RFR: 8351623: VectorAPI: Add SVE implementation of subword gather load operation [v6]
Xiaohong Gong
xgong at openjdk.org
Mon Sep 22 02:24:14 UTC 2025
On Thu, 18 Sep 2025 19:58:09 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:
> > I would want them to consider if the approach with the special vector nodes `VectorConcatenateAndNarrow` and `VectorMaskWiden` are really desirable. The complexity needs to go somewhere, but I'm not sure if it is better in the C2 IR or in the backend.
>
> > It would just be nice to build on "simple" building blocks and not have too many complex nodes, that have very special semantics (widen + split into two)
>
> Intuitively this seems like the right way to think about it, although I don't have a proposed solution, i am really just agreeing with the above sentiment - a compositional solution, if possible, with the right primitive building blocks will likely be superior.
Thanks for your input @PaulSandoz ! And I agree with making the IR simple enough. I'm now working on finding a better way for these two complex operations. Hope I can fix it soon. Thanks!
-------------
PR Comment: https://git.openjdk.org/jdk/pull/26236#issuecomment-3316545262
More information about the hotspot-compiler-dev
mailing list