試做自己的Unity SRP

目標：先以LightWeight SRP為參考，並在過程中更理解Unity繪圖流程的細節，最終設計出自己覺得適合的SRP。

架構流程如下：

Light List相關(Setup light constants pass)：
從代碼得知，Culling list裡的Light list是目前camera view所看到的lights，經過最大燈光的
數量的限制及SetGlobalVetorArray會造成物件燈光計算的錯誤，因為是view culling list，並不是Object List…
void SetupAdditionalLightConstants(CommandBuffer cmd, ref LightData lightData){
            List lights = lightData.visibleLights;
            if (lightData.totalAdditionalLightsCount > 0)
            {
                int localLightsCount = 0;
                for (int i = 0; i < lights.Count && localLightsCount < maxVisibleLocalLights; ++i)
                {
                    VisibleLight light = lights[i];
                    if (light.lightType != LightType.Directional)
                    {
                        InitializeLightConstants(lights, i, out m_LightPositions[localLightsCount],
                            out m_LightColors[localLightsCount],
                            out m_LightDistanceAttenuations[localLightsCount],
                            out m_LightSpotDirections[localLightsCount],
                            out m_LightSpotAttenuations[localLightsCount]);
                        localLightsCount++;
                    }
                }

                cmd.SetGlobalVector(LightConstantBuffer._AdditionalLightCount, new Vector4(lightData.pixelAdditionalLightsCount,
                    lightData.totalAdditionalLightsCount, 0.0f, 0.0f));

                // if not using a compute buffer, engine will set indices in 2 vec4 constants
                // unity_4LightIndices0 and unity_4LightIndices1
                if (perObjectLightIndices != null)
                    cmd.SetGlobalBuffer("_LightIndexBuffer", perObjectLightIndices);
            }
            else
            {
                cmd.SetGlobalVector(LightConstantBuffer._AdditionalLightCount, Vector4.zero);
            }
            cmd.SetGlobalVectorArray(LightConstantBuffer._AdditionalLightPosition, m_LightPositions);
            cmd.SetGlobalVectorArray(LightConstantBuffer._AdditionalLightColor, m_LightColors);
            cmd.SetGlobalVectorArray(LightConstantBuffer._AdditionalLightDistanceAttenuation, m_LightDistanceAttenuations);
            cmd.SetGlobalVectorArray(LightConstantBuffer._AdditionalLightSpotDir, m_LightSpotDirections);
            cmd.SetGlobalVectorArray(LightConstantBuffer._AdditionalLightSpotAttenuation, m_LightSpotAttenuations);
        }

Object light list的處理：原來Unity底層有支援RendererConfiguration.PerObjectLightIndices8，
可以用來解決這個問題，如果有支援StructuredBuffer的話，可以走另一個流程
RendererConfiguration.ProvideLightIndices，相關代碼如下：
//這裡主要是在處理方向光需要從project light indices中移掉，交由MainLight接手。
public void SetupPerObjectLightIndices(ref CullResults cullResults, ref LightData lightData)
{
...
}


public static RendererConfiguration GetRendererConfiguration(int localLightsCount){
            RendererConfiguration configuration = RendererConfiguration.PerObjectReflectionProbes | RendererConfiguration.PerObjectLightmaps | RendererConfiguration.PerObjectLightProbe;
            if (localLightsCount > 0)
            {
                if (useComputeBufferForPerObjectLightIndices)
                    configuration |= RendererConfiguration.ProvideLightIndices;
                else
                    configuration |= RendererConfiguration.PerObjectLightIndices8;
            }

            return configuration;
        }

相關Shader如下：會利用GetLight這個函式來取得正確的Light index。
Light GetLight(half i, float3 positionWS)
{
    LightInput lightInput;

#if USE_STRUCTURED_BUFFER_FOR_LIGHT_DATA
    int lightIndex = _LightIndexBuffer[unity_LightIndicesOffsetAndCount.x + i];
#else
    // The following code is more optimal than indexing unity_4LightIndices0.
    // Conditional moves are branch free even on mali-400
    half i_rem = (i < 2.0h) ? i : i - 2.0h;
    half2 lightIndex2 = (i < 2.0h) ? unity_4LightIndices0.xy : unity_4LightIndices0.zw;
    int lightIndex = (i_rem < 1.0h) ? lightIndex2.x : lightIndex2.y;
#endif

    // The following code will turn into a branching madhouse on platforms that don't support
    // dynamic indexing. Ideally we need to configure light data at a cluster of
    // objects granularity level. We will only be able to do that when scriptable culling kicks in.
    // TODO: Use StructuredBuffer on PC/Console and profile access speed on mobile that support it.
    lightInput.position = _AdditionalLightPosition[lightIndex];
    lightInput.color = _AdditionalLightColor[lightIndex].rgb;
    lightInput.distanceAttenuation = _AdditionalLightDistanceAttenuation[lightIndex];
    lightInput.spotDirection = _AdditionalLightSpotDir[lightIndex];
    lightInput.spotAttenuation = _AdditionalLightSpotAttenuation[lightIndex];

    half4 directionAndRealtimeAttenuation = GetLightDirectionAndAttenuation(lightInput, positionWS);

    Light light;
    light.index = lightIndex;
    light.direction = directionAndRealtimeAttenuation.xyz;
    light.attenuation = directionAndRealtimeAttenuation.w;
    light.subtractiveModeAttenuation = lightInput.distanceAttenuation.w;
    light.color = lightInput.color;

    return light;
}

切割Pass原因：
主要是為容易切割處理流程、共用程式碼、以及提供足夠的時機做相關後置特效處理。另外也利用了切pass的時機，copy了render過程的結果，以便於除錯及觀察繪製流程。

接下來要做的事：

1. 影子pass處理，預計會採用跟Unity一樣的CSM shadow map全場景投影的做法
2. PBR材質
3. 支持forward plus rendering，目前Unity是將這部份放在HRDP，估計應該是因為
   compute shader light list的問題，如果最後不行的話，最差的狀況至少也可以先為
   Vulkan做準備。

除錯資料（左上開始的小圖），由左而右，是繪製流程的順序，依序為
after draw opaques, depth, after draw sky, after draw transparencies.
另外也跟LWSRP一樣支持render scale，簡單來說就是不直接Render到viewport上，而是先
畫到RenderTexture，最後再Final Blt到viewport上，好處是：

1. 讓3D採用較低的解析度，可以大大減少GPU處理像素的數量，提升效能。
2. 可以直接使用NativeDepthBuffer，不用增加DrawCall。
3. 由GUI一般會使用point filter，而3D大都會採用linear filter，可以切開。

相關影片如下：




Dream continues in...