AW: Remove stream usage from Symtab.lookupPackage()

[christoph.dreis at freenet.de]

Hi Vicente,

thanks for taking the time. I’ve added a comment in the PR with screenshots from async-profiler runs etc. that are somewhat easier to post there compared to the mailing list. But to share the TL;DR:


  *   ~2500 classes
  *   Average size: ~76 lines
  *   Largest class: ~2500 lines
  *   Mostly hand-written, but some auto-generated protobuf files (I'd say 90/10 split)

Overall timings didn’t improve (or regress), but the allocations of Symtab.lookupPackage() went down from 1.6% to 0.1% overall, while CPU frames went down from 1.6% to 0.9%.

Let me know if you need more information and I’ll post this to the issue on GitHub.

Cheers,
Christoph


Am 22.02.23, 04:16 schrieb "Vicente Romero" <vicente.romero at oracle.com>:

Hi Christoph,

Could you please share more info like what type of sources are you compiling: machine generated or not, very long sources, etc. Also what's the performance improvement you get with your patch, etc.

Thanks,
Vicente
On 2/21/23 16:03, christoph.dreis at freenet.de<mailto:christoph.dreis at freenet.de> wrote:

Hi,

I'm currently profiling some compilation phases of internal projects and noticed in allocation profiles that Symtab.lookupPackage takes up ~2% overall. The majority of this is spent in .stream().anyMatch() usages to find out if the given module symbol depends on the unnamed module.



The PR under https://github.com/openjdk/jdk/pull/12700 desugars the code into a simple loop.

If you think this is worthwhile I'd appreciate if this is sponsored. I'd also need a ticket number for that because I can't create tickets. CLA should be signed though.

Let me know what you think.
Cheers,
Christoph





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/compiler-dev/attachments/20230222/efe14ef8/attachment-0001.htm>