[FFM performance] Intermediate buffer allocation when returning structs

[Jorn Vernee]

I've seen several similar crashes before, e.g.: 
https://bugs.openjdk.org/browse/JDK-8316756

So, there might be some handling in the EA code missing for the pattern 
of nodes that the intrinsic creates.

Jorn

On 17-1-2025 15:25, Matthias Ernst wrote:
> Disabling the intrinsics appears to both fix the scalar replacement 
> and the crasher:
>
> java -XX:DisableIntrinsic=_Continuation_pin --add-opens 
> java.base/jdk.internal.misc=ALL-UNNAMED --add-opens 
> java.base/jdk.internal.vm=ALL-UNNAMED -Xlog:gc Repro
> [0.078s][info][gc] Using G1
> ################# Expect allocations #################
> ################# Expect silence(release) or crash(fastdebug) 
> #################
> ################# Done. #################
>
>
> On Fri, Jan 17, 2025 at 3:10 PM Alan Bateman <alan.bateman at oracle.com> 
> wrote:
>
>     On 17/01/2025 13:48, Maurizio Cimadamore wrote:
>     >
>     > Thanks Matthias, adding loom-dev
>     >
>     The intrinsics for pin/unpin are new in JDK 24. Can you do a quick
>     test
>     with -XX:DisableIntrinsic=_Continuation_pin to see if it reproduces?
>
>     -Alan
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/panama-dev/attachments/20250117/682e89e3/attachment-0001.htm>