TDA4VM: Cannot compile any sample model in EdgeAI TIDL Tools using onnxrt_ep.py and tflrt_delegate.py

Hi,

Could you please let us know on which branch you are ?

You can use below command : 

git branch

Regards,

Pratik

You may want to look at https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1220814/tda4vm-bug-reports-for-tidl.

I got dozens of segfaults because I recompiled TIDL with optimisation using newer tools, and they were caused by these bugs (mostly the missing return statements interacting badly with the compiler's optimisation levels).

I'm on 'rel_8_6'

Hi,

Thanks for confirmation.

Could you please disable the AVX support and check again.

You can set TIDL_RT_AVX_REF to 0 to disable avx support, refer below command for the same.

export TIDL_RT_AVX_REF=0

Regards,

Pratik

I tried and nothing changes, I still get both the same outcomes with the two scripts.

On a more general note, following the above quoted discussion I tried to use docker over Windows and over Ubuntu 18.04, and in both cases it doesn't work.
With Windows Docker the setup script fails installing onnx:

root@81b10238f65f:/home/root# source ./setup.sh --skip_cpp_deps
X64 Architecture
Installing python packages...
Collecting git+https://github.com/kumardesappan/caffe2onnx (from -r ./requirements_pc.txt (line 12))
  Cloning https://github.com/kumardesappan/caffe2onnx to /tmp/pip-co6ws7zn-build
Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from -r ./requirements_pc.txt (line 1))
Collecting pyyaml (from -r ./requirements_pc.txt (line 2))
  Using cached https://files.pythonhosted.org/packages/b3/85/79b9e5b4e8d3c0ac657f4e8617713cca8408f6cdc65d2ee6554217cedff1/PyYAML-6.0-cp36-cp36m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl
Collecting protobuf==3.8.0 (from -r ./requirements_pc.txt (line 3))
  Using cached https://files.pythonhosted.org/packages/d2/fb/29de8d08967f0cce1bb10b39846d836b0f3bf6776ddc36aed7c73498ca7e/protobuf-3.8.0-cp36-cp36m-manylinux1_x86_64.whl
Collecting onnx==1.9.0 (from -r ./requirements_pc.txt (line 4))
  Using cached https://files.pythonhosted.org/packages/73/e9/5b953497c0e36df589fc60cc6c6b35a65eb67d9ad1e45a9163663e43426e/onnx-1.9.0.tar.gz
    Complete output from command python setup.py egg_info:
    fatal: not a git repository (or any of the parent directories): .git
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-build-9alp896m/onnx/setup.py", line 86, in <module>
        assert CMAKE, 'Could not find "cmake" executable!'
    AssertionError: Could not find "cmake" executable!

    ----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /tmp/pip-build-9alp896m/onnx/
Installing python osrt packages...
protobuf requires Python '>=3.7' but the running Python is 3.6.9
skipping gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu download: found /home/root/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu

I'm not being able of fixing that, and also it's weird because protobuf complains about the python version, but I read that with until 3.19.4 it should work with python 3.6. As you can see from the snippet, I tried to lower the version down to 3.8.0 with no success.

Same happens with Docker on Ubuntu 18:

root@cce65d232e79:/home/root# source ./setup.sh
X64 Architecture
Installing python packages...
Collecting git+https://github.com/kumardesappan/caffe2onnx (from -r ./requirements_pc.txt (line 12))
  Cloning https://github.com/kumardesappan/caffe2onnx to /tmp/pip-71_rn3kt-build
Collecting numpy (from -r ./requirements_pc.txt (line 1))
  Downloading https://files.pythonhosted.org/packages/45/b2/6c7545bb7a38754d63048c7696804a0d947328125d81bf12beaa692c3ae3/numpy-1.19.5-cp36-cp36m-manylinux1_x86_64.whl (13.4MB)
    100% |################################| 13.4MB 92kB/s
Collecting pyyaml (from -r ./requirements_pc.txt (line 2))
  Downloading https://files.pythonhosted.org/packages/b3/85/79b9e5b4e8d3c0ac657f4e8617713cca8408f6cdc65d2ee6554217cedff1/PyYAML-6.0-cp36-cp36m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl (603kB)
    100% |################################| 604kB 1.8MB/s
Collecting protobuf==3.19.4 (from -r ./requirements_pc.txt (line 3))
  Downloading https://files.pythonhosted.org/packages/c6/1c/f18d97fc479b4fb6f72bbb0e41188575362e3bbd31014cf294ef0fdec8bf/protobuf-3.19.4-py2.py3-none-any.whl (162kB)
    100% |################################| 163kB 4.6MB/s
Collecting onnx==1.9.0 (from -r ./requirements_pc.txt (line 4))
  Downloading https://files.pythonhosted.org/packages/73/e9/5b953497c0e36df589fc60cc6c6b35a65eb67d9ad1e45a9163663e43426e/onnx-1.9.0.tar.gz (9.8MB)
    100% |################################| 9.9MB 125kB/s
    Complete output from command python setup.py egg_info:
    fatal: not a git repository (or any of the parent directories): .git
    Traceback (most recent call last):
      File "<string>", line 1, in <module>
      File "/tmp/pip-build-ovum_1rs/onnx/setup.py", line 86, in <module>
        assert CMAKE, 'Could not find "cmake" executable!'
    AssertionError: Could not find "cmake" executable!

    ----------------------------------------
Command "python setup.py egg_info" failed with error code 1 in /tmp/pip-build-ovum_1rs/onnx/
Installing python osrt packages...
protobuf requires Python '>=3.7' but the running Python is 3.6.9
skipping gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu download: found /home/root/gcc-arm-9.2-2019.12-x86_64-aarch64-none-linux-gnu
Installing:onnxruntime
ln: failed to create symbolic link 'libonnxruntime.so.1.7.0': File exists
Installing:tflite_2.8
Installing:opencv
Installing:dlr
root@cce65d232e79:/home/root#

(On Ubuntu native I had managed to fix the linker problem with libonnxruntime.so: https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1235227/sk-tda4vm-pc-setup-for-model-compilation-error-while-building).

Basically, I cannot even manage to complete the setup and try to run the example scripts. Let me know if you had experienced something similar.

As per posted logs,

"Illegal Instruction (Core Dump) " 

edeai-tidl-tools are built using libs which requires AVX/AVX2 support.

Could you please confirm your machine is AVX supported ?

You can use below cmd ,

grep avx /proc/cpuinfo

OR

grep avx2 /proc/cpuinfo

Regards,

Pratik

I would say AVX is supported, while avx2 no because nothing happens. Could this be the reason?

Okay, then that must be the reason, thanks.

Do you know how I can fix the problems I encountered while trying the setup on docker over Windows? (I checked the the host laptop supports AVX and AVX2, thus there should be no problems)