search

koenvervloesem / tensorflow-addons-on-arm

Build TensorFlow Addons for ARM on the Raspberry Pi
MIT License
5 stars 0 forks source link

think its running with all 64bits koen so maybe try with armv7l #4

Open StuartIanNaylor opened 3 years ago

StuartIanNaylor commented 3 years ago

https://github.com/Qengineering/TensorFlow-Raspberry-Pi_64-bit/issues/3

I just copied what you had done but the ld linker could not find _pywrap_tensorflow_internal.so

so ld.gold -l_pywrap_tensorflow_internal --verbose

Then I just sym linked lib_pywrap_tensorflow_internal.so to its actual location and thing seem to work

koenvervloesem commented 3 years ago

Nice, thanks for the tip! I still have to try TensorFlow and TensorFlow Addons on aarch64, I'll see if I can find some time in a few weeks.

StuartIanNaylor commented 3 years ago

I didn't seem to find any reason for bazel build --enable_runfiles --linkopt=-latomic build_pip_pkg

After the sym link 

# This script links project with TensorFlow dependency
python3 ./configure.py

bazel build build_pip_pkg
bazel-bin/build_pip_pkg artifacts

pip install artifacts/tensorflow_addons-*.whl

As the repo https://github.com/tensorflow/addons

https://github.com/Qengineering/TensorFlow-Raspberry-Pi_64-bit/issues/3#issuecomment-791944393

def is_raspi_arm():
    return (os.uname()[4] == "armv7l") or (os.uname()[4] == "aarch64")  

    if is_macos() or is_linux():
#        write("build --copt=-mavx") <-not known in gcc aarch64 
        write("build --cxxopt=-std=c++14")
        write("build --host_cxxopt=-std=c++14")