Landing: 47f2a222ceda

← All landings

Project / Subsystem

gcc / risc-v

Date

2026-04-15

Author

Bohan Lei

Commit

47f2a222ceda5f9834e0840a14cc6155f7d15190

Source

github

Perf win

Yes

Breaking

View on GitHub →

All attributes

project: gcc
subsystem: risc-v
patch_id: —
commit_hash: 47f2a222ceda5f9834e0840a14cc6155f7d15190
source_type: github
headline: RISC-V: Use vnsrl instruction for even-odd shuffles.
tldr: GCC now uses vnsrl instruction in RISC-V for even-odd shuffles, potentially improving performance versus vcompress.
author: Bohan Lei
outcome: committed
performance_win: true
breaking_change: false
series_id: —
series_parts: []
tags: • risc-v
• optimization
• vectorization
• performance
discussion_id_link: —
bugzilla_pr: —
url: https://github.com/gcc-mirror/gcc/commit/47f2a222ceda5f9834e0840a14cc6155f7d15190
date: 2026-04-15T00:00:00.000Z

This change replaces the vcompress instruction with vnsrl for implementing even-odd shuffles on RISC-V. The vcompress instruction is slower, and requires a mask load, so using vnsrl should improve performance on many implementations. This optimization aligns with LLVM’s current behavior.