[ARROW-16850] [C++] Improve csv writer by copying data and delimiter separately

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 9.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/32176
• Assignee: Yibo Cai
• Reporter: Yibo Cai
• Created: 17/Jun/22 09:07
• Updated: 11/Jan/23 11:46
• Resolved: 29/Jun/22 09:11

Description

Profiling csv writer with Arm Statistical Profiling Extension (SPE) reveals there is relatively high branch misprediction rate from memcpy, due to interleaving copying data field and delimiter which is very short(1 or 2 chars). Copying delimiter with specific code, memcpy only the data field improves performance significantly.

Issue Links

• links to: GitHub Pull Request #13394