JDK-8302071 : StressIGVN: not just initial shuffle_worklist: always pick next node at random
  • Type: Enhancement
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 21
  • Priority: P4
  • Status: In Progress
  • Resolution: Unresolved
  • Submitted: 2023-02-08
  • Updated: 2023-06-05
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
Other
tbdUnresolved
Related Reports
Relates :  
JDK-8298951 - Umbrella: improve CCP and IGVN verification
Description
With StressIGVN, we shuffle_worklist at the beginning of PhaseIterGVN::optimize and then linearly work down the worklist. This means that nodes that get put on the worklist during IGVN are not randomized.

We should instead do it like in PhaseCCP::analyze, where we always ask for the next node via PhaseCCP::fetch_next_node, which picks a random next node from the worklist if StressCCP is enabled.
Comments
I tried this a while ago, and it was difficult to avoid timeouts. I don't know exactly why. But one explanation may be that the delays become quite unpredictable. The issue is that a single node may get delayed many time. An alternative idea: Delay specific nodes in IGVN for X times. Delay at most Y many nodes. This makes termination more deterministic. And the delays are a bit more tagetted.
05-06-2023