Unity performance optimization direction

1 Cpu

1.1 Rendering

1.1.1 Reduce drawcall

adopt batch Reduce ; Note that the bus bandwidth should not be allowed after approval gpu overpressure
Reduce the number of rendered objects ;

1.1.2 resource optimization

Irregular grid 、 Texture and so on

1.1.3 LOD

High accuracy for close objects , Low precision model for long-distance objects

1.1.4 Occlusion culling

Objects that are completely occluded are not rendered

1.1.5 Long distance culling

Objects too far away are not rendered , It is also applicable to calculation , such as MC Block mechanism

1.2 UI

1.2.1 Combined batch

UI Pay attention to atlas allocation , There are also some special components like mask、outline And so on will interrupt the approval

1.2.2 The reconstruction

One node update will lead to the whole canvas The reconstruction , Although each of them canvas Occupied a draw call, But also be careful not to pile up too many irrelevant modules in the same canvas Lead to frequent reconstruction .

1.3 load

1.3.1 Inclusion dependency

If there is no correct inclusion dependency , It will cause the public resources to be packaged N Time

1.3.2 Preloading

If the resource is not loaded ,Instantiate Resources will be loaded before instantiation , Will slow down the speed , So we need to preload these resources when loading levels according to the configuration . Anyway loading Medium slow is slow .

1.3.3 Code efficiency

The algorithm takes time complexity into consideration ,update Be careful about the calculation in , Don't count the cached values every time .

1.3.4 Object pool optimization

Destory Objects also drag cpu, Consider object pool reuse , Slow release pool to slowly release . Destroying a large number of objects at the same time will jam

2 Gpu

2.1 Reduce rendering pixels

2.1.1 Map optimization Reduce the area

Remove transparent invalid areas

2.1.2 Add vertex Reduce the area

2.1.3 Customize shader Reduce the number of faces

Avoid overlapping multiple textures , Customize shader Sample from multiple textures

2.1.4 UI

The nine palaces are hollowed out , avoid mask, Eliminate transparent buttons

2.1.5 Reduce shader Complexity

2.1.6 HSR Optimize

2.1.7 Reduce unnecessary post-treatment

2.1.8 Mesh Instead of Alpha Test

2.1.9 Bandwidth bottleneck

1. Reduce Cache miss
2. Turn on mipmap
3. Use hardware compression format , Try to use a larger compression ratio in the acceptable range
4. Use the nine palace grid sprite Reduce UI Map size
5. Check the map , Avoid too large maps
6. Reduce Lightmap Size
7. Use Occlusion culling Reduce the render target
8. Use LOD grid
9. Use gpuinstancing Instead of mesh merging
10. Reduce resolution
11. Try to use constant registers gpu instancing

3 Memory

Memory allocation Mono33% texture 33% other 33%

3.1 Resource memory

3.1.1 texture

3.1.1.1 Texture compression format

consider ETC2,ASTC

3.1.1.2 Texture size

Can be small

3.1.1.3 mipmap

Generally used for scenes and characters , Memory will increase 1.33 times , Pay attention to that UI Invalid ,UI Don't open

3.1.1.4 Read & Write

Turning on read and write will double the memory consumption

3.1.1.5 Resource redundancy

Instantiate and modify materials 、ab Package policies and so on cause resources to occupy memory repeatedly

3.1.2 grid

3.1.2.1 Normal、Color and Tangent

Don't import unneeded attributes

3.2 The memory of the engine module itself

3.2.1 ab Package instance release

reasonable ab Package management policy , Avoid memory leaks

3.2.2 Reasonable component management

For abandoned objects, clean them up at the first time

3.3 Managed heap memory

3.3.1 Avoid allocating large amounts of unnecessary memory

Mono Virtual machine will not reduce memory consumption after we release memory , That is to say, only increase but not decrease , Avoid improving Mono Peak in memory .

3.3.2 Avoid allocating memory with high-frequency functions

Various update Be careful when allocating memory in