Tiny CodeQuality issue in string.AsSpan

[Tornhoof]

I noticed in .NET Core 2.1.0-rtm-26502-02 that the following code

private static readonly byte[] inputBytes = new byte[] {1, 2, 3, 4};
private static string inputString = "Hello World";

[Benchmark]
public ReadOnlySpan<byte> CodeQualityBytes()
{
    return inputBytes.AsSpan();
}

[Benchmark]
public ReadOnlySpan<char> CodeQualityString()
{
    return inputString.AsSpan();
}

has a tiny code quality issue for the CodeQualityString, it has a couple of unused
mov rdx,rcx and mov rcx,rdx at the end.
(Those do not appear to be Benchmark.Net artifacts, they show up with longer methods too).

I think I've seen this before, but couldn't find anything, if this is a duplicate, please close.

The output of Benchmark.Net 0.10.14 for CodeQualityString is:

00007ffb`c5580f90 SpanJson.Benchmarks.SelectedBenchmarks.CodeQualityString()
IL_0000: ldsfld System.String SpanJson.Benchmarks.SelectedBenchmarks::inputString
IL_0005: call System.ReadOnlySpan`1<System.Char> System.MemoryExtensions::AsSpan(System.String)
IL_000a: ret
00007ffb`c5580f98 48b9b0fd62c5fb7f0000 mov rcx,7FFBC562FDB0h
00007ffb`c5580fa2 ba08000000      mov     edx,8
00007ffb`c5580fa7 e844fcac5f      call    coreclr!coreclr_shutdown_2+0x39680 (00007ffc`25050bf0)
00007ffb`c5580fac 48b8a89f7629c3010000 mov rax,1C329769FA8h
00007ffb`c5580fb6 488b00          mov     rax,qword ptr [rax]
00007ffb`c5580fb9 4885c0          test    rax,rax
00007ffb`c5580fbc 7506            jne     00007ffb`c5580fc4
00007ffb`c5580fbe 33d2            xor     edx,edx
00007ffb`c5580fc0 33c9            xor     ecx,ecx
00007ffb`c5580fc2 eb15            jmp     00007ffb`c5580fd9
00007ffb`c5580fc4 488d500c        lea     rdx,[rax+0Ch]
00007ffb`c5580fc8 488bca          mov     rcx,rdx
00007ffb`c5580fcb 8b4008          mov     eax,dword ptr [rax+8]
00007ffb`c5580fce 488bd1          mov     rdx,rcx
00007ffb`c5580fd1 488bca          mov     rcx,rdx
00007ffb`c5580fd4 488bd1          mov     rdx,rcx
00007ffb`c5580fd7 8bc8            mov     ecx,eax

The output of Benchmark.NET 0.10.14 for CodeQualityBytes is:

00007ffb`c5590f90 SpanJson.Benchmarks.SelectedBenchmarks.CodeQualityBytes()
IL_0000: ldsfld System.Byte[] SpanJson.Benchmarks.SelectedBenchmarks::inputBytes
IL_0005: call System.Span`1<!!0> System.MemoryExtensions::AsSpan<System.Byte>(!!0[])
IL_000a: call System.ReadOnlySpan`1<!0> System.Span`1<System.Byte>::op_Implicit(System.Span`1<!0>)
IL_000f: ret
00007ffb`c5590f98 48b9b0fd63c5fb7f0000 mov rcx,7FFBC563FDB0h
00007ffb`c5590fa2 ba08000000      mov     edx,8
00007ffb`c5590fa7 e844fcab5f      call    coreclr!coreclr_shutdown_2+0x39680 (00007ffc`25050bf0)
00007ffb`c5590fac 48b8a09f0bd6a4010000 mov rax,1A4D60B9FA0h
00007ffb`c5590fb6 488b00          mov     rax,qword ptr [rax]
00007ffb`c5590fb9 4885c0          test    rax,rax
00007ffb`c5590fbc 7506            jne     00007ffb`c5590fc4
00007ffb`c5590fbe 33d2            xor     edx,edx
00007ffb`c5590fc0 33c9            xor     ecx,ecx
00007ffb`c5590fc2 eb07            jmp     00007ffb`c5590fcb
00007ffb`c5590fc4 488d5010        lea     rdx,[rax+10h]
00007ffb`c5590fc8 8b4808          mov     ecx,dword ptr [rax+8]

No ILOffsetMap found
System.MemoryExtensions.AsSpan(!!0[])

category:cq
theme:register-allocator
skill-level:expert
cost:medium

[AndyAyersMS]

Looks like the same issue noted at the end of the comment stream for #7279.

[AndyAyersMS]

We're currently building up an inventory of all the open codegen issues and it seems like this problem of register shuffling likely deserves an anchor issue of its own, as it has appeared in several places.

So I'm going to leave this issue open for now. When we get further along in our issue review (hopefully just later this week), we will revisit.

[mikedn]

Seems similar to #9519 as well.

[AndyAyersMS]

And similar to issues in #9165.

[CarolEidt]

This should be analyzed to determine if it's still happening.

[kunalspathak]

I do not see the pattern in the issue anymore on Windows/x64.

