How does TFLite observe the operators in the model running on which kernel in the phone CPU?

[Jake Mao]

How does TFLite observe the operators in the model running on which kernel in the phone CPU? I use perfetto to observe, but only the inference time of the entire model does not include the operation time of individual operators such as convolution and running on that kernel. The TFLite version I am using is 2.9.0, and I manually compile and generate aar to import into the Android app for image recognition.