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.

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

JIT should recognize a popular type check pattern #11396

Closed
omariom opened this issue Nov 3, 2018 · 3 comments
Closed

JIT should recognize a popular type check pattern #11396

omariom opened this issue Nov 3, 2018 · 3 comments

Comments

@omariom
Copy link
Contributor

omariom commented Nov 3, 2018

I have found the trick in @GrabYourPitchforks dotnet/coreclr#20386
JIT could recognize the pattern and generate optimized code for sealed types.

        [MethodImpl(MethodImplOptions.NoInlining)]
        public static int Is(object obj)
        {
            if (obj is string str)
                return str.Length;

            return 7;
        }

        [MethodImpl(MethodImplOptions.NoInlining)]
        public static int As(object obj)
        {
            string str = obj as string;

            if (str != null)
                return str.Length;

            return 7;
        }

        [MethodImpl(MethodImplOptions.NoInlining)]
        public static int SafeUnsafe(object obj)
        {
            if (obj != null && obj.GetType() == typeof(string))
                return Unsafe.As<string>(obj).Length;

            return 7;
        }
See the generated code

; Assembly listing for method Program:Is(ref):int
; Emitting BLENDED_CODE for X64 CPU with AVX
; optimized code
; rsp based frame
; partially interruptible
; Final local variable assignments
;
;  V00 arg0         [V00,T01] (  3,  3   )     ref  ->  rcx         class-hnd
;  V01 loc0         [V01,T02] (  3,  2.50)     ref  ->  rax         class-hnd
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+0x00]  
;  V03 tmp1         [V03,T00] (  5,  6.74)     ref  ->  rax         class-hnd
;
; Lcl frame size = 0

G_M12761_IG01:
       0F1F440000           nop      

G_M12761_IG02:
       488BC1               mov      rax, rcx
       4885C0               test     rax, rax
       7411                 je       SHORT G_M12761_IG03
       48BA00F50A96FE7F0000 mov      rdx, 0x7FFE960AF500
       483910               cmp      qword ptr [rax], rdx
       7402                 je       SHORT G_M12761_IG03
       33C0                 xor      rax, rax

G_M12761_IG03:
       4885C0               test     rax, rax
       7404                 je       SHORT G_M12761_IG05
       8B4008               mov      eax, dword ptr [rax+8]

G_M12761_IG04:
       C3                   ret      

G_M12761_IG05:
       B807000000           mov      eax, 7

G_M12761_IG06:
       C3                   ret      

; Total bytes of code 45, prolog size 5 for method Program:Is(ref):int
; ============================================================


; Assembly listing for method Program:As(ref):int
; Emitting BLENDED_CODE for X64 CPU with AVX
; optimized code
; rsp based frame
; partially interruptible
; Final local variable assignments
;
;  V00 arg0         [V00,T01] (  3,  3   )     ref  ->  rcx         class-hnd
;  V01 loc0         [V01,T02] (  3,  2.50)     ref  ->  rax         class-hnd
;# V02 OutArgs      [V02    ] (  1,  1   )  lclBlk ( 0) [rsp+0x00]  
;  V03 tmp1         [V03,T00] (  5,  6.74)     ref  ->  rax         class-hnd
;
; Lcl frame size = 0

G_M12497_IG01:
       0F1F440000           nop      

G_M12497_IG02:
       488BC1               mov      rax, rcx
       4885C0               test     rax, rax
       7411                 je       SHORT G_M12497_IG03
       48BA00F50A96FE7F0000 mov      rdx, 0x7FFE960AF500
       483910               cmp      qword ptr [rax], rdx
       7402                 je       SHORT G_M12497_IG03
       33C0                 xor      rax, rax

G_M12497_IG03:
       4885C0               test     rax, rax
       7404                 je       SHORT G_M12497_IG05
       8B4008               mov      eax, dword ptr [rax+8]

G_M12497_IG04:
       C3                   ret      

G_M12497_IG05:
       B807000000           mov      eax, 7

G_M12497_IG06:
       C3                   ret      

; Total bytes of code 45, prolog size 5 for method Program:As(ref):int
; ============================================================


; Assembly listing for method Program:SafeUnsafe(ref):int
; Emitting BLENDED_CODE for X64 CPU with AVX
; optimized code
; rsp based frame
; partially interruptible
; Final local variable assignments
;
;  V00 arg0         [V00,T00] (  5,  4   )     ref  ->  rcx         class-hnd
;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+0x00]  
;
; Lcl frame size = 0

G_M6777_IG01:
       0F1F440000           nop      

G_M6777_IG02:
       4885C9               test     rcx, rcx
       7413                 je       SHORT G_M6777_IG04
       48B800F50A96FE7F0000 mov      rax, 0x7FFE960AF500
       483901               cmp      qword ptr [rcx], rax
       7504                 jne      SHORT G_M6777_IG04
       8B4108               mov      eax, dword ptr [rcx+8]

G_M6777_IG03:
       C3                   ret      

G_M6777_IG04:
       B807000000           mov      eax, 7

G_M6777_IG05:
       C3                   ret      

; Total bytes of code 35, prolog size 5 for method Program:SafeUnsafe(ref):int
; ============================================================

And to save rax it would be nice to have

cmp      qword ptr [rcx], 0x7FFE960AF500

instead of

mov      rax, 0x7FFE960AF500
cmp      qword ptr [rcx], rax
@AndyAyersMS
Copy link
Member

This is a known area for improvement (see #9117).

Most x64 instructions can't handle large literal constant operands, which is why you see the mov; cmp like you do above.

@omariom
Copy link
Contributor Author

omariom commented Nov 3, 2018

The week of dups :)
Closing.

@omariom omariom closed this as completed Nov 3, 2018
@AndyAyersMS
Copy link
Member

It's useful to have (and to be reminded of) these kinds of examples.

Please don't hold back reporting things because you're worried about reporting something already known. Sometimes similar looking things aren't all that similar in the end.

@msftgits msftgits transferred this issue from dotnet/coreclr Jan 31, 2020
@dotnet dotnet locked as resolved and limited conversation to collaborators Dec 15, 2020
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants