Test failure JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd

[VincentBu]

Run: runtime-coreclr jitstress-isas-x86 20211120.1

Failed test:

CoreCLR windows x86 Checked jitstress_isas_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd

CoreCLR windows x86 Checked jitstress_isas_x86_nosse2 @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd

CoreCLR windows x86 Checked jitstress_isas_2_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd

CoreCLR windows x64 Checked jitstress_isas_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd

CoreCLR OSX x64 Checked jitstress_isas_x86_nosse2 @ OSX.1014.Amd64.Open
- JIT/HardwareIntrinsics/X86/X86Base/Pause_r/Pause_r.sh
- JIT/HardwareIntrinsics/X86/X86Base/Pause_ro/Pause_ro.sh

CoreCLR Linux x64 Checked jitstress_isas_2_x86_nohwintrinsic @ Ubuntu.1804.Amd64.Open
- JIT/HardwareIntrinsics/X86/X86Base/Pause_r/Pause_r.sh
- JIT/HardwareIntrinsics/X86/X86Base/Pause_ro/Pause_ro.sh

Error message:

Return code:      1
Raw output file:      C:\h\w\AE9209AF\w\B22209B2\uploads\Reports\JIT.HardwareIntrinsics\X86\X86Base\Pause_ro\Pause_ro.output.txt
Raw output:
BEGIN EXECUTION
"C:\h\w\AE9209AF\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Pause_ro.dll
Expected: 100
Actual: 0
END EXECUTION - FAILED
FAILED
Test Harness Exitcode is : 1
To run the test:

set CORE_ROOT=C:\h\w\AE9209AF\p
C:\h\w\AE9209AF\w\B22209B2\e\JIT\HardwareIntrinsics\X86\X86Base\Pause_ro\Pause_ro.cmd
Expected: True
Actual:   False


Stack trace
   at JIT_HardwareIntrinsics._X86_X86Base_Pause_ro_Pause_ro_._X86_X86Base_Pause_ro_Pause_ro_cmd()

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

Issue Details

---

Run: runtime-coreclr jitstress-isas-x86 20211120.1

Failed test:

CoreCLR windows x86 Checked jitstress_isas_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd

CoreCLR windows x86 Checked jitstress_isas_x86_nosse2 @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd

CoreCLR windows x86 Checked jitstress_isas_2_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd

CoreCLR windows x64 Checked jitstress_isas_x86_nosse @ Windows.10.Amd64.Open
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_r\\Pause_r.cmd
- JIT\\HardwareIntrinsics\\X86\\X86Base\\Pause_ro\\Pause_ro.cmd

CoreCLR OSX x64 Checked jitstress_isas_x86_nosse2 @ OSX.1014.Amd64.Open
- JIT/HardwareIntrinsics/X86/X86Base/Pause_r/Pause_r.sh
- JIT/HardwareIntrinsics/X86/X86Base/Pause_ro/Pause_ro.sh

CoreCLR Linux x64 Checked jitstress_isas_2_x86_nohwintrinsic @ Ubuntu.1804.Amd64.Open
- JIT/HardwareIntrinsics/X86/X86Base/Pause_r/Pause_r.sh
- JIT/HardwareIntrinsics/X86/X86Base/Pause_ro/Pause_ro.sh

Error message:

Return code:      1
Raw output file:      C:\h\w\AE9209AF\w\B22209B2\uploads\Reports\JIT.HardwareIntrinsics\X86\X86Base\Pause_ro\Pause_ro.output.txt
Raw output:
BEGIN EXECUTION
"C:\h\w\AE9209AF\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Pause_ro.dll
Expected: 100
Actual: 0
END EXECUTION - FAILED
FAILED
Test Harness Exitcode is : 1
To run the test:

set CORE_ROOT=C:\h\w\AE9209AF\p
C:\h\w\AE9209AF\w\B22209B2\e\JIT\HardwareIntrinsics\X86\X86Base\Pause_ro\Pause_ro.cmd
Expected: True
Actual:   False


Stack trace
   at JIT_HardwareIntrinsics._X86_X86Base_Pause_ro_Pause_ro_._X86_X86Base_Pause_ro_Pause_ro_cmd()

Author:	VincentBu
Assignees:	-
Labels:	arch-x86, os-linux, os-mac-os-x, os-windows, JitStress, arch-x64, area-CodeGen-coreclr, untriaged
Milestone:	-

[echesakov]

The similar issue (related to usage of Arm.ArmBase.Yield intrinsic) can be reproduced on Arm64 under the same environment

COMPlus_FeatureSIMD=0
COMPlus_TieredCompilation=0

    JIT\HardwareIntrinsics\Arm\ArmBase\Yield_r\Yield_r.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\B92509B8\w\9AD10870\uploads\Reports\JIT.HardwareIntrinsics\Arm\ArmBase\Yield_r\Yield_r.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\B92509B8\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Yield_r.dll 
      Expected: 100
      Actual: 0
      END EXECUTION - FAILED
      FAILED
      Test Harness Exitcode is : 1
      To run the test:
      > set CORE_ROOT=D:\h\w\B92509B8\p
      > D:\h\w\B92509B8\w\9AD10870\e\JIT\HardwareIntrinsics\Arm\ArmBase\Yield_r\Yield_r.cmd
      Expected: True
      Actual:   False
      Stack Trace:
           at JIT_HardwareIntrinsics._Arm_ArmBase_Yield_r_Yield_r_._Arm_ArmBase_Yield_r_Yield_r_cmd()
      Output:
        
        Return code:      1
        Raw output file:      D:\h\w\B92509B8\w\9AD10870\uploads\Reports\JIT.HardwareIntrinsics\Arm\ArmBase\Yield_r\Yield_r.output.txt
        Raw output:
        BEGIN EXECUTION
         "D:\h\w\B92509B8\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Yield_r.dll 
        Expected: 100
        Actual: 0
        END EXECUTION - FAILED
        FAILED
        Test Harness Exitcode is : 1
        To run the test:
        > set CORE_ROOT=D:\h\w\B92509B8\p
        > D:\h\w\B92509B8\w\9AD10870\e\JIT\HardwareIntrinsics\Arm\ArmBase\Yield_r\Yield_r.cmd
    JIT\HardwareIntrinsics\Arm\ArmBase\Yield_ro\Yield_ro.cmd [FAIL]
      
      Return code:      1
      Raw output file:      D:\h\w\B92509B8\w\9AD10870\uploads\Reports\JIT.HardwareIntrinsics\Arm\ArmBase\Yield_ro\Yield_ro.output.txt
      Raw output:
      BEGIN EXECUTION
       "D:\h\w\B92509B8\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Yield_ro.dll 
      Expected: 100
      Actual: 0
      END EXECUTION - FAILED
      FAILED
      Test Harness Exitcode is : 1
      To run the test:
      > set CORE_ROOT=D:\h\w\B92509B8\p
      > D:\h\w\B92509B8\w\9AD10870\e\JIT\HardwareIntrinsics\Arm\ArmBase\Yield_ro\Yield_ro.cmd
      Expected: True
      Actual:   False
      Stack Trace:
           at JIT_HardwareIntrinsics._Arm_ArmBase_Yield_ro_Yield_ro_._Arm_ArmBase_Yield_ro_Yield_ro_cmd()
      Output:
        
        Return code:      1
        Raw output file:      D:\h\w\B92509B8\w\9AD10870\uploads\Reports\JIT.HardwareIntrinsics\Arm\ArmBase\Yield_ro\Yield_ro.output.txt
        Raw output:
        BEGIN EXECUTION
         "D:\h\w\B92509B8\p\corerun.exe" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  Yield_ro.dll 
        Expected: 100
        Actual: 0
        END EXECUTION - FAILED
        FAILED
        Test Harness Exitcode is : 1
        To run the test:
        > set CORE_ROOT=D:\h\w\B92509B8\p
        > D:\h\w\B92509B8\w\9AD10870\e\JIT\HardwareIntrinsics\Arm\ArmBase\Yield_ro\Yield_ro.cmd

[echesakov]

I think Yield/Pause should belong to Category_Scalar rather than Category_Special and table-driven importation can be just extended in the following way

--- a/src/coreclr/jit/hwintrinsic.cpp
+++ b/src/coreclr/jit/hwintrinsic.cpp
@@ -986,8 +986,8 @@ GenTree* Compiler::impHWIntrinsic(NamedIntrinsic        intrinsic,
         switch (numArgs)
         {
             case 0:
-                assert(!isScalar);
-                retNode = gtNewSimdHWIntrinsicNode(retType, intrinsic, simdBaseJitType, simdSize);
+                retNode = isScalar ? gtNewScalarHWIntrinsicNode(retType, intrinsic) : gtNewSimdHWIntrinsicNode(retType, intrinsic, simdBaseJitType, simdSize);
                 break;

             case 1:

@tannergooding What do you think?

[tannergooding]

These are strictly not scalar intrinsics; they are special just like LoadFence, MemoryFence, StoreFence, and others.

The issue with COMPlus_FeatureSIMD would be independent and is possibly due to X86Base and ArmBase not being correctly categorized. We should investigate what the issue there is and fix it so that special/helper intrinsics can still be correctly handled.