[NativeAOT] Accessing static variables, ideally, should not call runtime helpers.

[VSadov]

It is a bit surprising that in a completely static scenario, statically compiled code calls helpers like __GetGCStaticBase/__GetNonGCStaticBase.

I am porting some c++ code to c#/NativeAOT and I see impact on performance from this in a few tight spots.
Can the emit for statics be closer to the native compilers?

Repro:

using System.Runtime.CompilerServices;

internal class Program
{
    static int something1;
    static string something2;

    private static void Main(string[] args)
    {
        Test();
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
        something2 = "qwer";
        Console.WriteLine(something1 + something2);

        System.Diagnostics.Debugger.Break();
    }
}

After compiling for nativeAOT/release, I see in debugger:

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
00007FF6ABFE1CA0  push        rdi  
00007FF6ABFE1CA1  push        rsi  
00007FF6ABFE1CA2  sub         rsp,28h  
00007FF6ABFE1CA6  call        __GetNonGCStaticBase_ConsoleApp11_Program (07FF6ABED12F6h)    <===== ????
00007FF6ABFE1CAB  mov         rsi,rax  
00007FF6ABFE1CAE  mov         dword ptr [rsi],4D2h  
        something2 = "qwer";
00007FF6ABFE1CB4  call        __GetGCStaticBase_ConsoleApp11_Program (07FF6ABED1E0Dh)      <======????
00007FF6ABFE1CB9  mov         rdi,rax  
00007FF6ABFE1CBC  lea         rcx,[__Str_qwer (07FF6AC17D1B8h)]  
00007FF6ABFE1CC3  mov         qword ptr [rdi+8],rcx  
        Console.WriteLine(something1 + something2);
00007FF6ABFE1CC7  mov         ecx,dword ptr [rsi]  
00007FF6ABFE1CC9  call        S_P_CoreLib_System_Number__Int32ToDecStr (07FF6ABF62BD0h)  
00007FF6ABFE1CCE  mov         rcx,rax  
00007FF6ABFE1CD1  mov         rdx,qword ptr [rdi+8]  
00007FF6ABFE1CD5  call        String__Concat_5 (07FF6ABF4D150h)  
00007FF6ABFE1CDA  mov         rcx,rax  
00007FF6ABFE1CDD  call        System_Console_System_Console__WriteLine_12 (07FF6AC00ED00h)  

        System.Diagnostics.Debugger.Break();
00007FF6ABFE1CE2  call        S_P_CoreLib_System_Diagnostics_Debugger__Break (07FF6ABFA6D30h)  
    }
00007FF6ABFE1CE7  nop  
00007FF6ABFE1CE8  add         rsp,28h  
00007FF6ABFE1CEC  pop         rsi  
00007FF6ABFE1CED  pop         rdi  
00007FF6ABFE1CEE  ret  

[ghost]

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

Issue Details

---

It is a bit surprising that in a completely static scenario on a completely statically compiled runtime code calls helpers like __GetGCStaticBase/__GetNonGCStaticBase.

I am porting some c++ code and see impact on performance from this in a few tight spots.
Can the emit for statics be closer to the native compilers?

Repro:

using System.Runtime.CompilerServices;

internal class Program
{
    static int something1;
    static string something2;

    private static void Main(string[] args)
    {
        Test();
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
        something2 = "qwer";
        Console.WriteLine(something1 + something2);

        System.Diagnostics.Debugger.Break();
    }
}

After compiling for nativeAOT/release, I see in debugger:

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
00007FF6ABFE1CA0  push        rdi  
00007FF6ABFE1CA1  push        rsi  
00007FF6ABFE1CA2  sub         rsp,28h  
00007FF6ABFE1CA6  call        __GetNonGCStaticBase_ConsoleApp11_Program (07FF6ABED12F6h)    <===== ????
00007FF6ABFE1CAB  mov         rsi,rax  
00007FF6ABFE1CAE  mov         dword ptr [rsi],4D2h  
        something2 = "qwer";
00007FF6ABFE1CB4  call        __GetGCStaticBase_ConsoleApp11_Program (07FF6ABED1E0Dh)      <======????
00007FF6ABFE1CB9  mov         rdi,rax  
00007FF6ABFE1CBC  lea         rcx,[__Str_qwer (07FF6AC17D1B8h)]  
00007FF6ABFE1CC3  mov         qword ptr [rdi+8],rcx  
        Console.WriteLine(something1 + something2);
00007FF6ABFE1CC7  mov         ecx,dword ptr [rsi]  
00007FF6ABFE1CC9  call        S_P_CoreLib_System_Number__Int32ToDecStr (07FF6ABF62BD0h)  
00007FF6ABFE1CCE  mov         rcx,rax  
00007FF6ABFE1CD1  mov         rdx,qword ptr [rdi+8]  
00007FF6ABFE1CD5  call        String__Concat_5 (07FF6ABF4D150h)  
00007FF6ABFE1CDA  mov         rcx,rax  
00007FF6ABFE1CDD  call        System_Console_System_Console__WriteLine_12 (07FF6AC00ED00h)  

        System.Diagnostics.Debugger.Break();
00007FF6ABFE1CE2  call        S_P_CoreLib_System_Diagnostics_Debugger__Break (07FF6ABFA6D30h)  
    }
00007FF6ABFE1CE7  nop  
00007FF6ABFE1CE8  add         rsp,28h  
00007FF6ABFE1CEC  pop         rsi  
00007FF6ABFE1CED  pop         rdi  
00007FF6ABFE1CEE  ret  

Author:	VSadov
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[ghost]

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

It is a bit surprising that in a completely static scenario on a completely statically compiled runtime code calls helpers like __GetGCStaticBase/__GetNonGCStaticBase.

