该项目为使用Deferred render path的URP加入SSAO的支持，修改成延迟渲染后，发现URP原来的SSAO贴图的渲染是正常的，但是最终画面没有SSAO效果：
所有应该是在延迟渲染的光照计算中没有考虑SSAO，考虑在Fragment Shader中直接采样贴图实现，先要找到URP的Deferred中计算光照的位置。
直接参考Forward中的SSAO实现方式修改代码：
// "Hidden/Universal Render Pipeline/StencilDeferred"
half4 DeferredShading(Varyings input) : SV_Target
{
UNITY_SETUP_INSTANCE_ID(input);
UNITY_SETUP_STEREO_EYE_INDEX_POST_VERTEX(input);
// Using SAMPLE_TEXTURE2D is faster than using LOAD_TEXTURE2D on iOS platforms (5% faster shader).
// Possible reason: HLSLcc upcasts Load() operation to float, which doesn't happen for Sample()?
float2 screen_uv = (input.screenUV.xy / input.screenUV.z);
float d = SAMPLE_TEXTURE2D_X_LOD(_CameraDepthTexture, my_point_clamp_sampler, screen_uv, 0).x;
// raw depth value has UNITY_REVERSED_Z applied on most platforms.
half4 gbuffer0 = SAMPLE_TEXTURE2D_X_LOD(_GBuffer0, my_point_clamp_sampler, screen_uv, 0);
half4 gbuffer1 = SAMPLE_TEXTURE2D_X_LOD(_GBuffer1, my_point_clamp_sampler, screen_uv, 0);
half4 gbuffer2 = SAMPLE_TEXTURE2D_X_LOD(_GBuffer2, my_point_clamp_sampler, screen_uv, 0);
//SSAO here
#if defined(_SCREEN_SPACE_OCCLUSION)
AmbientOcclusionFactor aoFactor;
//采样SSAO贴图
aoFactor.indirectAmbientOcclusion = SAMPLE_TEXTURE2D_X_LOD(_ScreenSpaceOcclusionTexture, my_point_clamp_sampler,screen_uv, 0);
aoFactor.directAmbientOcclusion = lerp(1.0, aoFactor.indirectAmbientOcclusion, _AmbientOcclusionParam.w);
#endif
#ifdef _DEFERRED_SUBTRACTIVE_LIGHTING
half4 gbuffer4 = SAMPLE_TEXTURE2D_X_LOD(_GBuffer4, my_point_clamp_sampler, screen_uv, 0);
half4 shadowMask = gbuffer4;
#else
half4 shadowMask = 1.0;
#endif
......
// 参考Forward中的方式，使得AO影响直接光和间接光
#if defined(_SCREEN_SPACE_OCCLUSION)
// 直接光：
unityLight.color *= aoFactor.directAmbientOcclusion;
// 间接光：
inputData.bakedGI *= aoFactor.indirectAmbientOcclusion;
#endif但是修改后发现相关跟Forward相比更加明亮：
仔细研究源码发现，inputData.bakedGI *= aoFactor.indirectAmbientOcclusion; 这行代码是不起作用的，在StencilDeferred中inputData是一直为零的，那么GI去了哪里呢？
FragmentOutput output;
output.GBuffer0 = half4(brdfData.diffuse.rgb, PackMaterialFlags(materialFlags)); // diffuse diffuse diffuse materialFlags (sRGB rendertarget)
output.GBuffer1 = half4(specular, brdfData.reflectivity); // specular specular specular reflectivity (sRGB rendertarget)
output.GBuffer2 = half4(packedNormalWS, packedSmoothness); // encoded-normal encoded-normal encoded-normal smoothness
output.GBuffer3 = half4(globalIllumination, 0); // GI GI GI [not_available] (lighting buffer)GI是保存在第四个GBuffer里面的，我们来找找urp对GBuffer3的操作：
？？？只有五行提到了？
再来看看StencilDeferred中的InputData是如何提取的：
InputData InputDataFromGbufferAndWorldPosition(half4 gbuffer2, float3 wsPos)
{
InputData inputData;
inputData.positionWS = wsPos;
#if _GBUFFER_NORMALS_OCT
half2 remappedOctNormalWS = Unpack888ToFloat2(gbuffer2.xyz); // values between [ 0, 1]
half2 octNormalWS = remappedOctNormalWS.xy * 2.0h - 1.0h; // values between [-1, +1]
inputData.normalWS = UnpackNormalOctQuadEncode(octNormalWS);
#else
inputData.normalWS = normalize(gbuffer2.xyz); // values between [-1, +1]
#endif
inputData.viewDirectionWS = SafeNormalize(GetWorldSpaceViewDir(wsPos.xyz));
// TODO: pass this info?
inputData.shadowCoord = (float4)0;
inputData.fogCoord = (half )0;
inputData.vertexLighting = (half3 )0;
inputData.bakedGI = (half3)0; // Note: this is not made available at lighting pass in this renderer - bakedGI contribution is included (with emission) in the value GBuffer3.rgb, that is used as a renderTarget during lighting
return inputData;
}原来GBuffer3是作为光照计算的renderTarget ，怪不得在光照计算的shader中全程没有提到GBuffer3了！
GBuffer3作为RenderTarget：
// ScriptableRenderPass
public override void Configure(CommandBuffer cmd, RenderTextureDescriptor cameraTextureDescripor)
{
// GBuffer3 -> GI -> m_DeferredLights.GbufferAttachmentIdentifiers[m_DeferredLights.GBufferLightingIndex]
RenderTargetIdentifier lightingAttachmentId = m_DeferredLights.GbufferAttachmentIdentifiers[m_DeferredLights.GBufferLightingIndex];
RenderTargetIdentifier depthAttachmentId = m_DeferredLights.DepthAttachmentIdentifier;
// TODO: Cannot currently bind depth texture as read-only!
