[llvm] [IR][RISCV] Remove @llvm.experimental.vp.splat (PR #171084)

Min-Yih Hsu via llvm-commits llvm-commits at lists.llvm.org
Mon Dec 8 11:50:05 PST 2025


mshockwave wrote:

> > > Do we need to add AutoUpgrade support for compatibility with old bitcode?
> > 
> > 
> > If we are to add this AutoUpgrade, what would we substitute the scalable vector version of this intrinsic with?
> > I can only think of store + zero-stride load, or multiple layers of vector.interleave
> 
> The VL doesn't matter so we should be able to use the splat idiom, insertelement+shufflevector.

I somehow had a wrong impression that shufflevector couldn't be used with scalable vectors.

https://github.com/llvm/llvm-project/pull/171084


More information about the llvm-commits mailing list