Landing: c8a9bc5cf5d7
Project / Subsystem
gcc / i386
Date
2026-08-28
Author
Hongyu Wang
Commit
c8a9bc5cf5d788a3bc796341f8070d073076b91b
Source
github
Perf win
Yes
Breaking
No
All attributes
- project
- gcc
- subsystem
- i386
- patch_id
- —
- commit_hash
- c8a9bc5cf5d788a3bc796341f8070d073076b91b
- source_type
- github
- headline
- Allow round for HFmode in i386 optab hook to aid vectorization.
- tldr
- Enables the 'round' operation for 16-bit float (HFmode) in the i386 backend's optab hook, preventing unnecessary conversions during vectorization.
- author
- Hongyu Wang
- outcome
- committed
- performance_win
- true
- breaking_change
- false
- series_id
- —
- series_parts
- []
- tags
-
- • i386
- • vectorization
- • optimization
- • float16
- discussion_id_link
- —
- bugzilla_pr
- —
- date
- 2026-08-28T00:00:00.000Z
This change updates the i386 backend’s ix86_optab_supported_p function to recognize the ‘round’ operation for HFmode (_Float16). Previously, this was not explicitly supported, leading to the compiler inserting unnecessary extend/truncate operations when vectorizing code that used _Float16 rounding. By allowing direct support, the compiler can avoid these extra operations, improving vectorization efficiency.