Landing: e891ad2e77a5

← All landings

Project / Subsystem

gcc / tree-optimization

Date

2026-08-18

Author

Richard Biener

Commit

e891ad2e77a5373d62af39c8b5f815993fccf253

Source

github

Perf win

No

Breaking

No

View on GitHub →

All attributes

project: gcc
subsystem: tree-optimization
patch_id: —
commit_hash: e891ad2e77a5373d62af39c8b5f815993fccf253
source_type: github
headline: Assign vertex weight to all SLP nodes, including those without representatives
tldr: SLP optimization now assigns vertex weights to all nodes, ensuring consistent cost analysis even for nodes without a representative.
author: Richard Biener
outcome: committed
performance_win: false
breaking_change: false
series_id: —
series_parts: []
tags: • tree-optimization
• vectorization
• cost-model
discussion_id_link: —
bugzilla_pr: —
url: https://github.com/gcc-mirror/gcc/commit/e891ad2e77a5373d62af39c8b5f815993fccf253
date: 2026-08-18T00:00:00.000Z

This change ensures that all nodes within GCC’s SLP (Straight-Line-Progression) vectorization process are assigned a vertex weight, which is used for cost-based analysis. Previously, nodes lacking an SLP_TREE_REPRESENTATIVE might not have had their weights initialized, potentially leading to incorrect cost estimations. The vect_slp_node_weight function is updated to use the region entry block for context when a representative is absent, and vect_optimize_slp_pass::start_choosing_layouts now unconditionally assigns a vertex weight to all nodes.