I am porting some c++ code and see impact on performance from this in a few tight spots.
Can the emit for statics be closer to the native compilers?

Repro:

using System.Runtime.CompilerServices;

internal class Program
{
    static int something1;
    static string something2;

    private static void Main(string[] args)
    {
        Test();
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
        something2 = "qwer";
        Console.WriteLine(something1 + something2);

        System.Diagnostics.Debugger.Break();
    }
}

After compiling for nativeAOT/release, I see in debugger:

    [MethodImpl(MethodImplOptions.NoInlining)]
    private static void Test()
    {
        something1 = 1234;
00007FF6ABFE1CA0  push        rdi  
00007FF6ABFE1CA1  push        rsi  
00007FF6ABFE1CA2  sub         rsp,28h  
00007FF6ABFE1CA6  call        __GetNonGCStaticBase_ConsoleApp11_Program (07FF6ABED12F6h)    <===== ????
00007FF6ABFE1CAB  mov         rsi,rax  
00007FF6ABFE1CAE  mov         dword ptr [rsi],4D2h  
        something2 = "qwer";
00007FF6ABFE1CB4  call        __GetGCStaticBase_ConsoleApp11_Program (07FF6ABED1E0Dh)      <======????
00007FF6ABFE1CB9  mov         rdi,rax  
00007FF6ABFE1CBC  lea         rcx,[__Str_qwer (07FF6AC17D1B8h)]  
00007FF6ABFE1CC3  mov         qword ptr [rdi+8],rcx  
        Console.WriteLine(something1 + something2);
00007FF6ABFE1CC7  mov         ecx,dword ptr [rsi]  
00007FF6ABFE1CC9  call        S_P_CoreLib_System_Number__Int32ToDecStr (07FF6ABF62BD0h)  
00007FF6ABFE1CCE  mov         rcx,rax  
00007FF6ABFE1CD1  mov         rdx,qword ptr [rdi+8]  
00007FF6ABFE1CD5  call        String__Concat_5 (07FF6ABF4D150h)  
00007FF6ABFE1CDA  mov         rcx,rax  
00007FF6ABFE1CDD  call        System_Console_System_Console__WriteLine_12 (07FF6AC00ED00h)  

        System.Diagnostics.Debugger.Break();
00007FF6ABFE1CE2  call        S_P_CoreLib_System_Diagnostics_Debugger__Break (07FF6ABFA6D30h)  
    }
00007FF6ABFE1CE7  nop  
00007FF6ABFE1CE8  add         rsp,28h  
00007FF6ABFE1CEC  pop         rsi  
00007FF6ABFE1CED  pop         rdi  
00007FF6ABFE1CEE  ret  

Author:	VSadov
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged, area-NativeAOT-coreclr
Milestone:	-

[VSadov]

Perhaps there are reasons for dynamic indirection, but it feels like - if compiler knows what helpers to call and compiler has generated those helpers, maybe it could just inline the results instead?

[MichalStrehovsky]

Perhaps there are reasons for dynamic indirection, but it feels like - if compiler knows what functions to call and compiler has generated those helpers, maybe it could just inline the results instead?

No reason - just a current codegen limitation - we cannot express the access pattern though JitInterface. This is one of the sub-bullets in #64242. Cc @EgorBo this is in your area of interests.

[EgorBo]

Closing as completed. Current codegen for Test()

; Method Program:Test()
G_M24707_IG01:              ;; offset=0000H
       56                   push     rsi
       4883EC20             sub      rsp, 32
G_M24707_IG02:              ;; offset=0005H
       C705FCFFFFFFD2040000 mov      dword ptr [(reloc 0x4000000000423508)], 0x4D2      ; static handle
       488B3500000000       mov      rsi, qword ptr [(reloc 0x4000000000423518)]
       488D0D00000000       lea      rcx, gword ptr [(reloc 0x4000000000423580)]
       48894E08             mov      gword ptr [rsi+08H], rcx
       8B0D00000000         mov      ecx, dword ptr [(reloc 0x4000000000423508)]
       E800000000           call     System.Number:Int32ToDecStr(int):System.String
       488BC8               mov      rcx, rax
       488B5608             mov      rdx, gword ptr [rsi+08H]
       E800000000           call     System.String:Concat(System.String,System.String):System.String
       488BC8               mov      rcx, rax
       E800000000           call     System.Console:WriteLine(System.String)
       E800000000           call     System.Diagnostics.Debugger:Break()
       90                   nop      
G_M24707_IG03:              ;; offset=0046H
       4883C420             add      rsp, 32
       5E                   pop      rsi
       C3                   ret      
; Total bytes of code: 76