Number of jit tests failing with needLabel assert

[AndyAyersMS]

Saw this first when testing OSR but it seems to be more widespread.

Eg from libraries-pgo run: https://dev.azure.com/dnceng/public/_build/results?buildId=1480057&view=ms.vss-test-web.build-test-results-tab

  Starting:    System.Text.RegularExpressions.Tests (parallel test collections = on, max threads = 4)

Assert failure(PID 27 [0x0000001b], Thread: 33 [0x0021]): Assertion failed 'needLabel' in '<>c__DisplayClass207_0:<TryGenerateSimplifiedGo>g__EmitSingleCharAtomicLoop|22(System.Text.RegularExpressions.RegexNode):this' during 'Generate code' (IL size 1706)

    File: /__w/1/s/src/coreclr/jit/codegenlinear.cpp Line: 348
    Image: /root/helix/work/correlation/dotnet

Underlying issue is that we end up deciding to put align padding before an unreachable block, and then assert that that block does not have a label.

cc @kunalspathak @dotnet/jit-contrib

[ghost]

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

Issue Details

---

Saw this first when testing OSR but it seems to be more widespread.

Eg from libraries-pgo run: https://dev.azure.com/dnceng/public/_build/results?buildId=1480057&view=ms.vss-test-web.build-test-results-tab

  Starting:    System.Text.RegularExpressions.Tests (parallel test collections = on, max threads = 4)

Assert failure(PID 27 [0x0000001b], Thread: 33 [0x0021]): Assertion failed 'needLabel' in '<>c__DisplayClass207_0:<TryGenerateSimplifiedGo>g__EmitSingleCharAtomicLoop|22(System.Text.RegularExpressions.RegexNode):this' during 'Generate code' (IL size 1706)

    File: /__w/1/s/src/coreclr/jit/codegenlinear.cpp Line: 348
    Image: /root/helix/work/correlation/dotnet

Underlying issue is that we end up deciding to put align padding before an unreachable block, and then assert that that block does not have a label.

cc @kunalspathak @dotnet/jit-contrib

Author:	AndyAyersMS
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged
Milestone:	-

[kunalspathak]

This is mostly causing because we sometimes can have back to back jumps as part of same IG. We might have an align instruction after the first jmp and then expect the next instruction (in this case a jmp) has its own label. #62025 will fix this and some other loop alignment related issues.

       E97AFFFFFF           jmp      G_M22669_IG12
       E9A0060000           jmp      G_M22669_IG78