game_engine/docs/superpowers/plans/2026-03-25-phase7-2-ssgi.md

Phase 7-2: SSGI Implementation Plan

For agentic workers: REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (- [ ]) syntax for tracking.

Goal: SSAO + Color Bleeding 기반 SSGI로 간접광과 앰비언트 오클루전 추가

Architecture: voltex_renderer에 ssgi.rs(리소스+커널 생성) + ssgi_shader.wgsl(SSGI 풀스크린 패스) 추가. 기존 deferred_lighting.wgsl의 Shadow+IBL 바인드 그룹에 SSGI 출력 텍스처를 추가하여 ambient에 적용.

Tech Stack: Rust, wgpu 28.0, WGSL

Spec: docs/superpowers/specs/2026-03-25-phase7-2-ssgi.md

---

File Structure

• crates/voltex_renderer/src/ssgi.rs — SsgiResources, SsgiUniform, 커널/노이즈 생성 (Create)
• crates/voltex_renderer/src/ssgi_shader.wgsl — SSGI 풀스크린 셰이더 (Create)
• crates/voltex_renderer/src/deferred_pipeline.rs — SSGI 파이프라인 + 바인드 그룹 레이아웃 추가 (Modify)
• crates/voltex_renderer/src/deferred_lighting.wgsl — SSGI 텍스처 읽어서 ambient 적용 (Modify)
• crates/voltex_renderer/src/lib.rs — ssgi 모듈 등록 (Modify)
• examples/deferred_demo/src/main.rs — SSGI 패스 통합 (Modify)

---

Task 1: SsgiResources + 커널/노이즈 생성

Files:

• Create: crates/voltex_renderer/src/ssgi.rs

• Modify: crates/voltex_renderer/src/lib.rs

• Step 1: ssgi.rs 작성

// crates/voltex_renderer/src/ssgi.rs
use bytemuck::{Pod, Zeroable};
use wgpu::util::DeviceExt;

pub const SSGI_OUTPUT_FORMAT: wgpu::TextureFormat = wgpu::TextureFormat::Rgba16Float;
pub const SSGI_KERNEL_SIZE: usize = 64;

#[repr(C)]
#[derive(Copy, Clone, Debug, Pod, Zeroable)]
pub struct SsgiUniform {
    pub projection: [f32; 16],
    pub view: [f32; 16],
    pub radius: f32,
    pub bias: f32,
    pub intensity: f32,
    pub indirect_strength: f32,
}

impl Default for SsgiUniform {
    fn default() -> Self {
        Self {
            projection: [0.0; 16],
            view: [0.0; 16],
            radius: 0.5,
            bias: 0.025,
            intensity: 1.5,
            indirect_strength: 0.5,
        }
    }
}

pub struct SsgiResources {
    pub output_view: wgpu::TextureView,
    pub kernel_buffer: wgpu::Buffer,
    pub noise_view: wgpu::TextureView,
    pub noise_sampler: wgpu::Sampler,
    pub uniform_buffer: wgpu::Buffer,
    pub width: u32,
    pub height: u32,
}

impl SsgiResources {
    pub fn new(device: &wgpu::Device, queue: &wgpu::Queue, width: u32, height: u32) -> Self {
        let output_view = create_ssgi_output(device, width, height);
        let kernel_data = generate_kernel(SSGI_KERNEL_SIZE);
        let kernel_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
            label: Some("SSGI Kernel"),
            contents: bytemuck::cast_slice(&kernel_data),
            usage: wgpu::BufferUsages::UNIFORM,
        });
        let noise_view = create_noise_texture(device, queue);
        let noise_sampler = device.create_sampler(&wgpu::SamplerDescriptor {
            label: Some("SSGI Noise Sampler"),
            address_mode_u: wgpu::AddressMode::Repeat,
            address_mode_v: wgpu::AddressMode::Repeat,
            mag_filter: wgpu::FilterMode::Nearest,
            min_filter: wgpu::FilterMode::Nearest,
            ..Default::default()
        });
        let uniform_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
            label: Some("SSGI Uniform"),
            contents: bytemuck::bytes_of(&SsgiUniform::default()),
            usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
        });
        Self { output_view, kernel_buffer, noise_view, noise_sampler, uniform_buffer, width, height }
    }

