[ARROW-13170] [C++] Reducing branching in compute/kernels/vector_selection.cc

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: C++
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: Niranda Perera
• Created: 24/Jun/21 19:00
• Updated: 12/Jul/22 14:04

Description

wesm pointed out that the selection operations can be improved by using a non-branching method in ML and yibocai confirmed this. 

https://lists.apache.org/thread.html/rcffa661ca3526863fc5148ed3c111a72f03b2ce2626178bd83570aa6%40%3Cdev.arrow.apache.org%3E

 

Evaluate the following

1. Using branch-less approach
2. Check if `BitmapWordReader` can achieve better performance

Issue Links

• links to: GitHub Pull Request #10679