ArmArch codegen for empty / simple methods is not optimal

[am11]

The codegen for empty ctor on arm/arm64 is not as optimal when compared with x64. e.g. passing --optimize to CG2 compiles this code:

class C
{
    bool Foo() => true;
}

for arm64 as:

C:.ctor():this:    ; <- this should be just "ret" ?
            stp     fp, lr, [sp,#-16]!
            mov     fp, sp
            ldp     fp, lr, [sp],#16
            ret     lr

C:Foo():bool:this:
            stp     fp, lr, [sp,#-16]!
            mov     fp, sp
            mov     w0, #1
            ldp     fp, lr, [sp],#16
            ret     lr

and x64:

C:.ctor():this:
       ret

C:Foo():bool:this:
       mov      eax, 1
       ret

Similarly for arm, interestingly, which has relatively better codegen than arm64.

codegen:
C#: https://godbolt.org/z/zPhbecqab
C: https://godbolt.org/z/cPoa4ohPo

[ghost]

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

Issue Details

---

The codegen for empty ctor on arm/arm64 is not as optimal when compared with x64. e.g. passing --optimize to CG2 compiles this code:

class C
{
    bool Foo() => true;
}

for arm64 as:

C:.ctor():this:    ; <- this should be just "ret" ?
            stp     fp, lr, [sp,#-16]!
            mov     fp, sp
            ldp     fp, lr, [sp],#16
            ret     lr

C:Foo():bool:this:
            stp     fp, lr, [sp,#-16]!
            mov     fp, sp
            mov     w0, #1
            ldp     fp, lr, [sp],#16
            ret     lr

and x64:

C:.ctor():this:
       ret

C:Foo():bool:this:
       mov      eax, 1
       ret

Similarly for arm, interestingly, which has relatively better codegen than arm64.

codegen:
C#: https://godbolt.org/z/zPhbecqab
C: https://godbolt.org/z/cPoa4ohPo

Author:	am11
Assignees:	-
Labels:	area-CodeGen-coreclr, untriaged
Milestone:	-

[SingleAccretion]

Duplicate of #35274.

[am11]

Yep. Thanks!