    pub fn resize(&mut self, device: &wgpu::Device, width: u32, height: u32) {
        self.output_view = create_ssgi_output(device, width, height);
        self.width = width;
        self.height = height;
    }
}

fn create_ssgi_output(device: &wgpu::Device, w: u32, h: u32) -> wgpu::TextureView {
    let tex = device.create_texture(&wgpu::TextureDescriptor {
        label: Some("SSGI Output"),
        size: wgpu::Extent3d { width: w, height: h, depth_or_array_layers: 1 },
        mip_level_count: 1,
        sample_count: 1,
        dimension: wgpu::TextureDimension::D2,
        format: SSGI_OUTPUT_FORMAT,
        usage: wgpu::TextureUsages::RENDER_ATTACHMENT | wgpu::TextureUsages::TEXTURE_BINDING,
        view_formats: &[],
    });
    tex.create_view(&wgpu::TextureViewDescriptor::default())
}

/// Generate hemisphere sample kernel for SSAO/SSGI.
/// Samples are distributed in a hemisphere (z >= 0) with more samples near center.
pub fn generate_kernel(count: usize) -> Vec<[f32; 4]> {
    let mut kernel = Vec::with_capacity(count);
    for i in 0..count {
        // Pseudo-random using simple hash
        let fi = i as f32;
        let x = pseudo_random(i * 2) * 2.0 - 1.0;
        let y = pseudo_random(i * 2 + 1) * 2.0 - 1.0;
        let z = pseudo_random(i * 3 + 7).max(0.05); // hemisphere, z > 0

        let len = (x * x + y * y + z * z).sqrt();
        let (nx, ny, nz) = (x / len, y / len, z / len);

        // Scale: more samples near center
        let mut scale = fi / count as f32;
        scale = 0.1 + scale * scale * 0.9; // lerp(0.1, 1.0, scale^2)

        kernel.push([nx * scale, ny * scale, nz * scale, 0.0]);
    }
    kernel
}

/// Generate 4x4 noise texture data (random tangent-space rotation vectors).
pub fn generate_noise_data() -> Vec<[f32; 4]> {
    let mut noise = Vec::with_capacity(16);
    for i in 0..16 {
        let x = pseudo_random(i * 5 + 13) * 2.0 - 1.0;
        let y = pseudo_random(i * 7 + 17) * 2.0 - 1.0;
        let len = (x * x + y * y).sqrt().max(0.001);
        noise.push([x / len, y / len, 0.0, 0.0]);
    }
    noise
}

fn create_noise_texture(device: &wgpu::Device, queue: &wgpu::Queue) -> wgpu::TextureView {
    let data = generate_noise_data();
    let bytes: Vec<u8> = data.iter().flat_map(|v| {
        v.iter().flat_map(|f| f.to_le_bytes())
    }).collect();

    let tex = device.create_texture(&wgpu::TextureDescriptor {
        label: Some("SSGI Noise"),
        size: wgpu::Extent3d { width: 4, height: 4, depth_or_array_layers: 1 },
        mip_level_count: 1,
        sample_count: 1,
        dimension: wgpu::TextureDimension::D2,
        format: wgpu::TextureFormat::Rgba32Float,
        usage: wgpu::TextureUsages::TEXTURE_BINDING | wgpu::TextureUsages::COPY_DST,
        view_formats: &[],
    });
    queue.write_texture(
        wgpu::TexelCopyTextureInfo { texture: &tex, mip_level: 0, origin: wgpu::Origin3d::ZERO, aspect: wgpu::TextureAspect::All },
        &bytes,
        wgpu::TexelCopyBufferLayout { offset: 0, bytes_per_row: Some(4 * 16), rows_per_image: None },
        wgpu::Extent3d { width: 4, height: 4, depth_or_array_layers: 1 },
    );
    tex.create_view(&wgpu::TextureViewDescriptor::default())
}

