bytecodealliance / bytecodealliance/wasmtime
Improve regalloc register priority order under Fastcall ABI
Open
cranelift:area:x64
- Dominant language
- Rust
- Stars
- 18.6k
- Forks
- 1.8k
- Avg merge
- 1d 18h
- Merged PRs (30d)
- 126
Description
On x86-64 with the new backend, the fastcall ABI implementation (#2678) currently results in suboptimal code in many cases because the register allocator's priority order, which is tuned to use SysV caller-saves (volatiles) first, allocates registers that are actually callee-saves and forces unnecessary clobber-saves. We should tune the ordering based on ABI in order to improve the generated code.
Contributor guide
Assessment
This issue has not been assessed yet.