Landing: 3565f0b78096
Project / Subsystem
gcc / gcc/vect
Date
2026-08-04
Author
Tamar Christina
Commit
3565f0b780965a097fdeea07b94fcfe768998a65
Source
github
Perf win
Yes
Breaking
No
All attributes
- project
- gcc
- subsystem
- gcc/vect
- patch_id
- —
- commit_hash
- 3565f0b780965a097fdeea07b94fcfe768998a65
- source_type
- github
- headline
- Expand complex multiplication patterns to cover all operand orderings
- tldr
- Enable vectorizer to consider all 8 multiplication orderings for complex patterns to match more code.
- author
- Tamar Christina
- outcome
- committed
- performance_win
- true
- breaking_change
- false
- series_id
- —
- series_parts
- []
- tags
-
- • gcc/vect
- • vectorization
- • complex numbers
- • optimization
- discussion_id_link
- —
- bugzilla_pr
- —
- date
- 2026-08-04T00:00:00.000Z
This commit enhances the vectorizer’s ability to match complex multiplication patterns by enabling it to try all eight possible orderings of operands and multiplication variants. Previously, only a limited set of operand permutations were considered. This broader matching, combined with existing checks like compatible_complex_nodes_p and data flow analysis, ensures that more valid complex multiplication code can be vectorized without accepting incorrect combinations. This change is a prerequisite for correctly applying the signed zero handling fix on AArch64.