GitHub_17777 test failing on linux-arm64

[am11]

From outerloop (Pri1) test log: https://helix.dot.net/api/2019-06-17/jobs/e865e8bd-955d-4179-af2a-4c8d0686f1cc/workitems/JIT.Regression.JitBlue/console

+ dotnet /root/helix/work/correlation/xunit/xunit.console.dll JIT/Regression/JIT.Regression.XUnitWrapper.dll -parallel collections -nocolor -noshadow -xml testResults.xml -trait TestGroup=JIT.Regression.JitBlue
Microsoft.DotNet.XUnitConsoleRunner v2.5.0 (64-bit .NET 6.0.0-rtm.21522.10)
  Discovering: JIT.Regression.XUnitWrapper (method display = ClassAndMethod, method display options = None)
  Discovered:  JIT.Regression.XUnitWrapper (found 224 of 1542 test cases)
  Starting:    JIT.Regression.XUnitWrapper (parallel test collections = on, max threads = 4)
    JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh [FAIL]
      
      Assert failure(PID 526 [0x0000020e], Thread: 526 [0x020e]): Assertion failed 'emitCurIG != emitPrologIG' in 'Repro.Program:Test(int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int):int' during 'Generate code' (IL size 2418)
      
          File: /__w/1/s/src/coreclr/jit/emit.cpp Line: 8610
          Image: /root/helix/work/correlation/corerun

This assert fails:

runtime/src/coreclr/jit/emit.cpp

Line 8610 in 5605bdf

assert(emitCurIG != emitPrologIG);

After that, createdump fails this assert in libunwind:

runtime/src/coreclr/pal/src/libunwind/src/dwarf/Gfind_proc_info-lsb.c

Line 929 in 5605bdf

assert (ip >= di->start_ip && ip < di->end_ip);

      createdump: /__w/1/s/src/coreclr/pal/src/libunwind/src/dwarf/Gfind_proc_info-lsb.c:929: int _Uaarch64_dwarf_search_unwind_table(unw_addr_space_t, unw_word_t, unw_dyn_info_t *, unw_proc_info_t *, int, void *): Assertion `ip >= di->start_ip && ip < di->end_ip' failed.
      /root/helix/work/workitem/e/JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh: line 411:   526 Aborted                 (core dumped) $LAUNCHER $ExePath "${CLRTestExecutionArguments[@]}"
      
      Return code:      1
      Raw output file:      /root/helix/work/workitem/uploads/Reports/JIT.Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.output.txt
      Raw output:
      BEGIN EXECUTION
      /root/helix/work/correlation/corerun -p System.Reflection.Metadata.MetadataUpdater.IsSupported=false GitHub_17777.dll ''
      Expected: 100
      Actual: 134
      END EXECUTION - FAILED
      Test Harness Exitcode is : 1
      To run the test:
      > set CORE_ROOT=/root/helix/work/correlation
      > /root/helix/work/workitem/e/JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh
      Expected: True
      Actual:   False
      Stack Trace:
           at JIT_Regression._JitBlue_GitHub_17777_GitHub_17777_GitHub_17777_._JitBlue_GitHub_17777_GitHub_17777_GitHub_17777_sh()

[ghost]

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

Issue Details

---

From outerloop (Pri1) test log: https://helix.dot.net/api/2019-06-17/jobs/e865e8bd-955d-4179-af2a-4c8d0686f1cc/workitems/JIT.Regression.JitBlue/console

+ dotnet /root/helix/work/correlation/xunit/xunit.console.dll JIT/Regression/JIT.Regression.XUnitWrapper.dll -parallel collections -nocolor -noshadow -xml testResults.xml -trait TestGroup=JIT.Regression.JitBlue
Microsoft.DotNet.XUnitConsoleRunner v2.5.0 (64-bit .NET 6.0.0-rtm.21522.10)
  Discovering: JIT.Regression.XUnitWrapper (method display = ClassAndMethod, method display options = None)
  Discovered:  JIT.Regression.XUnitWrapper (found 224 of 1542 test cases)
  Starting:    JIT.Regression.XUnitWrapper (parallel test collections = on, max threads = 4)
    JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh [FAIL]
      
      Assert failure(PID 526 [0x0000020e], Thread: 526 [0x020e]): Assertion failed 'emitCurIG != emitPrologIG' in 'Repro.Program:Test(int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int,int):int' during 'Generate code' (IL size 2418)
      
          File: /__w/1/s/src/coreclr/jit/emit.cpp Line: 8610
          Image: /root/helix/work/correlation/corerun

