forge-pbr-shading

Add physically-based rendering to an SDL GPU project. Implements a Cook-Torrance microfacet BRDF fragment shader that uses the same texture bindings and uniform layout as `forge_scene.h`'s built-in model pipeline, enabling side-by-side comparison with Blinn-Phong. ## When to use - Replacing Blinn-Phong with physically-based shading - Adding metallic-roughness material support - Comparing different lighting models on the same geometry - Creating a custom pipeline that coexists with forge_scene.h ## Key API calls - `forge_scene_create_shader()` — create vertex/fragment shaders from bytecode - `SDL_CreateGPUGraphicsPipeline()` — create custom pipeline matching model vertex layout - `SDL_BindGPUFragmentSamplers()` — bind per-material textures with fallbacks - `SDL_PushGPUFragmentUniformData()` — push material uniform data to fragment shader - `forge_scene_load_model()` — load pipeline-processed model - `forge_scene_draw_model()` — draw with built-in Blinn-Phong for comparison ## Correct order 1. Load model via `forge_scene_load_model()` 2. Create PBR fragment shader from compiled bytecode 3. Create vertex shader from `scene_model_vert_spirv/dxil/msl` (available via `FORGE_SCENE_IMPLEMENTATION`) 4. Create pipeline with model vertex layout using `sizeof(ForgeSceneModelVertex)` for pitch and `offsetof()` for attribute offsets 5. Release shaders after pipeline creation 6. Per frame: bind custom pipeline, iterate model submeshes, push uniforms with overridden PBR parameters, bind textures, draw ## Cook-Torrance BRDF components ```hlsl /* GGX Normal Distribution */ float D = alpha2 / (PI * pow(NdotH * NdotH * (alpha2 - 1.0) + 1.0, 2.0)); /* Schlick-GGX Geometry (Smith method) */ float k = (roughness + 1.0) * (roughness + 1.0) / 8.0; float G = (NdotV / (NdotV * (1-k) + k)) * (NdotL / (NdotL * (1-k) + k)); /* Schlick Fresnel */ float3 F = F0 + (1.0 - F0) * pow(1.0 - VdotH, 5.0); /* Full specular: D * G * F / (4 * NdotV * NdotL) */ /* Energy-conserving diffuse: (1 - F) * (1 - metallic) * albedo / PI */ ``` ## Common mistakes - **F0 for metals must use albedo color** — `lerp(0.04, albedo, metallic)`, not a fixed 0.04 for all materials - **alpha = roughness^2, not roughness** — GGX uses squared roughness - **k differs for direct vs IBL lighting** — direct: `(r+1)^2/8`, IBL: `r^2/2` - **Divide-by-zero guard** — clamp NdotV above zero (e.g. 0.001) in the Cook-Torrance denominator - **Diffuse must divide by pi** — without this, the surface reflects more energy than it receives ## Reference See [Lesson 51 — PBR Shading Model](../../../lessons/gpu/51-pbr-shading/) for the full implementation.