installation gcc errors

[kjhenner]

installation gcc errors

[rusty1s]

This is strange. Which PyTorch version are you using?

[kjhenner]

The pytorch version is 1.3.0a0+bb4f380

I noticed that those errors are all occurring in the environment of DATA_PTR, which is conditionally defined on the VERSION_GE_1_3 flag. If I comment out the logic for that in the setup.py so it doesn't pass that flag, it will install, but with the following errors:

/usr/lib/gcc/x86_64-linux-gnu/5/include/mwaitxintrin.h(36): error: identifier "__builtin_ia32_monitorx" is undefined

/usr/lib/gcc/x86_64-linux-gnu/5/include/mwaitxintrin.h(42): error: identifier "__builtin_ia32_mwaitx" is undefined

2 errors detected in the compilation of "/tmp/tmpxft_00001d27_00000000-7_graclus_kernel.cpp1.ii".

If I try to import any module, I get: ModuleNotFoundError: No module named 'torch_cluster.graclus_cpu'

If I comment out everything but sample_neighbors from the __init__.py, I can actually import that. It's all I need from pytorch-cluster, so I was hoping perhaps I could settle with that, but unfortunately I also get similar errors for dependencies in pytorch-sparse.

[kjhenner]

It looks like my version of PyTorch didn't have the data/data_ptr change. It seems like that should be in any version >= 1.3, so I'm not sure what was going on.

I'm stuck with cuda 9.1, though, and can't figure out if there's a way to get torch-scatter working with that, but I suppose that's a separate problem!

I'll leave the issue open for now in case you want to look into whether that PyTorch version check isn't sufficient to determine how to handle that DATA_PTR macro, but if not, feel free to close the issue.

[rusty1s]

You can check out the torch-scatter<2.0 release, modify the DATA_PTR macro and install it by running python setup.py install. This should work fine.

[kjhenner]

Thanks for this suggestion--I've been busy working on another project for the last few days, but I will give this a try as soon as I'm able!

[kjhenner]

Thanks--I was able to get this working torch-scatter<2.0 and a corresponding version of torch-sparse--0.4.3, I think.

[github-actions]

This issue had no activity for 6 months. It will be closed in 2 weeks unless there is some new activity. Is this issue already resolved?