This assert fails:

runtime/src/coreclr/jit/emit.cpp

Line 8610 in 5605bdf

assert(emitCurIG != emitPrologIG);

After that, createdump fails this assert in libunwind:

runtime/src/coreclr/pal/src/libunwind/src/dwarf/Gfind_proc_info-lsb.c

Line 929 in 5605bdf

assert (ip >= di->start_ip && ip < di->end_ip);

      createdump: /__w/1/s/src/coreclr/pal/src/libunwind/src/dwarf/Gfind_proc_info-lsb.c:929: int _Uaarch64_dwarf_search_unwind_table(unw_addr_space_t, unw_word_t, unw_dyn_info_t *, unw_proc_info_t *, int, void *): Assertion `ip >= di->start_ip && ip < di->end_ip' failed.
      /root/helix/work/workitem/e/JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh: line 411:   526 Aborted                 (core dumped) $LAUNCHER $ExePath "${CLRTestExecutionArguments[@]}"
      
      Return code:      1
      Raw output file:      /root/helix/work/workitem/uploads/Reports/JIT.Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.output.txt
      Raw output:
      BEGIN EXECUTION
      /root/helix/work/correlation/corerun -p System.Reflection.Metadata.MetadataUpdater.IsSupported=false GitHub_17777.dll ''
      Expected: 100
      Actual: 134
      END EXECUTION - FAILED
      Test Harness Exitcode is : 1
      To run the test:
      > set CORE_ROOT=/root/helix/work/correlation
      > /root/helix/work/workitem/e/JIT/Regression/JitBlue/GitHub_17777/GitHub_17777/GitHub_17777.sh
      Expected: True
      Actual:   False
      Stack Trace:
           at JIT_Regression._JitBlue_GitHub_17777_GitHub_17777_GitHub_17777_._JitBlue_GitHub_17777_GitHub_17777_GitHub_17777_sh()

Author:	am11
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged
Milestone:	-

[am11]

@jkotas, @janvorli, if it is not known, should we create a separate arm64 issue for libunwind assert during createdump?

[jkotas]

if it is not known, should we create a separate arm64 issue for libunwind assert during createdump?

Yes. Opened #64168

[AndyAyersMS]

@BruceForstall if you end up bumping up the IG size for this let's make sure it is big enough for OSR too.

[BruceForstall]

@BruceForstall if you end up bumping up the IG size for this let's make sure it is big enough for OSR too.

@AndyAyersMS Do you have any idea what the maximum number of instructions might be for OSR? Or, put another way, what is the maximum amount bigger (in number of instructions) that an OSR prolog might be compared to a non-OSR prolog (for arm64/x64)?

This issue has only occurred (AFAIK) now on arm64 due when we have very large frames that require loading large local variable constant offsets into a temp register for many initializations.

I'm inclined to just double the arm64 max IG size.

[AndyAyersMS]

Looks like OSR is trying to load up 25 enregistered locals, each load takes 3 instructions on arm64:

IN48985:                           movz    w8, #0xeaf0
IN48986:                           movk    w8, #4 LSL #16
IN48987:                           ldr     w10, [fp, x8]

and there are at least 25 other instructions in the prolog before this (and possibly a few after).

So perhaps 120 instructions worth?

If this is unworkable I can probably common some of the address math and squeeze this down a bit.

[AndyAyersMS]

I think we're ok on x64 for now, latest osr stress only has arm64 failures.

[BruceForstall]

Looks like OSR is trying to load up 25 enregistered locals

That's for this particular test case? Where does the number "25" come from?

In general, is this number capped (is there a maximum?), or can it be unbounded?

[AndyAyersMS]

It's limited by the number of allocatable registers (I think just int registers as we initialize FP registers outside the OS prolog).

[BruceForstall]

It's limited by the number of allocatable registers (I think just int registers as we initialize FP registers outside the OS prolog).

Actually, since we care here about the JIT prolog insGroup, it includes everything in the JIT prolog, including zero inits, moving registers to the LSRA assigned locations, etc. -- not just the OS prolog.

[AndyAyersMS]

Well then it might include initializing all the allocatable FP registers and all the allocatable INT registers.

[AndyAyersMS]

In case you're wondering -- OSR is "unusual" in that a lot more things can be live on entry than in a normal method. Anything live on entry and enregistered gets loaded up from the Tier0 frame in the OSR prolog.

I suppose we could defer all of this loading but it would complicate things quite a bit.