[llvm] [RISCV] Add TuneNoDefaultUnroll to generic CPUs (PR #135318)

Mikhail R. Gadelha via llvm-commits llvm-commits at lists.llvm.org
Mon Jun 23 06:30:40 PDT 2025


mikhailramalho wrote:

> > > > https://lnt.lukelau.me/db_default/v4/nts/665?compare_to=656
> > > > Some big regression in code size with rva22u64_v -O3 -flto. Only regression in exec time was 523.xalancbmk_r but we know that benchmark is usually noisy
> > > 
> > > 
> > > The code size regressions are expected, the performance has no observable change possibly because Spacemit X60 is just a dual-issues in-order core. I don't know if there are some data from other CPUs?
> > 
> > 
> > I can run this PR on a VisionFive V2 that I have available if you'd like, but it doesn't have RVV.
> 
> That'll be appreciated!

Hi, I ran the tests this weekend on the visionfive2. This is SPEC built with `-march=rv64gc -fuse-ld=lld -O3 -flto`: https://lnt.lukelau.me/db_default/v4/nts/684?compare_to=686. Unfortunately, there is no perf data available.



https://github.com/llvm/llvm-project/pull/135318


More information about the llvm-commits mailing list