Need a more descriptive error when running ROCm tests on a non-ROCm machine

[ahmadsharif1]

Hi,

I was trying to reproduce this error:

https://github.com/pytorch/pytorch/actions/runs/14086862973/job/39455645090

PYTORCH_OPINFO_SAMPLE_INPUT_INDEX=1 PYTORCH_TEST_WITH_ROCM=1 python test/test_ops.py TestCommonCUDA.test_noncontiguous_samples_native_layer_norm_cuda_float32
Traceback (most recent call last):
  File "/home/ahmads/personal/pytorch/test/test_ops.py", line 2825, in <module>
    instantiate_device_type_tests(TestCommon, globals())
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_device_type.py", line 928, in instantiate_device_type_tests
    device_type_test_class.instantiate_test(
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_device_type.py", line 536, in instantiate_test
    instantiate_test_helper(
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_device_type.py", line 443, in instantiate_test_helper
    test = decorator(test)
           ^^^^^^^^^^^^^^^
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_device_type.py", line 1784, in skipCUDAIfNoCusolver
    not has_cusolver() and not has_hipsolver(), "cuSOLVER not available"
                               ^^^^^^^^^^^^^^^
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_device_type.py", line 1776, in has_hipsolver
    rocm_version = _get_torch_rocm_version()
                   ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_cuda.py", line 267, in _get_torch_rocm_version
    return tuple(int(x) for x in rocm_version.split("."))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ahmads/personal/pytorch/torch/testing/_internal/common_cuda.py", line 267, in <genexpr>
    return tuple(int(x) for x in rocm_version.split("."))
                 ^^^^^^
ValueError: invalid literal for int() with base 10: 'None'

I think this error should be more descriptive and say something like "you need to run this on a ROCm machine" instead of this error.

cc @jeffdaily @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang @naromero77amd @malfet

[naromero77amd]

Just to be clear, you were running this test on a non-ROCm machine. You are referring to the very last error, the "ValueError" remark ?

[ahmadsharif1]

Just to be clear, you were running this test on a non-ROCm machine. You are referring to the very last error, the "ValueError" remark ?

Correct. Instead of complaining about base 10, it should have said you need a ROCm machine here.

[naromero77amd]

With the linked PR, it is now saying:

No ROCm runtime is found, using ROCM_HOME='/opt/rocm'

and improvement.