-
Notifications
You must be signed in to change notification settings - Fork 902
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
[BUG] Shared-memory from_arrow no longer zero-copy #8639
Comments
@beckernick no, the change happened when we switched Python to using |
This issue has been labeled |
Still needs to be addressed, AFAIK. |
Is it not possible for C++ to also zero copy similarly where applicable? The reason we moved to the C++ impl was to avoid duplicates logic between C++ and Python. |
@shwina yeah it would be possible, but the current implementation returns a |
cc @jrhemstad for advice on the API here, but maybe we could return something like a |
I think the new solutions to this issue will come from work on #14926. |
With the closing of #14926 this is resolved. We now use the C Data Interface for all Arrow interop, so no extraneous copies will be made. |
cc @trxcllnt @galipremsagar
Looking into implementing zero-copy
from_arrow
for node-rapids for Arrow arrays stored in IPC format in shared memory, we noticed that Python cudffrom_arrow
delegates tolibcudf.interop.from_arrow
which will always result in a copy. Evidently previously it was possible to avoid a copy when and arrow table had been created viaGpuArrowReader
from GPU memory.The text was updated successfully, but these errors were encountered: