pytorch

157310c8 - [inductor][triton] if device is a torch.device, then make cuda_properties index it correctly (#87174)

[inductor][triton] if device is a torch.device, then make cuda_properties index it correctly (#87174) Without this, I was running into obvious `KeyError`s that were assuming that the device was an integer when running `examples/imagenet`. ```python (pytorch) soumith@bluebox:~/code/examples/imagenet$ python main.py --gpu 0 /home/soumith/dataset/imagenet /home/soumith/code/vision/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: warn(f"Failed to load image Python extension: {e}") /home/soumith/code/examples/imagenet/main.py:100: UserWarning: You have chosen a specific GPU. This will completely disable data parallelism. warnings.warn('You have chosen a specific GPU. This will completely ' Use GPU: 0 for training => creating model 'resnet18' make_fallback(aten.unfold): a decomposition exists, we should switch to it make_fallback(aten.unfold_backward): a decomposition exists, we should switch to it Traceback (most recent call last): File "/home/soumith/code/pytorch/torch/_inductor/graph.py", line 254, in call_function return lowerings[target](*args, **kwargs) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 202, in wrapped return decomp_fn(*args, **kwargs) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 2994, in var_ diffs = square(sub(x, mean(x, axis, keepdim=True))) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 202, in wrapped return decomp_fn(*args, **kwargs) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 2983, in mean sum_result = sum_(x, axis, keepdim) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 202, in wrapped return decomp_fn(*args, **kwargs) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 3211, in sum_ return fn(x, axis, keepdims, dtype=dtype) File "/home/soumith/code/pytorch/torch/_inductor/lowering.py", line 2953, in inner result = Reduction.create( File "/home/soumith/code/pytorch/torch/_inductor/ir.py", line 714, in create hint, split = cls.num_splits( File "/home/soumith/code/pytorch/torch/_inductor/ir.py", line 454, in num_splits num_sm = get_device_properties(device).multi_processor_count File "/home/soumith/code/pytorch/torch/_inductor/cuda_properties.py", line 43, in get_device_properties return _properties()[_device(device)] KeyError: device(type='cuda', index=0) ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/87174 Approved by: https://github.com/yf225