Fix incorrect switch temp lcl type

[mikedn]

Fixes #13486

[mikedn]

x64 jit-diff fx:

Total bytes of diff: -7 (0.00% of base)
    diff is an improvement.
Total byte diff includes 0 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had    0 unique methods,        0 unique bytes
Top file regressions by size (bytes):
           2 : System.Diagnostics.Process.dasm (0.00% of base)
Top file improvements by size (bytes):
          -6 : Microsoft.CodeAnalysis.CSharp.dasm (0.00% of base)
          -2 : Microsoft.CodeAnalysis.VisualBasic.dasm (0.00% of base)
          -1 : Microsoft.CodeAnalysis.dasm (0.00% of base)
4 total files with size differences (3 improved, 1 regressed), 75 unchanged.
Top method regessions by size (bytes):
           2 : System.Diagnostics.Process.dasm - Process:GetShellError(long):int:this
           1 : Microsoft.CodeAnalysis.CSharp.dasm - BaseTypeAnalysis:IsManagedTypeHelper(ref):ubyte
           1 : Microsoft.CodeAnalysis.CSharp.dasm - SourceNamespaceSymbol:BuildSymbol(ref,ref):ref:this
           1 : Microsoft.CodeAnalysis.dasm - SwitchConstantValueHelper:IsValidSwitchCaseLabelConstant(ref):bool
Top method improvements by size (bytes):
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - Lexer:ScanIdentifier_FastPath(byref):bool:this
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - ConversionEasyOut:TypeToIndex(ref):int
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - BinopEasyOut:TypeToIndex(ref):struct
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - UnopEasyOut:TypeToIndex(ref):struct
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - VarianceSafety:IsVarianceUnsafe(ref,bool,bool,ref,ref,ref,ref):bool (2 methods)
16 total methods with size differences (12 improved, 4 regressed), 66256 unchanged.

x86 jit-diff fx:

Total bytes of diff: 11 (0.00% of base)
    diff is a regression.
Total byte diff includes 0 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had    0 unique methods,        0 unique bytes
Top file regressions by size (bytes):
          18 : System.Private.CoreLib.dasm (0.00% of base)
           1 : Microsoft.CodeAnalysis.dasm (0.00% of base)
Top file improvements by size (bytes):
          -5 : Microsoft.CodeAnalysis.CSharp.dasm (0.00% of base)
          -3 : Microsoft.CodeAnalysis.VisualBasic.dasm (0.00% of base)
4 total files with size differences (2 improved, 2 regressed), 75 unchanged.
Top method regessions by size (bytes):
          18 : System.Private.CoreLib.dasm - EventSource:GetHelperCallFirstArg(ref):int
           2 : Microsoft.CodeAnalysis.CSharp.dasm - ConversionEasyOut:TypeToIndex(ref):int
           1 : Microsoft.CodeAnalysis.CSharp.dasm - BaseTypeAnalysis:IsManagedTypeHelper(ref):ubyte
           1 : Microsoft.CodeAnalysis.CSharp.dasm - SourceNamespaceSymbol:BuildSymbol(ref,ref):ref:this
           1 : Microsoft.CodeAnalysis.dasm - SwitchConstantValueHelper:IsValidSwitchCaseLabelConstant(ref):bool
Top method improvements by size (bytes):
          -3 : Microsoft.CodeAnalysis.CSharp.dasm - BinderFactoryVisitor:VisitXmlNameAttribute(ref):ref:this
          -2 : Microsoft.CodeAnalysis.CSharp.dasm - DocumentationCommentCompiler:BindName(ref,ref,ref,byref,byref,ref)
          -2 : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:ReportOverloadResolutionFailureAndProduceBoundNode(ref,int,ref,struct,ref,struct,struct,ref,ref,ref,ref,bool,ref,ref,ref):ref:this
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - Lexer:ScanIdentifier_FastPath(byref):bool:this
          -1 : Microsoft.CodeAnalysis.CSharp.dasm - VarianceSafety:IsVarianceUnsafe(ref,bool,bool,ref,ref,ref,ref):bool (2 methods)
13 total methods with size differences (8 improved, 5 regressed), 66198 unchanged.

Most of the diffs look like:

        movsx    eax, al
-       cmp      al, 38
+       cmp      eax, 38
        ja       SHORT G_M33493_IG08

The 18 byte regression from EventSource:GetHelperCallFirstArg has to do with register allocation choices. While unrelated to this it looks kind of interesting so here's a full diff: https://gist.github.com/mikedn/15e61328a8fe0812268125af04fa2b89

[mikedn]

@dotnet-bot test Tizen armel Cross Debug Build

[mikedn]

It's probably worth pointing out that on x64 this bug results in silent bad code generation. For example the JIT generated:

       4883F802             cmp      rax, 2
       7748                 ja       SHORT G_M55888_IG08
       488D0D65000000       lea      rcx, [reloc @RWD00]
       8B0C81               mov      ecx, dword ptr [rcx+4*rax]

instead of the correct:

       83F802               cmp      eax, 2
       774A                 ja       SHORT G_M55888_IG08
       8BC0                 mov      eax, eax
       488D0D64000000       lea      rcx, [reloc @RWD00]
       8B0C81               mov      ecx, dword ptr [rcx+4*rax]

But it's unlikely to hit this in practice, at least in C# where this pattern occurs when switching on longs and the C# compiler adds its own long compare before the switch:

G_M55888_IG02:
       488BC1               mov      rax, rcx
       4883F802             cmp      rax, 2
       774E                 ja       SHORT G_M55888_IG08
G_M55888_IG03:
; yeah, it's wrong if the upper 32 bits aren't 0 but that will never happen due to the above check
       4883F802             cmp      rax, 2 
       7748                 ja       SHORT G_M55888_IG08
       488D0D65000000       lea      rcx, [reloc @RWD00]
       8B0C81               mov      ecx, dword ptr [rcx+4*rax]

[mazong1123]

@dotnet-bot test Ubuntu x64 corefx_baseline
@dotnet-bot test Windows_NT x86 corefx_baseline

Just make sure this PR fixes the CI failure...

[JosephTremoulet]

LGTM, thanks for fixing this.