Enable CEC for strings_udf - #11884
Conversation
Codecov ReportBase: 87.47% // Head: 88.09% // Increases project coverage by
Additional details and impacted files@@ Coverage Diff @@
## branch-22.12 #11884 +/- ##
================================================
+ Coverage 87.47% 88.09% +0.61%
================================================
Files 133 135 +2
Lines 21826 21984 +158
================================================
+ Hits 19093 19367 +274
+ Misses 2733 2617 -116
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
Co-authored-by: Vyas Ramasubramani <vyas.ramasubramani@gmail.com>
gmarkall
left a comment
There was a problem hiding this comment.
This looks good - just a couple of minor points on the diff.
gmarkall
left a comment
There was a problem hiding this comment.
This looks good - many thanks for the updates.
wence-
left a comment
There was a problem hiding this comment.
Sorry, I should have been clearer first time round.
|
rerun tests |
|
@gmarkall @wence- now that the relevant let me know if you have any further thoughts :) |
|
rerun tests |
I think some of the comments from my previous review (around unconditional initialisation of cuda contexts on import, and undefined |
made a few updates to this in 777e257 |
|
|
||
| def _get_ptx_file(): | ||
| if "RAPIDS_NO_INITIALIZE" in os.environ: | ||
| cc = int(os.environ.get("STRINGS_UDF_CC", "52")) |
There was a problem hiding this comment.
The idea here is that downstream consumers (like dask-cuda) that set RAPIDS_NO_INITIALIZE have a way to advertise the compute capability of the device they are running on (without creating a context). Do we care about bikeshedding this name?
|
|
|
@gpucibot merge |
This PR removes the runtime checks for CEC in
strings_udf.