[JitDiff X64] [EgorBo] JIT: Stack allocate non-escaping strings

[MihuBot]

Job completed in 22 minutes 54 seconds (remote runner delay: 1 minute 14 seconds).

Using arguments: -NoPRLink -nuget
Main commit: dotnet/runtime@aa87a4e
PR commit: EgorBo/runtime-1@f6bfc0b

Diffs

Found 110 files with textual diffs.

Summary of Code Size diffs:
(Lower is better)

Total bytes of base: 391415576
Total bytes of diff: 391415779
Total bytes of delta: 203 (0.00 % of base)
Total relative delta: Infinity
    diff is a regression.
    relative diff is a regression.


Total byte diff includes 0 bytes from reconciling methods
	Base had    0 unique methods,        0 unique bytes
	Diff had    1 unique methods,        0 unique bytes

Top file regressions (bytes):
         133 : PdfSharp.dasm (0.02 % of base)
          40 : Prometheus.NetStandard.dasm (0.02 % of base)
          30 : ClosedXML.dasm (0.00 % of base)

3 total files with Code Size differences (0 improved, 3 regressed), 757 unchanged.

Top method regressions (bytes):
         133 (2.99 % of base) : PdfSharp.dasm - PdfSharp.Pdf.PdfMetadata+MetadataEncoder:GetString(byte[]):System.String (FullOpts)
          40 (Infinity of base) : Prometheus.NetStandard.dasm - Prometheus.GaugeConfiguration:.cctor() (FullOpts)
          30 (20.83 % of base) : ClosedXML.dasm - ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:<ValidateName>b__0(char):bool:this (FullOpts)

Top method regressions (percentages):
          40 (Infinity of base) : Prometheus.NetStandard.dasm - Prometheus.GaugeConfiguration:.cctor() (FullOpts)
          30 (20.83 % of base) : ClosedXML.dasm - ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:<ValidateName>b__0(char):bool:this (FullOpts)
         133 (2.99 % of base) : PdfSharp.dasm - PdfSharp.Pdf.PdfMetadata+MetadataEncoder:GetString(byte[]):System.String (FullOpts)

3 total methods with Code Size differences (0 improved, 3 regressed), 2084329 unchanged.

--------------------------------------------------------------------------------

Artifacts:

• PackageList.json (203 KB)
• ShortDiffsRegressions.md (7 KB)
• diff-frameworks.txt (1 KB)
• LongDiffsRegressions.md (121 KB)
• jit-diffs-main.7z (375 MB)
• jit-diffs-pr.7z (376 MB)

[MihuBot]

Top method regressions

30 (20.83 % of base) - ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:b__0(char):bool:this

