Landing: a5da203dd959
Project / Subsystem
gcc / xtensa
Date
2026-08-29
Author
Takayuki 'January June' Suwa
Commit
a5da203dd95900ba8bab16432667989f3d018c76
Source
github
Perf win
Yes
Breaking
No
All attributes
- project
- gcc
- subsystem
- xtensa
- patch_id
- —
- commit_hash
- a5da203dd95900ba8bab16432667989f3d018c76
- source_type
- github
- headline
- Make use of CONST.S instruction
- tldr
- Enabled the use of the Xtensa `CONST.S` instruction for floating-point immediate assignments.
- author
- Takayuki 'January June' Suwa
- outcome
- committed
- performance_win
- true
- breaking_change
- false
- series_id
- —
- series_parts
- []
- tags
-
- • xtensa
- • optimization
- • floating-point
- discussion_id_link
- —
- bugzilla_pr
- —
- date
- 2026-08-29T00:00:00.000Z
This commit enables the Xtensa CONST.S instruction for loading specific single-precision floating-point constants (+0.f, 1.f, 2.f, 0.5f) into floating-point registers. It also allows the Reload/LRA pass to handle assignments of +0.f to general-purpose registers using the MOVI instruction. This optimization is available when floating-point instructions for division, reciprocal, or square root are configured, as CONST.S is then enabled in the Xtensa ISA.