On Thu, 3 Sep 2026 13:22:10 GMT, Aleksey Shipilev <[email protected]> wrote:
> > Question for @shipilev : is Renaissance dotty an outlier, or is it typical? > > I have not checked a lot of workloads, just picked a few tests that I knew > were JIT-sensitive. IIRC, on quite a few workloads there were regressions, > and I have not seen gains. After the issue showed up prominently on Dotty, I > stopped looking further and reported this hiccup, because it looks like > something that would taint perf comparisons. I can take a broader look, if > you want, but what happens on Dotty is enough of a red flag for me. No problem, I'm on it. Part of the fun here is that uncommon traps are very time sensitive. Compiling `HashMap` slightly too soon means that the first use of `LinkedHashMap` causes many `HashMap` methods to be deoptimized because all of a sudden all of their calls become bimorphic. And, of course, the more cores you have, the more likely this is to happen. Etc, etc. ------------- PR Comment: https://git.openjdk.org/jdk/pull/28541#issuecomment-5550921576
