Performance Optimization - Rendering Optimization Notes

Two basic parameters that affect rendering efficiency: DrawCall and Triangle
The subject range (5%~95%) of Batch on medium and low-end machines is controlled within [0,250]
The number of faces: lod to reduceRendering faces
Camera.Render function stack analysis:
The high consumption of RenderForward.RenderLoopJob is usually due to the high number of Batches.
Culling time is reasonable in the range of 10%~20%.If it is high, you can check several aspects
If the number of GameObject small objects is too large, you can use dynamic loading, block display, CullingGroup Culling Distance and other methods to optimize the time consuming of Culling
If multi-threaded rendering is used in the project and it is enabledOcclusionCulling usually leads to excessive pressure on sub-threads, resulting in high overall Culling. Although OcclusionCulling reduces rendering consumption, its own consumption is also worth noting. The R&D team needs to test the consumption and then determine whether to enable this function.
    Render.Mesh CorrespondingIt is the time-consuming rendering that cannot be batched, and its number of calls corresponds to the corresponding number of Batches
    ParticleSystem.ScheduleGeometryJobs: It means that the main thread needs to wait for the sub-thread to calculate the position of the Particle before Culling.It is recommended to reduce the complexity of particles
      ParticleSystem.Draw: the number of DrawCalls of particles If the function is too high, you can consider reducing the number of particle systems [Real Machine Test->Memory Management->Specific Resource Information->Particle System]
You can reduce the interleaving of particle rendering through the atlas and adjusting the Order in Layer
    Shader.CreateGPUProgram: The consumption of the shader when it is rendered for the first time is related to the complexity of the shader. For this, the Shader can be preloaded through the ShaderVariantCollection, and after loadingTrigger Shader.CreateGPUProgram through ShaderVariantCollection.WarmUp, and cache this csv to avoid repeated calls
    "An idea for Shader variant collection, packaging and compilation optimization" https://answer.uwa4d.com/question/5da86670e84db43d6efbda72
Turn on multi-threaded rendering: Since uwa online report CPU time usage only counts the time taken by the main thread, if multi-threaded rendering is turned on, it is recommended to submit two versions for internal testing to facilitate analysis and rendering.
GPU Instancing: for grassTrees, buildings and other things that appear repeatedly in the scene have been significantly optimized.
SPR Batcher: Note: Shader needs to be compatible with SRP Particles are not supported Shader variants will interrupt DC batching