Method of compile LibTorchSharp on ARM based Mac OS

[dayo05]

Do this on your risk!!

Before we go, please replace path/to/TorchSharp and path/to/pytorch to yours!

1. Build whole pytorch.
   A. Clone https://github.com/pytorch/pytorch.git
   B. install dependencies of compile. This is introduced on README.md on pytorch's github repo.
   C. Execute setup.py
2. Edit CMakeLists of LibTorchSharp

• Some feature is not exists on ARM. So, I just disable some features.
  A. Open /path/to/TorchSharp/src/Native/LibTorchSharp/CMakeLists.txt
  B. Remove crc32c.c and crc32c.h. Those files uses cpuinfo.h which not exists on ARM.

3. Build LibTorchSharp
   A. Generate MakeFile with this command: CMAKE_PREFIX_PATH=path/to/pytorch/torch/share/cmake Caffe2_DIR=path/to/pytorch/torch/share/cmake/Caffe2 cmake .
   B. Build with command make

If this method works fine, native library will exists on path/to/TorchSharp/src/Native/LibTorchSharp/libTorchSharp.dylib

I don't test this yet but maybe you can load the library via NativeLibrary.Load

[dayo05]

How to test:

1. Edit TorchSharpTest.csproj: You should change target platform from x64 to arm64.
2. Load Native Library on constructor. In my case, I did like this.

NativeLibrary.Load("/Users/dayo/TorchSharp/src/Native/pytorch/build/lib/libtorch.dylib");
NativeLibrary.Load("/Users/dayo/TorchSharp/src/Native/LibTorchSharp/libLibTorchSharp.dylib");        

I check that test passed.

[FancyFurret]

I have integrated an arm64 build into the build system using the arm64 torch binaries from anaconda: PR #903

If you're still interested in arm64 support, maybe you can help validate the PR? :)