[lworld] RFR: 8329196: [lworld] More fixes to value object construction in C2 [v2]

Tobias Hartmann thartmann at openjdk.org
Thu Mar 28 16:14:55 UTC 2024


On Thu, 28 Mar 2024 14:50:56 GMT, Tobias Hartmann <thartmann at openjdk.org> wrote:

>> Follow-up from [JDK-8325660](https://bugs.openjdk.org/browse/JDK-8325660). We can't rely on `GraphKit::replace_in_map` to reliably replace an InlineTypeNode that is initialized in the constructor in the caller map because there can be data flow merges. Instead, we need to explicitly update the node in the exit map. I added corresponding regression tests.
>> 
>> I also noticed that we are extensively cloning InlineTypeNodes during parsing, which often leads to an explosion of redundant nodes that then need to be removed. I added `InlineTypeNode::clone_if_required` to mitigate this, this can probably be further improved (I added corresponding ToDo's).
>> 
>> There are still issues with deoptimization from the constructor (`-XX:+DeoptimizeALot`). I'll address these in another fix.
>> 
>> I also disabled an IR rule in `TestZGCBarrierElision`. The issue will be fixed by [JDK-8329234](https://bugs.openjdk.org/browse/JDK-8329234).
>> 
>> Thanks,
>> Tobias
>
> Tobias Hartmann has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Handling abstract value class constructors with arguments

Found more issues when there are deeply nested constructors, i.e., calls from one constructor to another. Will continue to work on this after the easter break.

-------------

PR Comment: https://git.openjdk.org/valhalla/pull/1064#issuecomment-2025592430



More information about the valhalla-dev mailing list