Unity performance optimization reading notes - explore performance issues -profiler (2.1)

The author talked a lot .. About the importance of finding problems and locating root causes . I'll skip it here

The main contents of this chapter are as follows ：

1 How to use profiler Get running data

2 How to analyze profiler Analyze the performance bottleneck of data

3 How to analyze performance problems and locate root causes

Then I introduced profiler The function of ,profiler The functions of each subsystem of are as follows ：

1 see cpu Consume

2 About rendering and gpu Basis and details of

3 Memory application and overall consumption

4 Sound resource / Data usage

5 Network messaging and operational utilization

6 Video playback utilization

7 About ui Fundamentals and details of performance

8 Global illumination data

Then the author introduces two ways of observing performance , One is to use monitoring tools , One is to benchmark .

The detection tool can well observe the internal implementation of the program , Function calling behavior , And how the memory is applied for and other running data . But when you start looking for performance problems, you use detection tools , It's not a good way , Because the detection tool itself will have some performance consumption . When compiling games in development mode , Additional compilation options will be enabled , Cause the system to call some special events . These events will be profiler Record and save . This will result in additional cpu And memory consumption . If it's in editor Perform performance test on , Things will get worse . because editor You also need to render some additional windows and handle background tasks . Especially in large projects , Use profiler It may lead to different performance of the application . The timing of event generation and the change of asynchronous invocation will lead to some exceptions . This is in use profiler Some costs that must be paid when performing performance troubleshooting , So it's using profiler Pay special attention to this point when .

Before starting to analyze each line of code , It is best to run on the test equipment first , Under specific test samples , Get some preliminary operation data . The test sample can be playing a game for a few seconds , Play of transition animation , Or play games in some levels . Objective To get a general user experience , And carefully observe when the performance gets worse , These problems sometimes need further analysis .

This process is often called benchmarking . The indicators we need to pay attention to are FPS,CPU Operating condition ,CPU and GPU And so on . These are better performance indicators .

After benchmarking , And when in-depth analysis is needed , It can be used profiler Performance testing . And try to test as many models as possible to obtain operation data . And don't use editor To test , Because there will be additional performance consumption .

Here's a hint , Sometimes I find editor On the contrary, the running index of the real machine is better . as a result of editor Will cache some sound effects ,prefab Data like this , It will run faster .

In conclusion, it is , First run it on the real machine to see if there is a performance bottleneck , And then in Profiler Find and locate the problem on .Development Build There will be additional performance consumption , use Profiler There will be some differences between the viewed data and the actual data . And try not to Ediotr Perform performance test on , because Editor It also has performance consumption .