Test failure: JIT\\jit64\\opt\\lur\\lur_02\\lur_02.cmd

[BruceForstall]

arm/arm64, JitStress

https://dev.azure.com/dnceng-public/public/_build/results?buildId=37829&view=ms.vss-test-web.build-test-results-tab&runId=754244&resultId=103099&paneView=debug

    JIT\jit64\opt\lur\lur_02\lur_02.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\9EAB08FE\w\A4B20948\uploads\Reports\JIT.jit64\opt\lur\lur_02\lur_02.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\9EAB08FE\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  lur_02.dll 
      FAIL: test_05(3)
      FAIL: test_05(5)
      Expected: 100
      Actual: 1
      END EXECUTION - FAILED
      FAILED

Possibly related: #76506

@dotnet/jit-contrib

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

arm/arm64, JitStress

https://dev.azure.com/dnceng-public/public/_build/results?buildId=37829&view=ms.vss-test-web.build-test-results-tab&runId=754244&resultId=103099&paneView=debug

    JIT\jit64\opt\lur\lur_02\lur_02.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\9EAB08FE\w\A4B20948\uploads\Reports\JIT.jit64\opt\lur\lur_02\lur_02.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\9EAB08FE\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  lur_02.dll 
      FAIL: test_05(3)
      FAIL: test_05(5)
      Expected: 100
      Actual: 1
      END EXECUTION - FAILED
      FAILED

Possibly related: #76506

@dotnet/jit-contrib

Author:	BruceForstall
Assignees:	-
Labels:	JitStress, area-CodeGen-coreclr
Milestone:	8.0.0

[BruceForstall]

@AndyAyersMS This is due to #76283

I can repro with:

set COMPlus_TieredCompilation=0
set COMPlus_JitStressModeNames=STRESS_BB_PROFILE STRESS_UNROLL_LOOPS

RBO with Phis is also what is creating the weird flow graph (at least in conjunction with odd stress-mode weights).

If I change:

return optJumpThreadPhi(block, tree, treeNormVN);

to:

return false;

the failure doesn't repro.

[BruceForstall]

It looks to me like RBO is changing the flow graph but not updating/moving PHIs. E.g., for test_01 BB46 has one pred and BB47 two, with two PHIs. After, BB46 had two preds and BB47 one, but the PHIs remained in BB47 and no PHIs were added to BB46. CSE follows and it looks like we generate an illegal CSE of a loop-dependent variable V01 and use it outside the loop.

[AndyAyersMS]

Yes, I am seeing a similar issue with #76636.

I think I may need to disable optJumpThreadPhi until I can add proper safeguards.

[BruceForstall]

Fixed by #76695