-
Notifications
You must be signed in to change notification settings - Fork 21.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
xpu: support torch.utils.data.DataLoader(pin_memory_device='xpu') #126491
Comments
Issue at torch-xpu-ops (filed as noted in the log): intel/torch-xpu-ops#261 |
cc @albanD just in case this is related to unified API for accelerator backends |
The unified API for pin memory is already in PyTorch master, right? |
According a printed call stack - yes. Once pin_memory op will be implemented in torch_xpu_ops, then someone will need to recheck whether pin_memory API works for XPU (I will do that once fix or PR will be available).
Pytorch has better visibility than torch_xpu_ops. To me it does make sense to have issue here in pytorch rather than in torch_xpu_ops. Besides this issue affect xpu backend at pytorch API level - another reason to have it as pytorch issue. This way it can be tracked towards xpu backend readiness for specific pytorch target. |
If need more attention, pls. add @EikanWang , @guangyey , @fengyuan14 and me as assignee or reviewer for all xpu related PR and issues. It's more helpful to solve the issue in short-term if urgent. Thanks. |
Good to know. However, I don't have the rights to add add assignees or reviewers. But I will be adding to CC in comments. As of now I do not request special attention to this issue from my side and can wait for the fix. Mostly it's to give heads up to other people who might be trying XPU backend that this feature is missing. |
It's already in our plan, actually. We will take attention on it. Thanks. |
torch.utils.data.DataLoader(pin_memory_device='xpu')
is currently not supported with upstream PyTorch XPU backend. I know that with IPEX this feature was supported. Please, support the feature if it's relevant for XPU or update PyTorch documentation with optimization notice that feature is not needed for XPU.Relevant environment:
This can be reproduced with:
Gives this output:
CC: @jgong5 @mingfeima @XiaobingSuper @ashokei @jingxu10 @gujinghui @EikanWang @fengyuan14 @guangyey
cc @andrewkho @gokulavasan @ssnl @VitalyFedyunin @dzhulgakov @gujinghui @EikanWang @fengyuan14 @guangyey
The text was updated successfully, but these errors were encountered: