Supportment for 16KB page sizes on other processor architectures #43
Description
Issue type
Feature Request
Have you reproduced the bug with TensorFlow Nightly?
Yes
Source
source
TensorFlow version
tensorflow-lite 2.16.1
Custom code
Yes
OS platform and distribution
Android 15
Mobile device
No response
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?
The tensorflow-lite does not support 16KB page sizes on x86, x86_64, armeavi-v7a architecture.
Android 15, which will be released next year, supports 16 KB page sizes.
I'm going to introduce it in my app to improve performance.
However, tensorflow-lite does not seem to support 16 KB page sizes.
Is it possible to support 16 KB page sizes?
My app supports the architecture below.
- x86
- x86_64
- arm64-v8a
- armeabi-v7a
I found that support for arm64-v8a was supported through the following issues.
tensorflow/tensorflow#69459
Please review whether it is possible to support x86, x86_64, armeavi-v7a architecture also.
The android guide related to this is as follows.
https://developer.android.com/guide/practices/page-sizes
Standalone code to reproduce the issue
1. APK with tensorflow-lite v2.16.1
2. Create an alignment.sh file for 16KB alignment test with reference to the link below
https://developer.android.com/guide/practices/page-sizes?hl=en#test
3. Run below command
$ ./alignment.sh ${UNZIPPED_APK_FOLDER}
currently the output is below,
apk/lib/x86/libtensorflowlite_jni.so: UNALIGNED (2**12)
apk/lib/x86_64/libtensorflowlite_jni.so: UNALIGNED (2**12)
apk/lib/arm64-v8a/libtensorflowlite_jni.so: UNALIGNED (2**12)
apk/lib/armeabi-v7a/libtensorflowlite_jni.so: UNALIGNED (2**12)Relevant log output
No response