Landing: b4de9d73b479

Project / Subsystem

gcc / tree-optimization

Date

2026-10-05

Author

Richard Biener

Commit

b4de9d73b47972bc89df288997c8f3a3862bc1b1

Source

github

Perf win

Yes

Breaking

No

All attributes

project
gcc
subsystem
tree-optimization
patch_id
—
commit_hash
b4de9d73b47972bc89df288997c8f3a3862bc1b1
source_type
github
headline
FRE optimization improved by fixing aliasing walk logic
tldr
Fixes a bug in alias analysis that caused incorrect handling of PHI nodes, enabling more effective loop-invariant code motion.
author
Richard Biener
outcome
committed
performance_win
true
breaking_change
false
series_id
—
series_parts
[]
tags
  • • optimization
  • • loop optimization
  • • alias analysis
discussion_id_link
—
bugzilla_pr
—
date
2026-10-05T00:00:00.000Z

This commit addresses an issue in GCC’s alias analysis that prevented loop-invariant code motion (LICM) from being applied effectively in certain loop scenarios. The problem occurred when translating aliased uses, causing the get_continuation_for_phi function to fail. By clearing the visited bitmap during translation and removing an unnecessary abort_on_visited flag, the analysis can now correctly traverse complex control flow, enabling more FRE optimizations.