Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

JIT does not always fold inlined constants #62604

Open
jasper-d opened this issue Dec 9, 2021 · 2 comments
Open

JIT does not always fold inlined constants #62604

jasper-d opened this issue Dec 9, 2021 · 2 comments
Labels
area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Milestone
Future

Comments

@jasper-d
Copy link
Contributor

jasper-d commented Dec 9, 2021 •
edited by BruceForstall
Loading

It appears that JIT (x64) sometimes misses opportunities for folding inlined(?) constants.
The following code demonstrates the issue. Depending on the order of the addends in C#, constants are more or less folded.
I did assume that order in C# would not impact JITs ability to fold them.

There are two pairs of methods, CalculateMaxSize[No]ExtraAdd and CalculateMaxSize[No]ExtraInc.
The *No* variants look like what I had expected for either implementation.

using System.Runtime.CompilerServices;
internal readonly struct SubId {
    public static int MaxSize => 16;
}

internal readonly struct MyStruct
{
    public static ReadOnlySpan<byte> EoL => new []{(byte)'\r', (byte)'\n'};
    internal readonly byte[] _subject;

    public MyStruct(byte[] subject){
        _subject = subject;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 18
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeNoExtraAdd()
    {
        return SubId.MaxSize + EoL.Length + _subject.Length;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 16
    add      eax, 2
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeExtraAdd()
    {
        // This will emit two add instructions even when making SubId.MaxSize a const field
        return _subject.Length + SubId.MaxSize + EoL.Length;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 17
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeNoExtraInc()
    {
        return _subject.Length + SubId.MaxSize + 1;
    }


    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    inc      eax                     // This does not repro on .NET 7.0, it now emits a single add eax, 17 ❤
    add      eax, 16
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeExtraInc()
    {
        return _subject.Length + 1 + SubId.MaxSize;
    }
}

Sharplap and JIT dump running 1d4b5f6

; Assembly listing for method foo.MyStruct:CalculateMaxSizeNoExtraAdd():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 1 inlinees with PGO data; 5 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;* V01 loc0         [V01    ] (  0,  0   )  struct (16) zero-ref    ld-addr-op
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;* V03 tmp1         [V03    ] (  0,  0   )     int  ->  zero-ref    "impAppendStmt"
;* V04 tmp2         [V04    ] (  0,  0   )  struct (16) zero-ref    "NewObj constructor temp"
;* V05 tmp3         [V05    ] (  0,  0   )  struct ( 8) zero-ref    "NewObj constructor temp"
;* V06 tmp4         [V06    ] (  0,  0   )   byref  ->  zero-ref    single-def V01._pointer(offs=0x00) P-INDEP "field V01._pointer (fldOffset=0x0)"
;* V07 tmp5         [V07    ] (  0,  0   )     int  ->  zero-ref    single-def V01._length(offs=0x08) P-INDEP "field V01._length (fldOffset=0x8)"
;* V08 tmp6         [V08    ] (  0,  0   )   byref  ->  zero-ref    V04._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"
;* V09 tmp7         [V09    ] (  0,  0   )     int  ->  zero-ref    V04._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)"
;* V10 tmp8         [V10    ] (  0,  0   )   byref  ->  zero-ref    single-def V05._value(offs=0x00) P-INDEP "field V05._value (fldOffset=0x0)"
;
; Lcl frame size = 0

G_M57648_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M57648_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 18
                                                ;; bbWeight=1    PerfScore 4.25
G_M57648_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 10, prolog size 0, PerfScore 6.25, instruction count 4, allocated bytes for code 10 (MethodHash=f2321ecf) for method foo.MyStruct:CalculateMaxSizeNoExtraAdd():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeExtraAdd():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 1 inlinees with PGO data; 5 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;* V01 loc0         [V01    ] (  0,  0   )  struct (16) zero-ref    ld-addr-op
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V03 tmp1         [V03,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;  V04 tmp2         [V04,T02] (  2,  4   )     int  ->  rax         "impAppendStmt"
;* V05 tmp3         [V05    ] (  0,  0   )  struct (16) zero-ref    "NewObj constructor temp"
;* V06 tmp4         [V06    ] (  0,  0   )  struct ( 8) zero-ref    "NewObj constructor temp"
;* V07 tmp5         [V07    ] (  0,  0   )   byref  ->  zero-ref    single-def V01._pointer(offs=0x00) P-INDEP "field V01._pointer (fldOffset=0x0)"
;* V08 tmp6         [V08    ] (  0,  0   )     int  ->  zero-ref    single-def V01._length(offs=0x08) P-INDEP "field V01._length (fldOffset=0x8)"
;* V09 tmp7         [V09    ] (  0,  0   )   byref  ->  zero-ref    V05._pointer(offs=0x00) P-INDEP "field V05._pointer (fldOffset=0x0)"
;* V10 tmp8         [V10    ] (  0,  0   )     int  ->  zero-ref    V05._length(offs=0x08) P-INDEP "field V05._length (fldOffset=0x8)"
;* V11 tmp9         [V11    ] (  0,  0   )   byref  ->  zero-ref    single-def V06._value(offs=0x00) P-INDEP "field V06._value (fldOffset=0x0)"
;
; Lcl frame size = 0

