraction_data);
+  #endif
+    }
+    else {
+  #ifdef PRINCIPLED_ANY
+      /* Un-optimized case. */
+      result = closure_eval(diffuse_data, reflection_data, coat_data, refraction_data);
+  #endif
+    }
+    Closure emission_cl = closure_eval(emission_data);
+    Closure transparency_cl = closure_eval(transparency_data);
+    result = closure_add(result, emission_cl);
+    result = closure_add(result, transparency_cl);
+#else
   /* Transparency component. */
   if (true) {
     ClosureTransparency transparency_data;
@@ -251,4 +447,5 @@ void node_bsdf_principled(vec4 base_color,
   }
 
   result = Closure(0);
+#endif
 }

source/blender/gpu/shaders/material/gpu_shader_material_vector_math.glsl

@@ -4,10 +4,23 @@
 
 #pragma BLENDER_REQUIRE(gpu_shader_common_math_utils.glsl)
 
+float len_sqred(vec2 a)
+{
+  return dot(a, a);
+}
+float len_sqred(vec3 a)
+{
+  return dot(a, a);
+}
+float len_sqred(vec4 a)
+{
+  return dot(a, a);
+}
+
 vec3 vector_math_safe_normalize(vec3 a)
 {
   /* Match the safe normalize function in Cycles by defaulting to vec3(0.0) */
-  float length_sqr = length_squared(a);
+  float length_sqr = len_sqred(a);
   return (length_sqr > 1e-35f) ? a * inversesqrt(length_sqr) : vec3(0.0);
 }
 

source/blender/nodes/shader/nodes/node_shader_bsdf_principled.cc

@@ -343,6 +343,24 @@ static int node_shader_gpu_bsdf_principled(GPUMaterial *mat,
     flag |= GPU_MATFLAG_COAT;
   }
 
+  /* Ref. #98190: Defines are optimizations for old compilers.
+   * Might become unnecessary with EEVEE-Next. */
+  if (use_diffuse == false && use_refract == false && use_coat == true) {
+    flag |= GPU_MATFLAG_PRINCIPLED_COAT;
+  }
+  else if (use_diffuse == false && use_refract == false && use_coat == false) {
+    flag |= GPU_MATFLAG_PRINCIPLED_METALLIC;
+  }
+  else if (use_diffuse == true && use_refract == false && use_coat == false) {
+    flag |= GPU_MATFLAG_PRINCIPLED_DIELECTRIC;
+  }
+  else if (use_diffuse == false && use_refract == true && use_coat == false) {
+    flag |= GPU_MATFLAG_PRINCIPLED_GLASS;
+  }
+  else {
+    flag |= GPU_MATFLAG_PRINCIPLED_ANY;
+  }
+
   if (use_subsurf) {
     bNodeSocket *socket = (bNodeSocket *)BLI_findlink(&node->runtime->original->inputs,
                                                       SOCK_SUBSURFACE_RADIUS_ID);
@@ -352,11 +370,23 @@ static int node_shader_gpu_bsdf_principled(GPUMaterial *mat,
   }
 
   float use_multi_scatter = (node->custom1 == SHD_GLOSSY_MULTI_GGX) ? 1.0f : 0.0f;
+  float use_sss = (use_subsurf) ? 1.0f : 0.0f;
+  float use_diffuse_f = (use_diffuse) ? 1.0f : 0.0f;
+  float use_coat_f = (use_coat) ? 1.0f : 0.0f;
+  float use_refract_f = (use_refract) ? 1.0f : 0.0f;
 
   GPU_material_flag_set(mat, flag);
 
-  return GPU_stack_link(
-      mat, node, "node_bsdf_principled", in, out, GPU_constant(&use_multi_scatter));
+  return GPU_stack_link(mat,
+                        node,
+                        "node_bsdf_principled",
+                        in,
+                        out,
+                        GPU_constant(&use_diffuse_f),
+                        GPU_constant(&use_coat_f),
+                        GPU_constant(&use_refract_f),
+                        GPU_constant(&use_multi_scatter),
+                        GPU_constant(&use_sss));
 }
 
 static void node_shader_update_principled(bNodeTree *ntree, bNode *node)