Is it possible to get GPU acceleration support on the M1 chip?

[HuangStomach]

🚀 The feature, motivation and pitch

You can see from pytorch's blog that the latest version of pytorch already supports GPU acceleration on M1 chips, which can be easily achieved by directly using to('mps'), may I ask if pyg also has plans to do similar support?

Alternatives

No response

Additional context

No response

[rusty1s]

I assume that you can get GPU acceleration on all/most of PyTorch functionality, which should be directly visible when running operators that do not make use of specialized functionality in torch-scatter or torch-sparse, e.g., GCNConv or SAGEConv. For our own external operators, I am afraid that M1 support seems to be of lower priority. We would need to integrate this with the help of Apply or the community.

[leo-oneill]

Not much of a coder, but it be awesome to have pyg support of GPU acceleration!