OverheadJitting  1: 1 op, 5186100.00 ns, 5.1861 ms/op
; Assembly listing for method MiniBench._10296:CodeQualityString():System.ReadOnlySpan`1[Char]:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Windows
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; Final local variable assignments
;
;* V00 this         [V00    ] (  0,  0   )     ref  ->  zero-ref    this class-hnd
;  V01 RetBuf       [V01,T00] (  5,  5   )   byref  ->  rsi
;  V02 OutArgs      [V02    ] (  1,  1   )  lclBlk (32) [rsp+00H]   "OutgoingArgSpace"
;* V03 tmp1         [V03    ] (  0,  0   )  struct (16) zero-ref    "Inline return value spill temp"
;  V04 tmp2         [V04,T01] (  4,  6   )     ref  ->  rcx         class-hnd "Inlining Arg"
;* V05 tmp3         [V05    ] (  0,  0   )  struct (16) zero-ref    "NewObj constructor temp"
;* V06 tmp4         [V06    ] (  0,  0   )  struct (16) zero-ref    ld-addr-op "Inline ldloca(s) first use temp"
;  V07 tmp5         [V07,T02] (  3,  3   )     int  ->  rbx         "Inlining Arg"
;  V08 tmp6         [V08,T06] (  2,  2   )   byref  ->  rdi         "Inlining Arg"
;* V09 tmp7         [V09    ] (  0,  0   )  struct ( 8) zero-ref    "NewObj constructor temp"
;  V10 tmp8         [V10,T07] (  2,  2   )    bool  ->  rcx         "Inlining Arg"
;* V11 tmp9         [V11,T13] (  0,  0   )     ref  ->  zero-ref    class-hnd "Inlining Arg"
;  V12 tmp10        [V12,T03] (  3,  2   )     ref  ->  rdx         class-hnd "Inlining Arg"
;  V13 tmp11        [V13,T04] (  3,  2   )   byref  ->  rax         V03._pointer(offs=0x00) P-INDEP "field V03._pointer (fldOffset=0x0)"
;  V14 tmp12        [V14,T05] (  3,  2   )     int  ->  rdi         V03._length(offs=0x08) P-INDEP "field V03._length (fldOffset=0x8)"
;  V15 tmp13        [V15,T09] (  2,  1   )   byref  ->  rax         V05._pointer(offs=0x00) P-INDEP "field V05._pointer (fldOffset=0x0)"
;  V16 tmp14        [V16,T12] (  2,  1   )     int  ->  rdi         V05._length(offs=0x08) P-INDEP "field V05._length (fldOffset=0x8)"
;  V17 tmp15        [V17,T10] (  2,  1   )   byref  ->  rax         V06._pointer(offs=0x00) P-INDEP "field V06._pointer (fldOffset=0x0)"
;* V18 tmp16        [V18,T14] (  0,  0   )     int  ->  zero-ref    V06._length(offs=0x08) P-INDEP "field V06._length (fldOffset=0x8)"
;  V19 tmp17        [V19,T11] (  2,  1   )   byref  ->  rdi         V09._value(offs=0x00) P-INDEP "field V09._value (fldOffset=0x0)"
;  V20 cse0         [V20,T08] (  2,  1   )     ref  ->  rdx         "CSE - moderate"
;
; Lcl frame size = 32

G_M38128_IG01:              ;; offset=0000H
       57                   push     rdi
       56                   push     rsi
       53                   push     rbx
       4883EC20             sub      rsp, 32
       488BF2               mov      rsi, rdx
                                                ;; bbWeight=1    PerfScore 3.50
G_M38128_IG02:              ;; offset=000AH
       48B990D15873F87F0000 mov      rcx, 0x7FF87358D190
       BA21000000           mov      edx, 33
       E8028A395F           call     CORINFO_HELP_GETSHARED_NONGCSTATIC_BASE
       48B9D88B36B500020000 mov      rcx, 0x200B5368BD8
       488B09               mov      rcx, gword ptr [rcx]
       4885C9               test     rcx, rcx
       7506                 jne      SHORT G_M38128_IG04
                                                ;; bbWeight=1    PerfScore 5.00
G_M38128_IG03:              ;; offset=0030H
       33C0                 xor      rax, rax
       33FF                 xor      edi, edi
       EB2C                 jmp      SHORT G_M38128_IG07
                                                ;; bbWeight=0.50 PerfScore 1.25
G_M38128_IG04:              ;; offset=0036H
       488D790C             lea      rdi, bword ptr [rcx+12]
       8B5908               mov      ebx, dword ptr [rcx+8]
       8BCB                 mov      ecx, ebx
       F7D1                 not      ecx
       C1E91F               shr      ecx, 31
       48BA203036B500020000 mov      rdx, 0x200B5363020
       488B12               mov      rdx, gword ptr [rdx]
       85C9                 test     ecx, ecx
       7508                 jne      SHORT G_M38128_IG06
                                                ;; bbWeight=0.50 PerfScore 4.25
G_M38128_IG05:              ;; offset=0055H
       488BCA               mov      rcx, rdx
       E8C3D990FF           call     System.Diagnostics.Debug:Fail(System.String,System.String)
                                                ;; bbWeight=0.25 PerfScore 0.31
G_M38128_IG06:              ;; offset=005DH
       488BC7               mov      rax, rdi
       8BFB                 mov      edi, ebx
                                                ;; bbWeight=0.50 PerfScore 0.25
G_M38128_IG07:              ;; offset=0062H
       488906               mov      bword ptr [rsi], rax
       897E08               mov      dword ptr [rsi+8], edi
       488BC6               mov      rax, rsi
                                                ;; bbWeight=1    PerfScore 2.25
G_M38128_IG08:              ;; offset=006BH
       4883C420             add      rsp, 32
       5B                   pop      rbx
       5E                   pop      rsi
       5F                   pop      rdi
       C3                   ret
                                                ;; bbWeight=1    PerfScore 2.75

; Total bytes of code 115, prolog size 7, PerfScore 31.06, instruction count 36, allocated bytes for code 115 (MethodHash=544d6b0f) for method MiniBench._10296:CodeQualityString():System.ReadOnlySpan`1[Char]:this
; ============================================================

WorkloadJitting  1: 1 op, 13667200.00 ns, 13.6672 ms/op