Problems with TFLite operator performance profiling

[pkgoogle]

Original Issue: tensorflow/tensorflow#83331
Original Author: @4570235

Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

binary

TensorFlow version

tf 2.16.1

Custom code

No

OS platform and distribution

No response

Mobile device

Android Google Pixel 6

Python version

No response

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

I want to profile the operator performance of my custom tflite model in android phone 'Google Pixel 6'.
I have followed the tutorial measurement and built a very simple test program [source code]. However, I encounter some problems.

Firstly, when I tested the model 'mobilenetv1.tflite' from tflite example image_classification, if my program depends on the newest version of TFLite

implementation 'org.tensorflow:tensorflow-lite:2.16.1'
implementation 'org.tensorflow:tensorflow-lite-support:0.4.4'

then the Android Studio CPU Profiler shows

I cannot see all the operators. After some testing, I figure out that the maximum version of TFLite that works is tensorflow-lite:2.12.0 with tensorflow-lite-support:0.4.0.

implementation 'org.tensorflow:tensorflow-lite:2.12.0'
implementation 'org.tensorflow:tensorflow-lite-support:0.4.0'

Here's the result:

So, is this a bug? How can I trace operator performance with the newest version of TFLite?

Secondly, with tensorflow-lite:2.12.0 and tensorflow-lite-support:0.4.0 mentioned above, I tried to trace my custom model which is very similar to Qualcomm's 'quicksrnetsmall.tflite'. Let's take 'quicksrnetsmall.tflite' as an sample and the Profiler shows

I cannot see any operators. How can I fix it?

Standalone code to reproduce the issue

https://github.com/4570235/LiteRTStarter/blob/master/app/src/main/java/com/handleychen/litertstarter/Benchmark.java

Relevant log output

No response

[pkgoogle]

Hi @4570235, can you update to the LiteRT packages and see if this works or not? Thanks.