[BUG] cudaErrorIllegalAddress while computing ranking... #9703
Comments
rilango
added
Needs Triage
|
Thank you for reportng @rilango -- is there any chance you could share the contents of the |
shwina
added
Python
|
@shwina Please find the data at https://drive.google.com/file/d/1-hIshmdiYd70BhDnfgYk-BSF89JU3HiG/view?usp=sharing |
|
Thank you @rilango -- two questions:
Again, public data is much more preferable here, but I'd understand if the issue is not easily reproducible for a generic dataset. |
|
Sorry. Please try now. It is now shared with public. |
|
FYI, I am able to narrow down the repro to: import cudf
df = cudf.read_csv('ranking.csv', usecols=['1557'])
print(df.rank()) |
|
repro without csv. import cudf
df = cudf.DataFrame({'0': range(0, 1557)})
df.rank() #no error
df = cudf.DataFrame({'0': range(0, 1558)})
df.rank() #error |
|
memcheck yield error when size is 1538 (close to ~1024+512) It is share memory write out of bounds error in |
Fixes #9703 add __host__ for nvcc return type deduction to work. replaced `auto` (generic lambda) with size_type.
|
This issue will be fixed by PR #9778 |
|
Cause of this issue is that return type is automatically assumed |
Fixes #9703 replace device lambda with device functor with return type. (due to [14. extended-lambda-restrictions](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#extended-lambda-restrictions) ) ~add `__host__` to lambda for nvcc return type deduction to work properly.~ ~replaced `auto` (generic lambda) with `size_type`.~ fixes shared memory write error caused in #9703 Authors: - Karthikeyan (https://github.com/karthikeyann) Approvers: - Vyas Ramasubramani (https://github.com/vyasr) - David Wendt (https://github.com/davidwendt) - Jake Hemstad (https://github.com/jrhemstad) URL: #9778
|
Issue is caused by return type deduction of device lambda limitation. Point 14 in https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#extended-lambda-restrictions |
Describe the bug
While attempting to compute rank(cudf.rank), we face the following error
Steps/Code to reproduce bug
Data file is at https://drive.google.com/file/d/1-hIshmdiYd70BhDnfgYk-BSF89JU3HiG/view?usp=sharing
Expected behavior
Return the dataframe with ranks
Environment overview (please complete the following information)
Environment details
Click here to see environment details
The text was updated successfully, but these errors were encountered: