JIT: Enable CSE for VectorX.Create

[EgorBo]

A low hanging fruit a noticed while I was triaging #13811, but it'd be also nice to allow CSE for GT_LIST-type of args in future.

This PR enables CSE (and hoisting) for Vector128.Create (and other functions with variable amount of args where the actual amount of them is zero or one).

Example:

voidTest(){for(inti=0;i<100;i++){varx=Sse2.MoveMask(Vector128.Create(42).AsByte());Console.WriteLine(x);}}

codegen diff: https://www.diffchecker.com/LMZuC6QS

jit-diff (-f):

PMI CodeSize Diffs for System.Private.CoreLib.dll, framework assemblies for default jit
Summary of Code Size diffs:
(Lower is better)
Total bytes of base: 53048870
Total bytes of diff: 53048870
Total bytes of delta: 0 (0.00% of base)
0 total files with Code Size differences (0 improved, 0 regressed), 271 unchanged.
Top method regressions (bytes):
8 ( 3.01% of base) : System.Private.CoreLib.dasm - HexConverter:EncodeToUtf16_Ssse3(ReadOnlySpan`1,Span`1,int)
Top method improvements (bytes):
-4 (-1.41% of base) : System.Private.CoreLib.dasm - ASCIIUtility:GetIndexOfFirstNonAsciiChar_Default(long,long):long
-4 (-1.40% of base) : System.Private.CoreLib.dasm - Latin1Utility:GetIndexOfFirstNonLatin1Char_Default(long,long):long
Top method regressions (percentages):
8 ( 3.01% of base) : System.Private.CoreLib.dasm - HexConverter:EncodeToUtf16_Ssse3(ReadOnlySpan`1,Span`1,int)
Top method improvements (percentages):
-4 (-1.41% of base) : System.Private.CoreLib.dasm - ASCIIUtility:GetIndexOfFirstNonAsciiChar_Default(long,long):long
-4 (-1.40% of base) : System.Private.CoreLib.dasm - Latin1Utility:GetIndexOfFirstNonLatin1Char_Default(long,long):long
3 total methods with Code Size differences (2 improved, 1 regressed), 268544 unchanged.

The regression is actually a perf improvement: https://www.diffchecker.com/54f2vVgR

[EgorBo]

PTAL @dotnet/jit-contrib @tannergooding

[tannergooding]

Just wondering, why 0 or 1 and not 0, 1, or 2?

[tannergooding]

Variable length args only impact Vector.Create calls for ARM, but also impact cases like Sse.ReciprocalScalar for x86, where the upper limit is 2 args (and therefore op1 will not be a GT_LIST, but NumArgs will be -1).

[EgorBo]

Initially I wanted to add support for args >= 2 cases later but since you noticed this case I've added it as part of this pr.

Repro:

staticintTest(){returnSse2.MoveMask(Sse2.Add(Vector128.Create(1,1),Vector128.Create(1,1)).AsByte());}

New codegen:

; Method Program:Test():intG_M58954_IG01:vzeroupper ;; bbWeight=1 PerfScore 1.00G_M58954_IG02:vmovupdxmm0, xmmword ptr [reloc @RWD00]vpaddqxmm0,xmm0,xmm0 vpmovmskb eax,xmm0 ;; bbWeight=1 PerfScore 4.33G_M58954_IG03:ret ;; bbWeight=1 PerfScore 1.00RWD00 dq	0000000000000001h,0000000000000001h; Total bytes of code: 20