G_M14833_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M14833_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 16
       add      eax, 2
                                                ;; bbWeight=1    PerfScore 4.50
G_M14833_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 13, prolog size 0, PerfScore 6.80, instruction count 5, allocated bytes for code 13 (MethodHash=cc45c60e) for method foo.MyStruct:CalculateMaxSizeExtraAdd():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeNoExtraInc():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V02 tmp1         [V02,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;
; Lcl frame size = 0

G_M11669_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M11669_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 17
                                                ;; bbWeight=1    PerfScore 4.25
G_M11669_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 10, prolog size 0, PerfScore 6.25, instruction count 4, allocated bytes for code 10 (MethodHash=d8a9d26a) for method foo.MyStruct:CalculateMaxSizeNoExtraInc():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeExtraInc():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V02 tmp1         [V02,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;
; Lcl frame size = 0

G_M56724_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M56724_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       inc      eax
       add      eax, 16
                                                ;; bbWeight=1    PerfScore 4.50
G_M56724_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 12, prolog size 0, PerfScore 6.70, instruction count 5, allocated bytes for code 12 (MethodHash=17e2226b) for method foo.MyStruct:CalculateMaxSizeExtraInc():int:this
; ============================================================

category:implementation
theme:inlining
skill-level:intermediate
cost:medium
impact:medium
@dotnet-issue-labeler dotnet-issue-labeler bot added area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI untriaged New issue has not been triaged by the area owner labels Dec 9, 2021
@ghost
Copy link

ghost commented Dec 9, 2021

Tagging subscribers to this area: @JulieLeeMSFT
See info in area-owners.md if you want to be subscribed.

Issue Details

It appears that JIT (x64) sometimes misses opportunities for folding inlined(?) constants.
The following code demonstrates the issue. Depending on the order of the addends in C#, constants are more or less folded.
I did assume that order in C# would not impact JITs ability to fold them.

There are two pairs of methods, CalculateMaxSize[No]ExtraAdd and CalculateMaxSize[No]ExtraInc.
The *No* variants look like what I had expected for either implementation.

using System.Runtime.CompilerServices;
internal readonly struct SubId {
    public static int MaxSize => 16;
}

internal readonly struct MyStruct
{
    public static ReadOnlySpan<byte> EoL => new []{(byte)'\r', (byte)'\n'};
    internal readonly byte[] _subject;

    public MyStruct(byte[] subject){
        _subject = subject;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 18
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeNoExtraAdd()
    {
        return SubId.MaxSize + EoL.Length + _subject.Length;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 16
    add      eax, 2
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeExtraAdd()
    {
        // This will emit two add instructions even when making SubId.MaxSize a const field
        return _subject.Length + SubId.MaxSize + EoL.Length;
    }

    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    add      eax, 17
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeNoExtraInc()
    {
        return _subject.Length + SubId.MaxSize + 1;
    }


    /*
    mov      rax, gword ptr [rdi]
    mov      eax, dword ptr [rax+8]
    inc      eax
    add      eax, 16
    */
    [MethodImpl(MethodImplOptions.NoInlining)]
    public int CalculateMaxSizeExtraInc()
    {
        return _subject.Length + 1 + SubId.MaxSize;
    }
}

Sharplap and JIT dump running 1d4b5f6

; Assembly listing for method foo.MyStruct:CalculateMaxSizeNoExtraAdd():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 1 inlinees with PGO data; 5 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;* V01 loc0         [V01    ] (  0,  0   )  struct (16) zero-ref    ld-addr-op
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;* V03 tmp1         [V03    ] (  0,  0   )     int  ->  zero-ref    "impAppendStmt"
;* V04 tmp2         [V04    ] (  0,  0   )  struct (16) zero-ref    "NewObj constructor temp"
;* V05 tmp3         [V05    ] (  0,  0   )  struct ( 8) zero-ref    "NewObj constructor temp"
;* V06 tmp4         [V06    ] (  0,  0   )   byref  ->  zero-ref    single-def V01._pointer(offs=0x00) P-INDEP "field V01._pointer (fldOffset=0x0)"
;* V07 tmp5         [V07    ] (  0,  0   )     int  ->  zero-ref    single-def V01._length(offs=0x08) P-INDEP "field V01._length (fldOffset=0x8)"
;* V08 tmp6         [V08    ] (  0,  0   )   byref  ->  zero-ref    V04._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"
;* V09 tmp7         [V09    ] (  0,  0   )     int  ->  zero-ref    V04._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)"
;* V10 tmp8         [V10    ] (  0,  0   )   byref  ->  zero-ref    single-def V05._value(offs=0x00) P-INDEP "field V05._value (fldOffset=0x0)"
;
; Lcl frame size = 0

