Landing: 61e2300bc936

← All landings

Project / Subsystem

gcc / gcc/range-op

Date

2026-05-30

Author

Andrew Pinski

Commit

61e2300bc9364a9060aca12ba189d5bbca950aff

Source

github

Perf win

Yes

Breaking

View on GitHub →

All attributes

project: gcc
subsystem: gcc/range-op
patch_id: —
commit_hash: 61e2300bc9364a9060aca12ba189d5bbca950aff
source_type: github
headline: Range-op: Teach multiplication about non-negative results.
tldr: The range analysis pass now understands that a*a is non-negative, which helps optimize RISC-V VSETVL.
author: Andrew Pinski
outcome: committed
performance_win: true
breaking_change: false
series_id: —
series_parts: []
tags: • optimization
• range analysis
• risc-v
• aarch64
discussion_id_link: —
bugzilla_pr: —
url: https://github.com/gcc-mirror/gcc/commit/61e2300bc9364a9060aca12ba189d5bbca950aff
date: 2026-05-30T00:00:00.000Z

The value range propagation (VRP) pass now recognizes that the square of a variable is non-negative. This helps the compiler deduce tighter bounds for expressions involving multiplication. AArch64 and RISC-V VSETVL instructions benefit from this optimization. Two new testcases, vrp-mult-nonneg-1.c and vrp-mult-nonneg-2.c, validate the fix.