JIT: some lclvars related cleanup

[AndyAyersMS]

Consolidate various compiler globals used when setting local var ref
counts by folding them into the visitor:

• lvaMarkRefsCurBlock
• lvaMarkRefsCurStmt
• lvaMarkRefsWeight

Remove the largely vestigial lvPrefReg and associated methods to set
or modify this field. Haven't verified but this is likely a remmant of
the legacy backend.

In the one remaning use (lcl var sorting predicates), swap in lvIsRegArg
instead, which gets most of the same cases.

[AndyAyersMS]

@dotnet/jit-contrib PTAL

Some minimal diffs:

PMI Diffs for System.Private.CoreLib.dll, framework assemblies for x64 default jit
Summary:
(Lower is better)
Total bytes of diff: 4 (0.00% of base)
    diff is a regression.
Top file regressions by size (bytes):
           9 : System.Private.DataContractSerialization.dasm (0.00% of base)
Top file improvements by size (bytes):
          -5 : Microsoft.CodeAnalysis.CSharp.dasm (0.00% of base)
2 total files with size differences (1 improved, 1 regressed), 128 unchanged.
Top method regressions by size (bytes):
           9 : System.Private.DataContractSerialization.dasm - XmlBufferReader:Equals2(int,int,ref):bool:this (3 methods)
Top method improvements by size (bytes):
          -5 : Microsoft.CodeAnalysis.CSharp.dasm - DataFlowPass:UseNonFieldSymbolUnsafely(ref):ref:this
2 total methods with size differences (1 improved, 1 regressed), 170552 unchanged.

Total bytes of diff: 13 (0.00% of base)
    diff is a regression.
Top file regressions by size (bytes):
           9 : System.Private.DataContractSerialization.dasm (0.00% of base)
           4 : Microsoft.CodeAnalysis.CSharp.dasm (0.00% of base)
2 total files with size differences (0 improved, 2 regressed), 128 unchanged.
Top method regressions by size (bytes):
           9 : System.Private.DataContractSerialization.dasm - XmlBufferReader:Equals2(int,int,ref):bool:this (3 methods)
           4 : Microsoft.CodeAnalysis.CSharp.dasm - Binder:BindConditionalLogicalOperator(ref,ref,ref,ref):ref:this
2 total methods with size differences (0 improved, 2 regressed), 142656 unchanged.

[CarolEidt]

Nice! I'm loving all this cleanup. I just had one comment suggestion.

[CarolEidt]

Might be nice to add a bit more detail here about what the "and more" is - e.g. "In addition to increment local var reference counts, it identifies additional local var properties such as those used by assertion prop."

[AndyAyersMS]

Thanks. I was surprised how much it did and never got back to filling it all in.

Ok if I do this in my next change? I need to make sure that the only required part for minopts is having nonzero ref counts.

[CarolEidt]

Fine with me, this is already way better than what was there!

[AndyAyersMS]

OSX failures look like the somewhat typical IO pal test failures:

file_io.GetFileSize.test1
file_io.GetFileSizeEx.test1
file_io.SetFilePointer.test5
file_io.SetFilePointer.test6
file_io.SetFilePointer.test7

@dotnet-bot retest OSX10.12 x64 Checked Innerloop Build and Test

[briansull]

Looks Good

[briansull]

The value here '0xdddddddd' depends upon the value JitDefaultFill, we should fix this to call the method:
T UninitializedWord(Compiler* comp)
so that it compares with the correct value when JitDefaultFill isn't 0xdd.
Also code like this cause problems if we try to set JitDefaultFill to the value of 0x00 and sometimes 0x01 causes problems.

[AndyAyersMS]

Now that the weight always comes in via a parameter and not a field on the compiler instance this check seems much less useful anyways.

So I'd suggest we just remove it. Will do this in my next round of cleanups.