[FEA] Optimize regexp_replace in multi-replace scenarios #7907
Assignees
Labels
performance
A performance related task/issue
Comments
NVnavkumar
added
feature request
3 tasks
rapids-bot bot
pushed a commit
to rapidsai/cudf
that referenced
this issue
Mar 23, 2023
Adds the JNI API for `stringReplace` using column vector arguments for `targets` and `repls` (to make this consistent with the C++ API). Also adds unit tests for the new API. Part of the work for NVIDIA/spark-rapids#7907. Authors: - Navin Kumar (https://github.com/NVnavkumar) Approvers: - Nghia Truong (https://github.com/ttnghia) URL: #12979
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Is your feature request related to a problem? Please describe.
When seeing regular expressions like
ab|cd|efin the replace API, the performance is not great for large strings. We can actually optimize this for running on the GPU using the cudf native multi-replace API. This API was recently updated for performance fixes for large strings in rapidsai/cudf#12858. So we should also measure how performant this optimization could be based on input string size and number of target strings.Describe the solution you'd like
The RAPIDS Accelerator for Apache Spark should utilize this optimization whenever it makes sense based on performance considerations.
Describe alternatives you've considered
regexp_replacedoes not perform well on the GPU when the input strings are very large. In addition, choices can add some complexity to the regular expression as there are more options. We can parallelize this on the GPU using cudf API described earlier.The text was updated successfully, but these errors were encountered: