On Fri, 23 Oct 2026 15:46:22 GMT, Jorn Vernee <[email protected]> wrote:
> I've implemented the feedback so far. PPC and s390 will now only widen > unsigned values to `long`. > > I'm working out some final details about the design with Maurizio, then I > will un-draft the PR. Thank you! I appreciate it. This way, we will have more time to think about the remaining topics and can handle them separately. jtreg:test/jdk/java/foreign tests have passed on linux ppc64le, linux s390 and AIX. I can rerun tests on your final version. > That's a good point. We could move some of this logic to the Windows > CallArranger as well, and convert to `long` first using > `Double::doubleToRawLongBits`. I prototyped this and it seems to just work. > The benefit of doing it in the backend is that we can handle the conversion > and move into target register both at once with a single move operation. Thanks for trying! I had similar thoughts. Unfortunately, we need all cases for PPC64: I2F, F2I, L2D, D2L May I ask what the goal of the intrinsification is? Is it to replace the DowncallLinker stubs completely including all complicated cases and features like capture state? Or is the goal more to improve performance? If so, we could focus on the most performance-critical cases and only intrinsify calls with typical signatures. Btw. your update looks good. Thanks for improving it! ------------- PR Comment: https://git.openjdk.org/jdk/pull/33417#issuecomment-6373927380 PR Comment: https://git.openjdk.org/jdk/pull/33417#issuecomment-6472647615
