Assertion failed 'comp->supportSIMDTypes()'

[BruceForstall]

Many tests in "runtime-coreclr jitstress-isas-x86" pipeline, job "CoreCLR OSX x64 Checked jitstress_isas_nosimd" fail, e.g., JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh:

Assert failure(PID 19764 [0x00004d34], Thread: 722343 [0xb05a7]): Assertion failed 'comp->supportSIMDTypes()' in 'r8NaNdiv:d_test2_f(System.String):double' during 'Rationalize IR' (IL size 33; hash 0xcd8ee9f9; FullOpts)

File: /Users/runner/work/1/s/src/coreclr/jit/rationalize.cpp Line: 781
Image: /private/tmp/helix/working/A85C0927/p/corerun
task_for_pid(19764) FAILED 5 (os/kern) failure
/private/tmp/helix/working/A85C0927/w/AA01093A/e/JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh: line 379: 19764 Abort trap: 6           (core dumped) $LAUNCHER $ExePath "${CLRTestExecutionArguments[@]}"

Return code:      1
Raw output file:      /tmp/helix/working/A85C0927/w/AA01093A/uploads/Reports/JIT.Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.output.txt
Raw output:
BEGIN EXECUTION
/tmp/helix/working/A85C0927/p/corerun -p System.Reflection.Metadata.MetadataUpdater.IsSupported=false r8NaNdiv_cs_do.dll ''
Expected: 100
Actual: 134
END EXECUTION - FAILED
Test Harness Exitcode is : 1
To run the test:

set CORE_ROOT=/tmp/helix/working/A85C0927/p
/private/tmp/helix/working/A85C0927/w/AA01093A/e/JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh
Expected: True
Actual:   False

@tannergooding @dotnet/jit-contrib

[ghost]

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

Issue Details

---

Many tests in "runtime-coreclr jitstress-isas-x86" pipeline, job "CoreCLR OSX x64 Checked jitstress_isas_nosimd" fail, e.g., JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh:

Assert failure(PID 19764 [0x00004d34], Thread: 722343 [0xb05a7]): Assertion failed 'comp->supportSIMDTypes()' in 'r8NaNdiv:d_test2_f(System.String):double' during 'Rationalize IR' (IL size 33; hash 0xcd8ee9f9; FullOpts)

File: /Users/runner/work/1/s/src/coreclr/jit/rationalize.cpp Line: 781
Image: /private/tmp/helix/working/A85C0927/p/corerun
task_for_pid(19764) FAILED 5 (os/kern) failure
/private/tmp/helix/working/A85C0927/w/AA01093A/e/JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh: line 379: 19764 Abort trap: 6           (core dumped) $LAUNCHER $ExePath "${CLRTestExecutionArguments[@]}"

Return code:      1
Raw output file:      /tmp/helix/working/A85C0927/w/AA01093A/uploads/Reports/JIT.Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.output.txt
Raw output:
BEGIN EXECUTION
/tmp/helix/working/A85C0927/p/corerun -p System.Reflection.Metadata.MetadataUpdater.IsSupported=false r8NaNdiv_cs_do.dll ''
Expected: 100
Actual: 134
END EXECUTION - FAILED
Test Harness Exitcode is : 1
To run the test:

set CORE_ROOT=/tmp/helix/working/A85C0927/p
/private/tmp/helix/working/A85C0927/w/AA01093A/e/JIT/Methodical/NaN/r8NaNdiv_cs_do/r8NaNdiv_cs_do.sh
Expected: True
Actual:   False

@tannergooding @dotnet/jit-contrib

Author:	BruceForstall
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	7.0.0

[tannergooding]

It's probably worth removing the COMPlus_EnableSIMD=0 switch. It doesn't exist/work for ARM64 and the code has been in long enough for x86/x64 that it's extremely well tested and so the original purpose for it doesn't really "exist" anymore. Additionally on Unix (and to an extent on x64 Windows) the support is "required" for ABI purposes to ensure types like Vector128<T> or HFAs/HVAs are correctly passed around.

Users still have the option of disabling individual ISAs or using the COMPlus_EnableHWIntrinsic, which would then leave on the general SIMD support needed for ABI, but while still allowing testing the various code paths their code has.

@BruceForstall, @echesakovMSFT, thoughts?

[BruceForstall]

You understand the support matrix here much better than I, so I will defer to your judgement. My main concern would be that we should be able to test all the configurations that we support. If COMPlus_EnableSIMD=0 doesn't improve our testability, then I'm fine to remove it.

[kunalspathak]

#66670 fixes this.