G_M57648_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M57648_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 18
                                                ;; bbWeight=1    PerfScore 4.25
G_M57648_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 10, prolog size 0, PerfScore 6.25, instruction count 4, allocated bytes for code 10 (MethodHash=f2321ecf) for method foo.MyStruct:CalculateMaxSizeNoExtraAdd():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeExtraAdd():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 1 inlinees with PGO data; 5 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;* V01 loc0         [V01    ] (  0,  0   )  struct (16) zero-ref    ld-addr-op
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V03 tmp1         [V03,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;  V04 tmp2         [V04,T02] (  2,  4   )     int  ->  rax         "impAppendStmt"
;* V05 tmp3         [V05    ] (  0,  0   )  struct (16) zero-ref    "NewObj constructor temp"
;* V06 tmp4         [V06    ] (  0,  0   )  struct ( 8) zero-ref    "NewObj constructor temp"
;* V07 tmp5         [V07    ] (  0,  0   )   byref  ->  zero-ref    single-def V01._pointer(offs=0x00) P-INDEP "field V01._pointer (fldOffset=0x0)"
;* V08 tmp6         [V08    ] (  0,  0   )     int  ->  zero-ref    single-def V01._length(offs=0x08) P-INDEP "field V01._length (fldOffset=0x8)"
;* V09 tmp7         [V09    ] (  0,  0   )   byref  ->  zero-ref    V05._pointer(offs=0x00) P-INDEP "field V05._pointer (fldOffset=0x0)"
;* V10 tmp8         [V10    ] (  0,  0   )     int  ->  zero-ref    V05._length(offs=0x08) P-INDEP "field V05._length (fldOffset=0x8)"
;* V11 tmp9         [V11    ] (  0,  0   )   byref  ->  zero-ref    single-def V06._value(offs=0x00) P-INDEP "field V06._value (fldOffset=0x0)"
;
; Lcl frame size = 0

G_M14833_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M14833_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 16
       add      eax, 2
                                                ;; bbWeight=1    PerfScore 4.50
G_M14833_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 13, prolog size 0, PerfScore 6.80, instruction count 5, allocated bytes for code 13 (MethodHash=cc45c60e) for method foo.MyStruct:CalculateMaxSizeExtraAdd():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeNoExtraInc():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V02 tmp1         [V02,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;
; Lcl frame size = 0

G_M11669_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M11669_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       add      eax, 17
                                                ;; bbWeight=1    PerfScore 4.25
G_M11669_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 10, prolog size 0, PerfScore 6.25, instruction count 4, allocated bytes for code 10 (MethodHash=d8a9d26a) for method foo.MyStruct:CalculateMaxSizeNoExtraInc():int:this
; ============================================================

; Assembly listing for method foo.MyStruct:CalculateMaxSizeExtraInc():int:this
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
; Tier-1 compilation
; optimized code
; rsp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
; Final local variable assignments
;
;  V00 this         [V00,T00] (  3,  3   )   byref  ->  rdi         this single-def
;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
;  V02 tmp1         [V02,T01] (  2,  4   )     int  ->  rax         "impAppendStmt"
;
; Lcl frame size = 0

G_M56724_IG01:
                                                ;; bbWeight=1    PerfScore 0.00
G_M56724_IG02:
       mov      rax, gword ptr [rdi]
       mov      eax, dword ptr [rax+8]
       inc      eax
       add      eax, 16
                                                ;; bbWeight=1    PerfScore 4.50
G_M56724_IG03:
       ret
                                                ;; bbWeight=1    PerfScore 1.00

; Total bytes of code 12, prolog size 0, PerfScore 6.70, instruction count 5, allocated bytes for code 12 (MethodHash=17e2226b) for method foo.MyStruct:CalculateMaxSizeExtraInc():int:this
; ============================================================
Author: jasper-d
Assignees: -
Labels:

area-CodeGen-coreclr, untriaged
Milestone: -

@AndyAyersMS
Copy link
Member

More cases that would benefit from forward substitution.

See #6973.

@AndyAyersMS AndyAyersMS mentioned this issue Jan 13, 2022
Merged
@JulieLeeMSFT JulieLeeMSFT added this to the Future milestone Jan 20, 2022
@JulieLeeMSFT JulieLeeMSFT removed the untriaged New issue has not been triaged by the area owner label Jan 20, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Projects
None yet
Milestone
Future
Development

No branches or pull requests
3 participants
@jasper-d @AndyAyersMS @JulieLeeMSFT