Test failure: JIT\\Regression\\JitBlue\\GitHub_23159\\GitHub_23159\\GitHub_23159.cmd

[BruceForstall]

win-arm64 random JitStress

https://dev.azure.com/dnceng-public/public/_build/results?buildId=47392&view=ms.vss-test-web.build-test-results-tab&runId=954268&resultId=102958&paneView=dotnet-dnceng.dnceng-build-release-tasks.helix-test-information-tab

set COMPlus_TieredCompilation=0
set COMPlus_JitStress=e7

    JIT\Regression\JitBlue\GitHub_23159\GitHub_23159\GitHub_23159.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\9EF6090D\w\AEE409BF\uploads\Reports\JIT.Regression\JitBlue\GitHub_23159\GitHub_23159\GitHub_23159.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\9EF6090D\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  GitHub_23159.dll 
      Expected: 100
      Actual: -1
      END EXECUTION - FAILED

fwiw, this test has a lot of Vector<T> and Unsafe code

@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

---

win-arm64 random JitStress

https://dev.azure.com/dnceng-public/public/_build/results?buildId=47392&view=ms.vss-test-web.build-test-results-tab&runId=954268&resultId=102958&paneView=dotnet-dnceng.dnceng-build-release-tasks.helix-test-information-tab

set COMPlus_TieredCompilation=0
set COMPlus_JitStress=e7

    JIT\Regression\JitBlue\GitHub_23159\GitHub_23159\GitHub_23159.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\9EF6090D\w\AEE409BF\uploads\Reports\JIT.Regression\JitBlue\GitHub_23159\GitHub_23159\GitHub_23159.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\9EF6090D\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  GitHub_23159.dll 
      Expected: 100
      Actual: -1
      END EXECUTION - FAILED

fwiw, this test has a lot of Vector<T> and Unsafe code

@dotnet/jit-contrib

Author:	BruceForstall
Assignees:	-
Labels:	JitStress, area-CodeGen-coreclr, blocking-clean-ci-optional
Milestone:	8.0.0

[BruceForstall]

More cases in non-random stress run, including MinOpts: https://dev.azure.com/dnceng-public/public/_build/results?buildId=47390&view=ms.vss-test-web.build-test-results-tab&runId=954158&resultId=106563&paneView=debug

[AndyAyersMS]

I'll take a look.

[AndyAyersMS]

Repros with

set COMPlus_TieredCompilation=0
set COMPlus_JitMinOptsName=BytesOrdinalEqualsStringAndAscii

[AndyAyersMS]

The AllTrue vector is not properly set up, and per element, contains FF00 instead of FFFF

runtime/src/tests/JIT/Regression/JitBlue/GitHub_23159/GitHub_23159.cs

Line 122 in 9bcbf50

var AllTrue = new Vector<ushort>(ushort.MaxValue);

AllTrue ==> <65280, 65280, 65280, 65280, 65280, 65280, 65280, 65280>

during codegen we have

Generating: N1014 (  1,  4) [000426] -c---------                  t426 =    CNS_INT   int    0xFFFF REG NA
                                                                        /--*  t426   int    
Generating: N1016 (  3,  2) [000427] -----------                  t427 = *  HWINTRINSIC simd16 ushort DuplicateToVector128 REG d0
IN0181:                           movi    v0.8h, #0xFF,  MSL #8

[AndyAyersMS]

Seems like we may be encoding this immediate incorrectly. We believe we're emitting MSL but we actually emit LSL

;; jit
IN01ae: 0006F4  4F07A7E0          movi    v0.8h, #0xFF,  MSL #8

;; vs disassembler
00007FFDB79542EC 4F07A7E0             movi        v0.8h,#0xFF,lsl #8

cc @tannergooding @TIHan

[AndyAyersMS]

Suspect IF_DV_1B needs to check for an MSL and do something different when it has EA_2BYTE elements, but don't know the encoding well enough to say for sure.

[AndyAyersMS]

Or maybe the issue is upstream? Per C7.2.204 MOVI in the arm64 specs it looks like there is no 16 bit MSL variant.

[AndyAyersMS]

FYI @tannergooding @TIHan -- suspect this may be an arm64 encoding issue.

[TIHan]

I'll take a look.

[TIHan]

I actually get an assert when trying to disasm this:

Assert failure(PID 32712 [0x00007fc8], Thread: 33816 [0x8418]): Assertion failed '!varTypeIsStruct(inlArgNode->gtType) && !varTypeIsStruct(sigType)' in 'GitHub_23159:BytesOrdinalEqualsStringAndAscii(System.String,System.Span`1[ubyte]):bool' during 'Morph - Inlining' (IL size 976; hash 0x1a3ab2eb; FullOpts)

    File: C:\work\runtime\src\coreclr\jit\importer.cpp Line: 13523
    Image: C:\work\runtime\artifacts\bin\coreclr\windows.x64.Checked\CoreRun.exe
ERROR:
Assert failure(PID 32712 [0x00007fc8], Thread: 33816 [0x8418]): Assertion failed '!varTypeIsStruct(inlArgNode->gtType) && !varTypeIsStruct(sigType)' in 'GitHub_23159:BytesOrdinalEqualsStringAndAscii(System.String,System.Span`1[ubyte]):bool' during 'Morph - Inlining' (IL size 976; hash 0x1a3ab2eb; FullOpts)

    File: C:\work\runtime\src\coreclr\jit\importer.cpp Line: 13523
    Image: C:\work\runtime\artifacts\bin\coreclr\windows.x64.Checked\CoreRun.exe

Probably unrelated and just a limitation of disasm since it is occurring in importer.cpp.

[BruceForstall]

@TIHan Note that the failure is on win-arm64 and you're using win-x64. If you're using altjit, it might not work.

[TIHan]

Yes, I was using altjit, but I did manage to isolate it:

    static int Main(string[] args)
    {
        var AllTrue = new Vector<ushort>(ushort.MaxValue);
        Console.Write(AllTrue);

        return 0;
    }

disasm shows:

movi    v16.8h, #0xFF,  MSL #8

[TIHan]

My guess is that emitIns_valid_imm_for_movi is returning true in this case, which it should not, or, there is logic that is not calling this.