Deferred Shading Pipeline延迟渲染管线

GBuffer LayoutGBuffer 布局

NeoX's deferred path uses a compact 3-RT GBuffer for the standard mode, or a 5-RT Single-Pass Deferred for mobile (TBDR) optimization. In the new engine the GBuffer layout changed: Roughness moved from RT1 to RT2, and the RT names were updated. A new shared gbuffer_struct.hlsl header is now used across all GBuffer passes.

NeoX 的延迟路径标准模式使用紧凑的 3-RT GBuffer，或为移动端（TBDR）优化使用 5-RT 单 Pass 延迟。新引擎 GBuffer 布局有变化：Roughness 从 RT1 移到 RT2，RT 名称也已更新。新引擎所有 GBuffer Pass 现在使用共享的 gbuffer_struct.hlsl 头文件。

GBuffer Write — ps_main()GBuffer 写入 — ps_main()

The GBuffer pixel shader now includes gbuffer_struct.hlsl (a shared struct definition), and the packed layout changed — Roughness is in RT2.g and Flags are in RT2.a:

GBuffer 像素着色器现在包含 gbuffer_struct.hlsl（共享结构体定义），打包布局变化——Roughness 在 RT2.g，Flags 在 RT2.a：

// From gbuffer_struct.hlsl (new shared header)
struct GBuffer {
  float4 AlbedoAO                         : SV_Target0;
  float4 Normal                            : SV_Target1;  // was NormalRoughness
  float4 MetallicRoughnessShadingModelFlags : SV_Target2;  // was MetallicShadingModelFlags
};

GBuffer ps_main(Fragment input, bool is_front_face : SV_IsFrontFace) {
  #if ENABLE_VIRTUALTEXTURE_SAMPLEING
    SamplingVirtualTexture(input);         // NEW: VT sampling before material
  #endif
  RawData raw_data = GetRawData(input);
  Material mtl = (Material)0;
  SetupMaterial(input, raw_data, mtl);
  #if DEFERRED_DECAL_ENABLE && !ENABLE_VIRTUALTEXTURE_SAMPLEING
    DeferredApplyNewDecal(input, mtl);     // conditional guard added
  #endif

  GBuffer gbuffer;
  gbuffer.AlbedoAO = float4(mtl.albedo, mtl.occlusion);
  gbuffer.Normal = float4(0.5 * (mtl.normal + 1.0), 0);  // A channel now free

  // Encode flags into alpha byte (was green byte in old engine)
  uint flags = 0;
  flags = EncodeShadingModel(flags, mtl.shading_model);
  flags = EncodeReceiveShadow(flags);
  #if SLPV_ENABLE && LIGHT_PROBE_ENABLE          // changed condition
    flags = EncodeClusteredShadingLightProbeVolumeEnable(flags);
  #endif

  gbuffer.MetallicRoughnessShadingModelFlags = float4(
    mtl.metalness,
    mtl.roughness,                        // NEW: Roughness in .g (was in RT1.a)
    0,
    float(flags) / 255.0                   // Flags in .a (was in .g)
  );
  return gbuffer;
}

Single-Pass Deferred — Mobile Optimization单 Pass 延迟 — 移动端优化

On tile-based deferred rendering (TBDR) GPUs (iOS/Android), NeoX supports Single-Pass Deferred Shading. This writes both GBuffer data AND lighting result in one pass, avoiding the costly off-chip GBuffer readback that destroys TBDR benefits.

在基于 Tile 的延迟渲染（TBDR）GPU 上（iOS/Android），NeoX 支持单 Pass 延迟着色。在一个 Pass 中同时写入 GBuffer 数据和光照结果，避免了破坏 TBDR 优势的昂贵离片 GBuffer 回读。

// Single-pass mode: output emissive directly + manual depth
#if SINGLE_PASS_DEFERRED_SHADING_ENABLE
  gbuffer.FinalResult = float4(mtl.emissive, 1);
  #ifdef NEOX_GLSL
    gbuffer.Depth = input.position.z;           // single float
  #else
    gbuffer.Depth = float4(input.position.z);   // packed
  #endif
#endif

Flag Encoding System标志位编码系统

The Alpha channel of MetallicRoughnessShadingModelFlags packs multiple boolean flags into a single byte (0-255). In the new engine the flags moved from the Green to the Alpha channel. Defined in flags.hlsl:

MetallicRoughnessShadingModelFlags 的 Alpha 通道将多个布尔标志打包到单个字节（0-255）中。新引擎中标志从 Green 通道移到 Alpha 通道。定义在 flags.hlsl：

NeoX vs UE5 GBuffer ComparisonNeoX vs UE5 GBuffer 对比