torch.nn.functional.{instance, batch}_norm trigger INTERNAL ASSERT FAILED when input is empty tensor with complex32 #71635
Description
🐛 Describe the bug
torch.nn.functional.{instance, batch}_norm trigger INTERNAL ASSERT FAILED when input is empty tensor with complex32
import torch
input = torch.rand([1, 0, 2], dtype=torch.complex32)
running_mean = torch.rand([])
running_var = torch.rand([])
torch.nn.functional.instance_norm(input)
# RuntimeError: falseINTERNAL ASSERT FAILED at "../aten/src/ATen/AccumulateType.cpp":16, please report a bug to PyTorch. Unrecognized ScalarType: ComplexHalf
torch.nn.functional.batch_norm(input, running_mean, running_var)
# RuntimeError: falseINTERNAL ASSERT FAILED at "../aten/src/ATen/AccumulateType.cpp":16, please report a bug to PyTorch. Unrecognized ScalarType: ComplexHalfVersions
pytorch: 1.10.1
cc @ezyang @anjali411 @dylanbespalko @mruberry @lezcano @nikitaved