/// Simple deterministic pseudo-random [0, 1) from integer seed.
fn pseudo_random(seed: usize) -> f32 {
    let n = seed.wrapping_mul(0x5DEECE66D).wrapping_add(0xB) & 0xFFFFFF;
    n as f32 / 0xFFFFFF as f32
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_kernel_hemisphere() {
        let kernel = generate_kernel(64);
        assert_eq!(kernel.len(), 64);
        for k in &kernel {
            assert!(k[2] >= 0.0, "kernel z must be >= 0 (hemisphere), got {}", k[2]);
            let len = (k[0] * k[0] + k[1] * k[1] + k[2] * k[2]).sqrt();
            assert!(len <= 1.01, "kernel sample must be within unit hemisphere, len={}", len);
        }
    }

    #[test]
    fn test_noise_data() {
        let noise = generate_noise_data();
        assert_eq!(noise.len(), 16);
        for n in &noise {
            assert!((n[2]).abs() < 1e-5, "noise z should be 0");
            let len = (n[0] * n[0] + n[1] * n[1]).sqrt();
            assert!((len - 1.0).abs() < 0.1, "noise vector should be roughly unit length, got {}", len);
        }
    }

    #[test]
    fn test_ssgi_uniform_default() {
        let u = SsgiUniform::default();
        assert!((u.radius - 0.5).abs() < 1e-5);
        assert!((u.bias - 0.025).abs() < 1e-5);
    }
}

• Step 2: lib.rs에 ssgi 모듈 등록

pub mod ssgi;
pub use ssgi::{SsgiResources, SsgiUniform, SSGI_OUTPUT_FORMAT};

• Step 3: 빌드 + 테스트

Run: cargo test -p voltex_renderer Expected: 기존 20 + 3 = 23 PASS

• Step 4: 커밋

git add crates/voltex_renderer/src/ssgi.rs crates/voltex_renderer/src/lib.rs
git commit -m "feat(renderer): add SSGI resources with hemisphere kernel and noise texture"

---

Task 2: SSGI 셰이더 + 파이프라인

Files:

• Create: crates/voltex_renderer/src/ssgi_shader.wgsl

• Modify: crates/voltex_renderer/src/deferred_pipeline.rs

• Step 1: ssgi_shader.wgsl 작성

// SSGI pass: screen-space ambient occlusion + color bleeding
// Reads G-Buffer position/normal/albedo, outputs AO + indirect color

// Group 0: G-Buffer (same layout as lighting pass)
@group(0) @binding(0) var t_position: texture_2d<f32>;
@group(0) @binding(1) var t_normal: texture_2d<f32>;
@group(0) @binding(2) var t_albedo: texture_2d<f32>;
@group(0) @binding(3) var s_gbuffer: sampler;

// Group 1: SSGI data
struct SsgiUniform {
    projection: mat4x4<f32>,
    view: mat4x4<f32>,
    radius: f32,
    bias: f32,
    intensity: f32,
    indirect_strength: f32,
};

struct SsgiKernel {
    samples: array<vec4<f32>, 64>,
};

@group(1) @binding(0) var<uniform> ssgi: SsgiUniform;
@group(1) @binding(1) var<uniform> kernel: SsgiKernel;
@group(1) @binding(2) var t_noise: texture_2d<f32>;
@group(1) @binding(3) var s_noise: sampler;

struct VertexOutput {
    @builtin(position) clip_position: vec4<f32>,
    @location(0) uv: vec2<f32>,
};

@vertex
fn vs_main(@location(0) position: vec2<f32>) -> VertexOutput {
    var out: VertexOutput;
    out.clip_position = vec4<f32>(position, 0.0, 1.0);
    out.uv = vec2<f32>(position.x * 0.5 + 0.5, 1.0 - (position.y * 0.5 + 0.5));
    return out;
}

@fragment
fn fs_main(in: VertexOutput) -> @location(0) vec4<f32> {
    let uv = in.uv;

    let world_pos = textureSample(t_position, s_gbuffer, uv).xyz;

    // Skip background
    if dot(world_pos, world_pos) < 0.001 {
        return vec4<f32>(1.0, 0.0, 0.0, 0.0); // AO=1 (no occlusion), indirect=0
    }

    let world_normal = normalize(textureSample(t_normal, s_gbuffer, uv).xyz * 2.0 - 1.0);

