[Feature Request / Question] Modifications Required to run on ARM / M1

[The-Real-Thisas]

Since I didn't see any mention of M1 or ARM and the current build fails. I was wondering what changes may be needed to support the architecture.

Short Version:

 ~ pip install --upgrade git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0

Building wheels for collected packages: torchsparse
  Building wheel for torchsparse (setup.py) ... error
  error: subprocess-exited-with-error

  × python setup.py bdist_wheel did not run successfully.
  │ exit code: 1
  ╰─> [58 lines of output]
      clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/thisas/miniforge3/include -arch arm64 -fPIC -O2 -isystem /Users/thisas/miniforge3/include -arch arm64 -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/torch/csrc/api/include -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/TH -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/THC -I/Users/thisas/miniforge3/include/python3.9 -c torchsparse/backend/convolution/convolution_cpu.cpp -o build/temp.macosx-11.0-arm64-cpython-39/torchsparse/backend/convolution/convolution_cpu.o -g -O3 -fopenmp -lgomp -DTORCH_API_INCLUDE_EXTENSION_H -DPYBIND11_COMPILER_TYPE=\"_clang\" -DPYBIND11_STDLIB=\"_libcpp\" -DPYBIND11_BUILD_ABI=\"_cxxabi1002\" -DTORCH_EXTENSION_NAME=backend -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++14
      clang: warning: -lgomp: 'linker' input unused [-Wunused-command-line-argument]
      clang: error: unsupported option '-fopenmp'
      error: command '/usr/bin/clang' failed with exit code 1
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
  ERROR: Failed building wheel for torchsparse
  Running setup.py clean for torchsparse
Failed to build torchsparse
Installing collected packages: torchsparse
  Running setup.py install for torchsparse ... error
  error: subprocess-exited-with-error

  × Running setup.py install for torchsparse did not run successfully.
  │ exit code: 1
  ╰─> [60 lines of output]
      running install
      clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/thisas/miniforge3/include -arch arm64 -fPIC -O2 -isystem /Users/thisas/miniforge3/include -arch arm64 -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/torch/csrc/api/include -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/TH -I/Users/thisas/miniforge3/lib/python3.9/site-packages/torch/include/THC -I/Users/thisas/miniforge3/include/python3.9 -c torchsparse/backend/convolution/convolution_cpu.cpp -o build/temp.macosx-11.0-arm64-cpython-39/torchsparse/backend/convolution/convolution_cpu.o -g -O3 -fopenmp -lgomp -DTORCH_API_INCLUDE_EXTENSION_H -DPYBIND11_COMPILER_TYPE=\"_clang\" -DPYBIND11_STDLIB=\"_libcpp\" -DPYBIND11_BUILD_ABI=\"_cxxabi1002\" -DTORCH_EXTENSION_NAME=backend -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++14
      clang: warning: -lgomp: 'linker' input unused [-Wunused-command-line-argument]
      clang: error: unsupported option '-fopenmp'
      error: command '/usr/bin/clang' failed with exit code 1
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> torchsparse

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

Full Log:
fail.log

[zhijian-liu]

Could you maybe try to use gcc instead of clang for the compilation?

[The-Real-Thisas]

I attempted.

I think maybe this is the wrong way to do it.

 ~ pip install --global-option build_ext --global-option --compiler=gcc --upgrade git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0

WARNING: Disabling all use of wheels due to the use of --build-option / --global-option / --install-option.
Collecting git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0
  Cloning https://github.com/mit-han-lab/torchsparse.git (to revision v1.4.0) to /private/var/folders/z8/sp2w4vyj3q71z22h0hz0vxs00000gn/T/pip-req-build-x1r2vkq5
  Running command git clone --filter=blob:none --quiet https://github.com/mit-han-lab/torchsparse.git /private/var/folders/z8/sp2w4vyj3q71z22h0hz0vxs00000gn/T/pip-req-build-x1r2vkq5
  Running command git checkout -q 74099d10a51c71c14318bce63d6421f698b24f24
  Resolved https://github.com/mit-han-lab/torchsparse.git to commit 74099d10a51c71c14318bce63d6421f698b24f24
  Preparing metadata (setup.py) ... done
Skipping wheel build for torchsparse, due to binaries being disabled for it.
Installing collected packages: torchsparse
  Running setup.py install for torchsparse ... error
  error: subprocess-exited-with-error

  × Running setup.py install for torchsparse did not run successfully.
  │ exit code: 1
  ╰─> [2 lines of output]
      running build_ext
      error: don't know how to compile C/C++ code on platform 'posix' with 'gcc' compiler
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: legacy-install-failure

× Encountered error while trying to install package.
╰─> torchsparse

note: This is an issue with the package mentioned above, not pip.
hint: See above for output from the failure.

