Describe the bug
Getting ipc handle using df['src']._column.data_array_view.get_ipc_handle() leads to AttributeError
Steps/Code to reproduce bug
import cudf
df =cudf.DataFrame()
df['src']=[1,2,3]
df['src']._column.data_array_view.get_ipc_handle()
Traceback (most recent call last):
File "", line 1, in
File "/home/iroy/.conda/envs/snmg_pr/lib/python3.7/site-packages/numba/cuda/cudadrv/devicearray.py", line 320, in get_ipc_handle
ipch = devices.get_context().get_ipc_handle(self.gpu_data)
File "/home/iroy/.conda/envs/snmg_pr/lib/python3.7/site-packages/numba/cuda/cudadrv/driver.py", line 847, in get_ipc_handle
memory.owner.handle,
AttributeError: 'Buffer' object has no attribute 'handle'
@Iroy30 would suggest raising an issue against Numba as well since this affects them directly outside of just cuDF.
cc @jakirkham since we've chatted about this. Seems like we may just want to ducktype a handle attribute to Buffer with the ctypes type that numba expects.
@Iroy30 would suggest raising an issue against Numba as well since this affects them directly outside of just cuDF.
Meanwhile I'll try ducktyping from cugraph if that works. I believe it expects ctype c_ulong. Thanks for looking into this!
I believe they define an actual type in their drvapi module that you can use that grabs it from the system
not sure i understand what you mean
This is not a bug of cuDF but is a bug of Numba. Suggested raising an issue against Numba offline, so closing this.
Raised as issue ( https://github.com/numba/numba/issues/5429 ).