Mac build is not working as monarch assumes the presence of cuda

[suo]

Trying to install monarch on my arm64 macbook, going through the readme:

conda create -n monarchenv python=3.10 -y
conda activate monarchenv

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
rustup toolchain install nightly
rustup default nightly

The next two instructions are not portable to mac, so I skipped them for now:

# Install non-python dependencies
conda install libunwind -y

# Install the correct cuda and cuda-toolkit versions for your machine
sudo dnf install cuda-toolkit-12-0 cuda-12-0 libnccl-devel clang-devel

Running:

pip install --no-build-isolation .

Yields the following output:

Processing /Users/suo/code/monarch
  Preparing metadata (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Preparing metadata (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [21 lines of output]
      Traceback (most recent call last):
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 389, in <module>
          main()
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 373, in main
          json_out["return_val"] = hook(**hook_input["kwargs"])
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 175, in prepare_metadata_for_build_wheel
          return hook(metadata_directory, config_settings)
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/setuptools/build_meta.py", line 374, in prepare_metadata_for_build_wheel
          self.run_setup()
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/setuptools/build_meta.py", line 512, in run_setup
          super().run_setup(setup_script=setup_script)
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/site-packages/setuptools/build_meta.py", line 317, in run_setup
          exec(code, locals())
        File "<string>", line 54, in <module>
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/_collections_abc.py", line 999, in update
          self[key] = other[key]
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/os.py", line 685, in __setitem__
          value = self.encodevalue(value)
        File "/Users/suo/miniconda3/envs/monarchenv/lib/python3.10/os.py", line 757, in encode
          raise TypeError("str expected, not %s" % type(value).__name__)
      TypeError: str expected, not NoneType
      [end of output]

[dcci]

The problem here is that setup.py does rely on CUDA being available:

This patch lets us go a little further

diff --git a/setup.py b/setup.py
index 567046e..012a4f4 100644
--- a/setup.py
+++ b/setup.py
@@ -55,7 +55,7 @@ os.environ.update(
     {
         "CXXFLAGS": f"-D_GLIBCXX_USE_CXX11_ABI={int(torch._C._GLIBCXX_USE_CXX11_ABI)}",
         "RUSTFLAGS": " ".join(["-Zthreads=16", ENABLE_MSG_LOGGING]),
-        "CUDA_HOME": CUDA_HOME,
+        #"CUDA_HOME": CUDA_HOME,
         "LIBTORCH_LIB": TORCH_LIB_PATH,
         "LIBTORCH_INCLUDE": ":".join(torch_include_paths()),
         "_GLIBCXX_USE_CXX11_ABI": str(int(torch._C._GLIBCXX_USE_CXX11_ABI)),

But after that we hit other errors

/Users/davidino/monarch/monarch/python/monarch/common/mock_cuda.cpp:20:10: fatal error: 'link.h' file not found                        
   20 | #include <link.h>           
      |          ^~~~~~~~                                                                                                                                   
1 error generated.                 

Let me re-title this task to capture.

[allenwang28]

This should work now with USE_TENSOR_ENGINE=0! Closing this out as I'm adding instructions in #648