[The-Real-Thisas]

Weirdly enough this conda package (https://anaconda.org/conda-forge/torchsparse) actually builds but when I execute an example script.

from torch import nn
from torchsparse import nn as spnn

model = nn.Sequential(
    spnn.Conv3d(in_channels, out_channels, kernel_size),
    spnn.BatchNorm(out_channels),
    spnn.ReLU(True),
)

It gives me an error.

(base)  ~ python3 temp.py
Traceback (most recent call last):
  File "/Users/thisas/temp.py", line 2, in <module>
    from torchsparse import nn as spnn
  File "/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/nn/__init__.py", line 1, in <module>
    from .modules import *
  File "/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/nn/modules/__init__.py", line 3, in <module>
    from .conv import *
  File "/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/nn/modules/conv.py", line 9, in <module>
    from torchsparse.nn import functional as F
  File "/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/nn/functional/__init__.py", line 2, in <module>
    from .conv import *
  File "/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/nn/functional/conv.py", line 7, in <module>
    import torchsparse.backend
ImportError: dlopen(/Users/thisas/miniforge3/lib/python3.9/site-packages/torchsparse/backend.cpython-39-darwin.so, 0x0002): symbol not found in flat namespace '__ZNK2at10TensorBase8data_ptrIlEEPT_v'

I think this is indicative of some sort of dependacy collision.

Related Stackoverflow

Python Version: Python 3.9.13
Pytorch: 1.11.0.post2

[zhijian-liu]

I think you could try setting CC or CXX instead to specify which compiler to use for building TorchSparse.

[The-Real-Thisas]

I'm sorry @zhijian-liu I'm not sure how I can do that. Could you maybe provide what flag I should set since I feel out of my depth here.

[zhijian-liu]

No worries! I'm not very certain either. Could you maybe try export CXX=<path-to-g++> before installation (see https://unix.stackexchange.com/questions/298810/how-to-change-the-default-cc-command-on-mac-os)?

[zhijian-liu]

@The-Real-Thisas, I'm just checking in to see if this issue has been resolved. Thanks!

[The-Real-Thisas]

Sorry @zhijian-liu I was busy for the past few weeks and was unable to focus on this.

I tested with setting CXX

(base)  ~ export CXX=/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin && pip install --upgrade git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0
Collecting git+https://github.com/mit-han-lab/torchsparse.git@v1.4.0
  Cloning https://github.com/mit-han-lab/torchsparse.git (to revision v1.4.0) to /private/var/folders/z8/sp2w4vyj3q71z22h0hz0vxs00000gn/T/pip-req-build-iomdfi44
  Running command git clone --filter=blob:none --quiet https://github.com/mit-han-lab/torchsparse.git /private/var/folders/z8/sp2w4vyj3q71z22h0hz0vxs00000gn/T/pip-req-build-iomdfi44
  Running command git checkout -q 74099d10a51c71c14318bce63d6421f698b24f24
  Resolved https://github.com/mit-han-lab/torchsparse.git to commit 74099d10a51c71c14318bce63d6421f698b24f24
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error

  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [6 lines of output]
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "/private/var/folders/z8/sp2w4vyj3q71z22h0hz0vxs00000gn/T/pip-req-build-iomdfi44/setup.py", line 7, in <module>
          from torch.utils.cpp_extension import (CUDA_HOME, BuildExtension, CppExtension,
      ModuleNotFoundError: No module named 'torch.utils.cpp_extension'
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

It's a different error so I think it is progress ?

I think it's a problem with pytorch since it gives a ModuleNotFoundError but I have torch installed, maybe it is a version issue ?

(base)  ~ pip show torch
Name: torch
Version: 1.12.0
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org/
Author: PyTorch Team
Author-email: packages@pytorch.org
License: BSD-3
Location: /Users/thisas/miniforge3/lib/python3.9/site-packages
Requires: typing-extensions
Required-by: timm, torchvision

I'm not sure where to go from here, maybe try building from the repo manually ?

[The-Real-Thisas]

@zhijian-liu, Sorry for the delay, just wondering what I can do from here ?

[zhijian-liu]

Sorry for the late reply. Could you double-check in your Python that you could import torch.utils.cpp_extension?

[The-Real-Thisas]

Nope it gives me a ModuleNotFoundError: No module named 'torch.utils.cpp_extension' however importing from torch import nn works.