On Fri, 31 Jul 2026 05:36:48 GMT, Jatin Bhateja <[email protected]> wrote:
> Hi All, > > This is a follow-up to JDK-8370691 (Float16Vector). As noted by > @xuemingshen-oracle and @PaulSandoz in this review comment on PR #28002 > (https://github.com/openjdk/jdk/pull/28002#issuecomment-4568001440), the > floating-point VECTOR_OP_OR fallback is a dead and semantically incorrect > code path: it computes the bitwise OR on the widened float32 encoding (e.g. > for Float16, Float16 → float32 → OR → Float16). > > The public OR operator is VO_NOFP and is rejected on FP vectors, so this path > was only reachable via the private OR_UNCHECKED op (which lacked VO_NOFP) and > had no call sites. This change removes the dead code entirely. As per > https://github.com/openjdk/jdk/pull/28002#issuecomment-4568642676, if an FP > OR is needed later, it can be re-added with proper intrinsic support. > > Kindly review and share your feedback. > > Best Regards, > Jatin > > --------- > - [x] I confirm that I make this contribution in accordance with the [OpenJDK > Interim AI Policy](https://openjdk.org/legal/ai). The changes ( (1) OR remains guarded by VO_NOFP, and the floating-point vector opCode validation rejects it before the fallback lookup, (2) and OR_UNCHECKED appears not to have been used since JDK-8291118) look good to me. Just wonder whether it would be worth adding a small test covering OR rejection for Float16Vector, FloatVector, and DoubleVector, though I don’t feel strongly about it. Saw the 'pre-submit tests" failed. I’m running Mach5 tiers 1–3 manually now: MDash: https://mach5.us.oracle.com/mdash/jobs/xueming.shen-jdk8385616-20260804-0100-48580758 ------------- PR Comment: https://git.openjdk.org/jdk/pull/32119#issuecomment-5173544862
