[Issue]: CUDA_HOME environment variable is needed

[JinghanHuang]

Problem Description

When running the following command:
python3 setup.py --variant rocm build develop
I encounter the following error:
OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

My understanding is that this repo should not require CUDA. Do I still need to install CUDA to compile this repository? If not, is there a recommended way to resolve this error? Thanks!

Operating System

CentOS Stream 9

CPU

AMD EPYC 9654 96-Core Processor

GPU

8 x AMD Instinct MI300X

ROCm Version

ROCm 6.4.2

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[liligwu]

Hi @JinghanHuang , your understanding is correct; it requires ROCm 6.3, not CUDA.
Could you please provide the entire building log? Unfortunately, the information you provided is insufficient for us to identify the source of the problem. Thanks.

[JinghanHuang]

Thanks for your reply! Here is output for python3 setup.py --variant rocm build develop

Building for variant: rocm
ROCm directory: /opt/rocm
rocSHMEM directory: /home/username/rocshmem
Detected OpenMPI directory: /home/username/ompi_install
Traceback (most recent call last):
  File "/home/username/DeepEP/setup.py", line 181, in <module>
    CUDAExtension(
  File "/home/username/.local/lib/python3.12/site-packages/torch/utils/cpp_extension.py", line 1337, in CUDAExtension
    library_dirs += library_paths(device_type="cuda")
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/username/.local/lib/python3.12/site-packages/torch/utils/cpp_extension.py", line 1548, in library_paths
    if (not os.path.exists(_join_cuda_home(lib_dir)) and
                           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/username/.local/lib/python3.12/site-packages/torch/utils/cpp_extension.py", line 2982, in _join_cuda_home
    raise OSError('CUDA_HOME environment variable is not set. '
OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.

[liligwu]

Hi @JinghanHuang , I'm sorry for the late reply.
The issue actually happens in Pytorch. Which Pytorch version were you using? Could you please use the nightly Pytorch build if possible? Thanks.
pip3 install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/rocm6.3