RFR: 8351833: Unexpected increase in live nodes when splitting Phis through MergeMems in PhiNode::Ideal [v3]
Daniel Lundén
dlunden at openjdk.org
Thu Apr 10 19:40:28 UTC 2025
On Thu, 10 Apr 2025 18:52:21 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:
> Is bailout due to infinite nodes generation (which is bug and can be fixed separately) or simply excessive numbers of node and by increasing bailout number limit the compilation will pass?
Compilation does complete with increased `MaxNodeLimit`. The node count is "just" (temporarily) excessive. Have a look at the plot from the PR description below, illustrating the live node count (orange) and IGVN worklist size (blue) throughout the specific problematic IGVN run (with `MaxNodeLimit` increased from the default 80 000 to allow the compilation to complete). The live node count peaks just beyond 80 000 nodes. After completing the IGVN run, the live node count looks normal (see the sharp decrease in the plots just at the end).
- Before the fix for [JDK-8333393](https://bugs.openjdk.org/browse/JDK-8333393) ("baseline")
- After the fix for [JDK-8333393](https://bugs.openjdk.org/browse/JDK-8333393) ("target-old")
- After the fix in this PR ("target")

-------------
PR Comment: https://git.openjdk.org/jdk/pull/24325#issuecomment-2794961261
More information about the hotspot-compiler-dev
mailing list