 ; Assembly listing for method ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:<ValidateName>b__0(char):bool:this (FullOpts)
 ; Emitting BLENDED_CODE for x64 + VEX + EVEX on Unix
 ; FullOpts code
 ; optimized code
 ; rbp based frame
 ; partially interruptible
 ; No PGO data
 ; 0 inlinees with PGO data; 12 single block inlinees; 3 inlinees without PGO data
 ; Final local variable assignments
 ;
 ;  V00 this         [V00,T05] (  3,  3   )     ref  ->  rdi         this class-hnd single-def <ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0>
 ;  V01 arg1         [V01,T06] (  3,  3   )  ushort  ->  rbx         ld-addr-op single-def
 ;# V02 OutArgs      [V02    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace" <Empty>
-;  V03 tmp1         [V03,T00] (  5,  8   )     ref  ->  r15         class-hnd exact single-def "impAppendStmt" <System.String>
+;  V03 tmp1         [V03,T01] (  5,  8   )     ref  ->  rax         class-hnd exact single-def "impAppendStmt" <System.String>
 ;  V04 tmp2         [V04,T02] (  3,  6   )     ref  ->  r15         class-hnd exact single-def "Inlining Arg" <System.String>
 ;* V05 tmp3         [V05    ] (  0,  0   )     ref  ->  zero-ref   
 ;* V06 tmp4         [V06    ] (  0,  0   )     ref  ->  zero-ref   
 ;  V07 tmp5         [V07,T03] (  3,  6   )     ref  ->  rax         class-hnd exact single-def "Inlining Arg" <System.Globalization.TextInfo>
-;  V08 tmp6         [V08,T07] (  2,  4   )  ushort  ->  rbx         "Inlining Arg"
-;  V09 tmp7         [V09,T01] (  4,  7   )     ref  ->  rax         class-hnd exact single-def "dup spill" <System.String>
+;  V08 tmp6         [V08,T07] (  2,  4   )  ushort  ->  rdx         "Inlining Arg"
+;  V09 tmp7         [V09,T00] (  6,  9   )    long  ->  rsi         class-hnd exact "dup spill" <System.String>
 ;* V10 tmp8         [V10    ] (  0,  0   )  struct ( 8) zero-ref    ld-addr-op "Inline stloc first use temp" <System.Runtime.CompilerServices.TypeHandle>
 ;* V11 tmp9         [V11    ] (  0,  0   )  struct ( 8) zero-ref    ld-addr-op "NewObj constructor temp" <System.Runtime.CompilerServices.TypeHandle>
 ;  V12 tmp10        [V12,T08] (  4,  2.50)   ubyte  ->  rax         "Inline return value spill temp"
-;  V13 tmp11        [V13,T09] (  2,  2   )   byref  ->  r15         single-def "impAppendStmt"
-;* V14 tmp12        [V14    ] (  0,  0   )    long  ->  zero-ref    "field V10.m_asTAddr (fldOffset=0x0)" P-INDEP
-;* V15 tmp13        [V15    ] (  0,  0   )    long  ->  zero-ref    "field V11.m_asTAddr (fldOffset=0x0)" P-INDEP
-;  V16 tmp14        [V16,T04] (  3,  6   )     ref  ->  rax         single-def "argument with side effect"
-;  V17 cse0         [V17,T10] (  2,  1   )     int  ->  rcx         "CSE #01: moderate"
+;  V13 tmp11        [V13,T09] (  2,  2   )   byref  ->  rdi         single-def "impAppendStmt"
+;  V14 tmp12        [V14    ] (  4,  4   )  struct (16) [rbp-0x20]  do-not-enreg[XS] addr-exposed "stack allocated string temp" <UNNAMED>
+;* V15 tmp13        [V15    ] (  0,  0   )    long  ->  zero-ref    "field V10.m_asTAddr (fldOffset=0x0)" P-INDEP
+;* V16 tmp14        [V16    ] (  0,  0   )    long  ->  zero-ref    "field V11.m_asTAddr (fldOffset=0x0)" P-INDEP
+;  V17 tmp15        [V17,T04] (  3,  6   )     ref  ->  rax         single-def "argument with side effect"
+;  V18 cse0         [V18,T10] (  3,  1.50)     int  ->  rdx         "CSE #01: moderate"
 ;
-; Lcl frame size = 0
+; Lcl frame size = 16
 