    // Transform to view space
    let view_pos = (ssgi.view * vec4<f32>(world_pos, 1.0)).xyz;
    let view_normal = normalize((ssgi.view * vec4<f32>(world_normal, 0.0)).xyz);

    // Random rotation from noise texture (4x4 tiling)
    let tex_dims = textureDimensions(t_position);
    let noise_scale = vec2<f32>(f32(tex_dims.x) / 4.0, f32(tex_dims.y) / 4.0);
    let random_vec = textureSample(t_noise, s_noise, uv * noise_scale).xyz;

    // Construct TBN in view space using Gram-Schmidt
    let tangent = normalize(random_vec - view_normal * dot(random_vec, view_normal));
    let bitangent = cross(view_normal, tangent);
    let TBN = mat3x3<f32>(tangent, bitangent, view_normal);

    var occlusion = 0.0;
    var indirect = vec3<f32>(0.0);

    for (var i = 0u; i < 64u; i++) {
        // Sample position in view space
        let sample_offset = TBN * kernel.samples[i].xyz;
        let sample_view_pos = view_pos + sample_offset * ssgi.radius;

        // Project to screen UV
        let clip = ssgi.projection * vec4<f32>(sample_view_pos, 1.0);
        var screen_uv = clip.xy / clip.w * 0.5 + 0.5;
        screen_uv.y = 1.0 - screen_uv.y;

        // Clamp to valid range
        screen_uv = clamp(screen_uv, vec2<f32>(0.001), vec2<f32>(0.999));

        // Read actual position at that screen location
        let actual_world_pos = textureSample(t_position, s_gbuffer, screen_uv).xyz;
        let actual_view_pos = (ssgi.view * vec4<f32>(actual_world_pos, 1.0)).xyz;

        // Occlusion: is the actual geometry closer to camera than our sample?
        let depth_diff = sample_view_pos.z - actual_view_pos.z;
        let range_check = smoothstep(0.0, 1.0, ssgi.radius / (abs(view_pos.z - actual_view_pos.z) + 0.001));

        if depth_diff > ssgi.bias && depth_diff < ssgi.radius {
            occlusion += range_check;
            // Color bleeding: sample albedo at occluder position
            let sample_albedo = textureSample(t_albedo, s_gbuffer, screen_uv).rgb;
            indirect += sample_albedo * range_check;
        }
    }

    let ao = clamp(1.0 - (occlusion / 64.0) * ssgi.intensity, 0.0, 1.0);
    indirect = indirect / 64.0 * ssgi.indirect_strength;

    return vec4<f32>(ao, indirect);
}

• Step 2: deferred_pipeline.rs에 SSGI 파이프라인 함수 추가

Add to deferred_pipeline.rs:

use crate::ssgi::SSGI_OUTPUT_FORMAT;

/// SSGI pass: reads G-Buffer (group 0) + SSGI data (group 1)
pub fn ssgi_gbuffer_bind_group_layout(device: &wgpu::Device) -> wgpu::BindGroupLayout {
    device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
        label: Some("SSGI GBuffer BGL"),
        entries: &[
            // position (non-filterable)
            wgpu::BindGroupLayoutEntry {
                binding: 0,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Texture {
                    sample_type: wgpu::TextureSampleType::Float { filterable: false },
                    view_dimension: wgpu::TextureViewDimension::D2,
                    multisampled: false,
                },
                count: None,
            },
            // normal (filterable)
            wgpu::BindGroupLayoutEntry {
                binding: 1,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Texture {
                    sample_type: wgpu::TextureSampleType::Float { filterable: true },
                    view_dimension: wgpu::TextureViewDimension::D2,
                    multisampled: false,
                },
                count: None,
            },
            // albedo (filterable)
            wgpu::BindGroupLayoutEntry {
                binding: 2,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Texture {
                    sample_type: wgpu::TextureSampleType::Float { filterable: true },
                    view_dimension: wgpu::TextureViewDimension::D2,
                    multisampled: false,
                },
                count: None,
            },
            // sampler (non-filtering for position)
            wgpu::BindGroupLayoutEntry {
                binding: 3,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::NonFiltering),
                count: None,
            },
        ],
    })
}

