Open fcharras opened 1 year ago
In fact there's the same isue with device="mps"
or device="cuda"
so it seems to be a pytorch bug and not an ipex bug.
+@gujinghui, @tye1
@fcharras Did you confirm on PyTorch master? Or, you will submit a issue to PyTorch?
I confirmed with latest stable distribution. I submitted the issue at https://github.com/pytorch/pytorch/issues/108408 . Also worth mentioning that a similar issue was recently reported and fixed on master, see https://github.com/pytorch/pytorch/issues/106773 , it also affected xpu and other backends.
The issue has been fixed upstream, will be available in further relases or builds from source on latest commits.
Shall we close here, then?
Describe the bug
Reproducer:
prints:
showing that
array
andarray_2
share the same memory (and indeed mutating one will mutate the other), despite passingcopy=True
.Versions