 G_M24541_IG01:
        push     rbp
        push     r15
        push     rbx
-       lea      rbp, [rsp+0x10]
+       sub      rsp, 16
+       lea      rbp, [rsp+0x20]
        mov      ebx, esi
-						;; size=11 bbWeight=1 PerfScore 3.75
+						;; size=15 bbWeight=1 PerfScore 4.00
 G_M24541_IG02:
        mov      r15, gword ptr [rdi+0x08]
        cmp      byte  ptr [r15], r15b
        mov      rax, 0xD1FFAB1E      ; code for System.Globalization.CultureInfo:get_CurrentCulture():System.Globalization.CultureInfo
        call     [rax]System.Globalization.CultureInfo:get_CurrentCulture():System.Globalization.CultureInfo
        mov      rdi, rax
        mov      rax, qword ptr [rax]
        mov      rax, qword ptr [rax+0x48]
        call     [rax+0x38]System.Globalization.CultureInfo:get_TextInfo():System.Globalization.TextInfo:this
        cmp      byte  ptr [rax], al
        mov      rdi, rax
        mov      rsi, r15
        mov      rax, 0xD1FFAB1E      ; code for System.Globalization.TextInfo:ChangeCaseCommon[System.Globalization.TextInfo+ToUpperConversion](System.Globalization.TextInfo,System.String):System.String
        call     [rax]System.Globalization.TextInfo:ChangeCaseCommon[System.Globalization.TextInfo+ToUpperConversion](System.Globalization.TextInfo,System.String):System.String
-       mov      r15, rax
-       movzx    rbx, bx
-       mov      rdi, 0xD1FFAB1E      ; System.String
-       mov      esi, 1
-       call     System.String:FastAllocateString(ptr,nint):System.String
-       mov      word  ptr [rax+0x0C], bx
-       cmp      byte  ptr [r15], r15b
-       cmp      r15, rax
+       movzx    rdx, bx
+       vxorps   xmm0, xmm0, xmm0
+       vmovups  xmmword ptr [rbp-0x20], xmm0
+       mov      rsi, 0xD1FFAB1E      ; System.String
+       mov      qword ptr [rbp-0x20], rsi
+       lea      rsi, [rbp-0x20]
+       mov      dword ptr [rsi+0x08], 1
+       lea      rsi, [rbp-0x20]
+       mov      word  ptr [rsi+0x0C], dx
+       cmp      byte  ptr [rax], al
+       cmp      rax, rsi
        je       SHORT G_M24541_IG05
-						;; size=90 bbWeight=1 PerfScore 26.50
+						;; size=104 bbWeight=1 PerfScore 29.33
 G_M24541_IG03:
-       mov      ecx, dword ptr [r15+0x08]
-       cmp      ecx, 1
+       test     rsi, rsi
+       je       SHORT G_M24541_IG04
+       mov      edx, dword ptr [rax+0x08]
+       cmp      edx, dword ptr [rsi+0x08]
        jne      SHORT G_M24541_IG04
-       add      r15, 12
-       movzx    rcx, word  ptr [r15]
-       cmp      cx, word  ptr [rax+0x0C]
-       sete     al
-       movzx    rax, al
+       lea      rdi, bword ptr [rax+0x0C]
+       add      edx, edx
+       add      rsi, 12
+       mov      rax, 0xD1FFAB1E      ; code for System.SpanHelpers:SequenceEqual(byref,byref,nuint):bool
+       call     [rax]System.SpanHelpers:SequenceEqual(byref,byref,nuint):bool
        jmp      SHORT G_M24541_IG06
-						;; size=29 bbWeight=0.50 PerfScore 5.12
+						;; size=37 bbWeight=0.50 PerfScore 6.25
 G_M24541_IG04:
        xor      eax, eax
        jmp      SHORT G_M24541_IG06
 						;; size=4 bbWeight=0.50 PerfScore 1.12
 G_M24541_IG05:
        mov      eax, 1
 						;; size=5 bbWeight=0.50 PerfScore 0.12
 G_M24541_IG06:
+       add      rsp, 16
        pop      rbx
        pop      r15
        pop      rbp
        ret      
-						;; size=5 bbWeight=1 PerfScore 2.50
+						;; size=9 bbWeight=1 PerfScore 2.75
 
-; Total bytes of code 144, prolog size 9, PerfScore 39.12, instruction count 43, allocated bytes for code 144 (MethodHash=ef50a022) for method ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:<ValidateName>b__0(char):bool:this (FullOpts)
+; Total bytes of code 174, prolog size 13, PerfScore 43.58, instruction count 50, allocated bytes for code 174 (MethodHash=ef50a022) for method ClosedXML.Excel.XLHelper+<>c__DisplayClass46_0:<ValidateName>b__0(char):bool:this (FullOpts)
 ; ============================================================

Note: some changes were skipped as they were too large to fit into a comment.

Larger list of diffs: https://gist.github.com/MihuBot/70c998eca57a00776f11ecd0610ab76d

[MihuBot]

@EgorBo