pub fn ssgi_data_bind_group_layout(device: &wgpu::Device) -> wgpu::BindGroupLayout {
    device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
        label: Some("SSGI Data BGL"),
        entries: &[
            // SsgiUniform
            wgpu::BindGroupLayoutEntry {
                binding: 0,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Buffer {
                    ty: wgpu::BufferBindingType::Uniform,
                    has_dynamic_offset: false,
                    min_binding_size: None,
                },
                count: None,
            },
            // kernel (uniform buffer, 64 * vec4 = 1024 bytes)
            wgpu::BindGroupLayoutEntry {
                binding: 1,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Buffer {
                    ty: wgpu::BufferBindingType::Uniform,
                    has_dynamic_offset: false,
                    min_binding_size: None,
                },
                count: None,
            },
            // noise texture (non-filterable, Rgba32Float)
            wgpu::BindGroupLayoutEntry {
                binding: 2,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Texture {
                    sample_type: wgpu::TextureSampleType::Float { filterable: false },
                    view_dimension: wgpu::TextureViewDimension::D2,
                    multisampled: false,
                },
                count: None,
            },
            // noise sampler
            wgpu::BindGroupLayoutEntry {
                binding: 3,
                visibility: wgpu::ShaderStages::FRAGMENT,
                ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::NonFiltering),
                count: None,
            },
        ],
    })
}

pub fn create_ssgi_pipeline(
    device: &wgpu::Device,
    gbuffer_layout: &wgpu::BindGroupLayout,
    data_layout: &wgpu::BindGroupLayout,
) -> wgpu::RenderPipeline {
    let shader = device.create_shader_module(wgpu::ShaderModuleDescriptor {
        label: Some("SSGI Shader"),
        source: wgpu::ShaderSource::Wgsl(include_str!("ssgi_shader.wgsl").into()),
    });

    let layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
        label: Some("SSGI Pipeline Layout"),
        bind_group_layouts: &[gbuffer_layout, data_layout],
        immediate_size: 0,
    });

    device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
        label: Some("SSGI Pipeline"),
        layout: Some(&layout),
        vertex: wgpu::VertexState {
            module: &shader,
            entry_point: Some("vs_main"),
            buffers: &[FullscreenVertex::LAYOUT],
            compilation_options: wgpu::PipelineCompilationOptions::default(),
        },
        fragment: Some(wgpu::FragmentState {
            module: &shader,
            entry_point: Some("fs_main"),
            targets: &[Some(wgpu::ColorTargetState {
                format: SSGI_OUTPUT_FORMAT,
                blend: None,
                write_mask: wgpu::ColorWrites::ALL,
            })],
            compilation_options: wgpu::PipelineCompilationOptions::default(),
        }),
        primitive: wgpu::PrimitiveState {
            topology: wgpu::PrimitiveTopology::TriangleList,
            ..Default::default()
        },
        depth_stencil: None,
        multisample: wgpu::MultisampleState::default(),
        multiview_mask: None,
        cache: None,
    })
}

• Step 3: 빌드 확인

Run: cargo build -p voltex_renderer Expected: 컴파일 성공

• Step 4: 커밋

git add crates/voltex_renderer/src/ssgi_shader.wgsl crates/voltex_renderer/src/deferred_pipeline.rs
git commit -m "feat(renderer): add SSGI shader and pipeline for screen-space GI"

---

Task 3: Lighting Pass에 SSGI 통합

Files:

• Modify: crates/voltex_renderer/src/deferred_lighting.wgsl

• Modify: crates/voltex_renderer/src/deferred_pipeline.rs

• Step 1: lighting_shadow_bind_group_layout에 SSGI binding 추가

현재 lighting_shadow_bind_group_layout에 binding 0-4 (shadow+IBL). 여기에 추가:

// binding 5: SSGI output texture
wgpu::BindGroupLayoutEntry {
    binding: 5,
    visibility: wgpu::ShaderStages::FRAGMENT,
    ty: wgpu::BindingType::Texture {
        sample_type: wgpu::TextureSampleType::Float { filterable: true },
        view_dimension: wgpu::TextureViewDimension::D2,
        multisampled: false,
    },
    count: None,
},
// binding 6: SSGI sampler
wgpu::BindGroupLayoutEntry {
    binding: 6,
    visibility: wgpu::ShaderStages::FRAGMENT,
    ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::Filtering),
    count: None,
},

• Step 2: deferred_lighting.wgsl에 SSGI 바인딩 + 적용 추가

Group 2에 추가:

@group(2) @binding(5) var t_ssgi: texture_2d<f32>;
@group(2) @binding(6) var s_ssgi: sampler;

Fragment shader에서 ambient 계산 부분 변경:

// 기존: let ambient = (diffuse_ibl + specular_ibl) * ao;
// 변경:
let ssgi_data = textureSample(t_ssgi, s_ssgi, uv);
let ssgi_ao = ssgi_data.r;
let ssgi_indirect = ssgi_data.gba;
let ambient = (diffuse_ibl + specular_ibl) * ao * ssgi_ao + ssgi_indirect;

• Step 3: 빌드 확인

Run: cargo build -p voltex_renderer Expected: 컴파일 성공

• Step 4: 커밋

git add crates/voltex_renderer/src/deferred_lighting.wgsl crates/voltex_renderer/src/deferred_pipeline.rs
git commit -m "feat(renderer): integrate SSGI output into deferred lighting pass"

---

Task 4: deferred_demo에 SSGI 패스 통합

Files:

• Modify: examples/deferred_demo/src/main.rs

NOTE: 이 태스크는 기존 deferred_demo를 확장하여 3-pass 렌더링으로 변경합니다.

변경사항:

1. SsgiResources::new() 호출하여 SSGI 리소스 생성
2. SSGI 파이프라인 + 바인드 그룹 레이아웃 생성
3. SSGI 바인드 그룹 2개 생성 (G-Buffer + SSGI data)
4. 기존 Shadow+IBL 바인드 그룹에 SSGI output texture + sampler 추가 (binding 5,6)
5. 렌더 루프에 SSGI 패스 삽입 (Pass 2: SSGI, 기존 Lighting은 Pass 3으로)
6. 매 프레임 SsgiUniform 업데이트 (view, projection 행렬)
7. 리사이즈 시 SSGI 리소스 + 바인드 그룹 재생성

이 태스크는 deferred_demo의 전체 구조를 이해해야 하므로 opus 모델로 실행.

• Step 1: deferred_demo 수정

Read the current examples/deferred_demo/src/main.rs first, then add SSGI integration.

• Step 2: 빌드 확인

Run: cargo build --bin deferred_demo Expected: 컴파일 성공

• Step 3: 커밋

git add examples/deferred_demo/src/main.rs
git commit -m "feat(renderer): add SSGI pass to deferred_demo (AO + color bleeding)"

---

Task 5: 문서 업데이트

Files:

• Modify: docs/STATUS.md

• Modify: docs/DEFERRED.md

• Step 1: STATUS.md에 Phase 7-2 추가

Phase 7-1 아래에:

### Phase 7-2: SSGI (Screen-Space Global Illumination)
- voltex_renderer: SsgiResources (hemisphere kernel, 4x4 noise, output texture)
- voltex_renderer: SSGI shader (SSAO + color bleeding in one pass)
- voltex_renderer: SSGI pipeline + bind group layouts
- voltex_renderer: Lighting pass SSGI integration (ambient * ssgi_ao + indirect)
- deferred_demo updated with 3-pass rendering (GBuffer → SSGI → Lighting)

테스트 수 업데이트 (voltex_renderer: 23).

• Step 2: DEFERRED.md에 Phase 7-2 미뤄진 항목

## Phase 7-2

- **Bilateral Blur** — SSGI 노이즈 제거 블러 미구현. 4x4 노이즈 타일링만.
- **반해상도 렌더링** — 풀 해상도에서 SSGI 실행. 성능 최적화 미적용.
- **Temporal Accumulation** — 프레임 간 누적 미구현. 매 프레임 독립 계산.
- **Light Probes** — 베이크 기반 GI 미구현.

• Step 3: 커밋

git add docs/STATUS.md docs/DEFERRED.md
git commit -m "docs: add Phase 7-2 SSGI status and deferred items"