ConfigureTarget(lightingAttachmentId, depthAttachmentId);
}
// ScriptableRenderPass
public override void Execute(ScriptableRenderContext context, ref RenderingData renderingData)
{
//在G-Buffer 3 的基础上进行其他光照的计算
m_DeferredLights.ExecuteDeferredPass(context, ref renderingData);
}看起来好像是先计算GI，然后再把其他光源和GI进行叠加，再来看看其他光源是怎么计算和叠加的：
// DeferredLights.cs
void RenderStencilDirectionalLights(CommandBuffer cmd, ref RenderingData renderingData, NativeArray<VisibleLight> visibleLights, int mainLightIndex)
{
cmd.EnableShaderKeyword(ShaderKeywordStrings._DIRECTIONAL);
// Directional lights.
// TODO bundle extra directional lights rendering by batches of 8.
// Also separate shadow caster lights from non-shadow caster.
for (int soffset = m_stencilVisLightOffsets[(int)LightType.Directional]; soffset < m_stencilVisLights.Length; ++soffset)
{
ushort visLightIndex = m_stencilVisLights[soffset];
VisibleLight vl = visibleLights[visLightIndex];
if (vl.lightType != LightType.Directional)
break;
Vector4 lightDir, lightColor, lightAttenuation, lightSpotDir, lightOcclusionChannel;
UniversalRenderPipeline.InitializeLightConstants_Common(visibleLights, visLightIndex, out lightDir, out lightColor, out lightAttenuation, out lightSpotDir, out lightOcclusionChannel);
int lightFlags = 0;
if (vl.light.bakingOutput.lightmapBakeType == LightmapBakeType.Mixed)
lightFlags |= (int)LightFlag.SubtractiveMixedLighting;
// Setup shadow paramters:
// - for the main light, they have already been setup globally, so nothing to do.
// - for other directional lights, it is actually not supported by URP, but the code would look like this.
bool hasDeferredShadows;
if (visLightIndex == mainLightIndex)
{
hasDeferredShadows = vl.light && vl.light.shadows != LightShadows.None;
cmd.DisableShaderKeyword(ShaderKeywordStrings._DEFERRED_ADDITIONAL_LIGHT_SHADOWS);
}
else
{
int shadowLightIndex = m_AdditionalLightsShadowCasterPass != null ? m_AdditionalLightsShadowCasterPass.GetShadowLightIndexFromLightIndex(visLightIndex) : -1;
hasDeferredShadows = vl.light && vl.light.shadows != LightShadows.None && shadowLightIndex >= 0;
CoreUtils.SetKeyword(cmd, ShaderKeywordStrings._DEFERRED_ADDITIONAL_LIGHT_SHADOWS, hasDeferredShadows);
cmd.SetGlobalInt(ShaderConstants._ShadowLightIndex, shadowLightIndex);
}
bool hasSoftShadow = hasDeferredShadows && renderingData.shadowData.supportsSoftShadows && vl.light.shadows == LightShadows.Soft;
CoreUtils.SetKeyword(cmd, ShaderKeywordStrings.SoftShadows, hasSoftShadow);
cmd.SetGlobalVector(ShaderConstants._LightColor, lightColor); // VisibleLight.finalColor already returns color in active color space
cmd.SetGlobalVector(ShaderConstants._LightDirection, lightDir);
cmd.SetGlobalInt(ShaderConstants._LightFlags, lightFlags);
// Lighting pass.
cmd.DrawMesh(m_FullscreenMesh, Matrix4x4.identity, m_StencilDeferredMaterial, 0, 3); // Lit
cmd.DrawMesh(m_FullscreenMesh, Matrix4x4.identity, m_StencilDeferredMaterial, 0, 4); // SimpleLit
}
cmd.DisableShaderKeyword(ShaderKeywordStrings._DEFERRED_ADDITIONAL_LIGHT_SHADOWS);
cmd.DisableShaderKeyword(ShaderKeywordStrings.SoftShadows);
cmd.DisableShaderKeyword(ShaderKeywordStrings._DIRECTIONAL);
}奇怪的是，在得到的最总结果中，光照是绘制了一个全屏的面片：
// Lighting pass.
cmd.DrawMesh(m_FullscreenMesh, Matrix4x4.identity, m_StencilDeferredMaterial, 0, 3); // Lit
cmd.DrawMesh(m_FullscreenMesh, Matrix4x4.identity, m_StencilDeferredMaterial, 0, 4); // SimpleLit而他的输出是不透明的：
这不就直接覆盖了吗，怎么做到的混合呢？
原来是再shader里面设置了Blend命令进行了混合，终于可以梳理下整个流程了：
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent