[ARROW-10304] [C++][Compute] Optimize variance kernel for integers

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/26295
• Assignee: Yibo Cai
• Reporter: Yibo Cai
• Created: 14/Oct/20 02:31
• Updated: 11/Jan/23 08:12
• Resolved: 22/Oct/20 08:57

Description

Current variance kernel converts all data type to `double` before calculation. It's sub-optimal for integers. Integer arithmetic is much faster than floating points, e.g., summation is 4x faster [1].

A quick test for calculating int32 variance shows up to 3x performance gain. Another benefit is that integer arithmetic is accurate.

[1] https://quick-bench.com/q/_Sz-Peq1MNWYwZYrTtQDx3GI7lQ

Issue Links

• links to: GitHub Pull Request #8466