Devirtualiazation and unboxing regression

[benaadams]

Seen in dotnet/coreclr#20936 (comment)

For AsyncTaskMethodBuilder``1:AwaitUnsafeOnCompleted(byref,byref):this

https://github.com/dotnet/coreclr/blob/463ba889b0c010eb9c6f9807eaf0ab4ab5624450/src/System.Private.CoreLib/src/System/Runtime/CompilerServices/AsyncMethodBuilder.cs#L382-L396

It looks like the boxing is back for ValueTaskAwaiter<T> and TaskAwaiter

G_M5080_IG02:
       488B4D10             mov      rcx, bword ptr [rbp+10H]
       488B5520             mov      rdx, bword ptr [rbp+20H]
       E8E3F6FFFF           call     AsyncTaskMethodBuilder`1:GetStateMachineBox(byref):ref:this
       488945E8             mov      gword ptr [rbp-18H], rax
       33D2                 xor      rdx, rdx
       488955E0             mov      qword ptr [rbp-20H], rdx
       488D55E0             lea      rdx, bword ptr [rbp-20H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E87EE8785F           call     CORINFO_HELP_BOX
       4885C0               test     rax, rax
       745B                 je       SHORT G_M5080_IG04
       488B5518             mov      rdx, bword ptr [rbp+18H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E866E8785F           call     CORINFO_HELP_BOX
       488945A8             mov      gword ptr [rbp-58H], rax
       488B55A8             mov      rdx, gword ptr [rbp-58H]
       48B9D0012EA0FA7F0000 mov      rcx, 0x7FFAA02E01D0
       E8AFD2785F           call     CORINFO_HELP_ISINSTANCEOFINTERFACE
       4885C0               test     rax, rax
       742C                 je       SHORT G_M5080_IG04
       488B4D18             mov      rcx, bword ptr [rbp+18H]
       E841EF8359           call     Unsafe:As(byref):byref
       488945D8             mov      bword ptr [rbp-28H], rax
       488B4DD8             mov      rcx, bword ptr [rbp-28H]
       488B09               mov      rcx, gword ptr [rcx]
       488B55E8             mov      rdx, gword ptr [rbp-18H]
       41B801000000         mov      r8d, 1
       E867268859           call     TaskAwaiter:UnsafeOnCompletedInternal(ref,ref,bool)
       90                   nop      

G_M5080_IG03:
       488D65F0             lea      rsp, [rbp-10H]
       5E                   pop      rsi
       5F                   pop      rdi
       5D                   pop      rbp
       C3                   ret      

G_M5080_IG04:
       33D2                 xor      rdx, rdx
       488955E0             mov      qword ptr [rbp-20H], rdx
       488D55E0             lea      rdx, bword ptr [rbp-20H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E805E8785F           call     CORINFO_HELP_BOX
       4885C0               test     rax, rax
       745E                 je       SHORT G_M5080_IG06
       488B5518             mov      rdx, bword ptr [rbp+18H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E8EDE7785F           call     CORINFO_HELP_BOX
       488945A0             mov      gword ptr [rbp-60H], rax
       488B55A0             mov      rdx, gword ptr [rbp-60H]
       48B910022EA0FA7F0000 mov      rcx, 0x7FFAA02E0210
       E836D2785F           call     CORINFO_HELP_ISINSTANCEOFINTERFACE
       4885C0               test     rax, rax
       742F                 je       SHORT G_M5080_IG06
       488B4D18             mov      rcx, bword ptr [rbp+18H]
       E898EDCDFF           call     Unsafe:As(byref):byref
       488945D0             mov      bword ptr [rbp-30H], rax
       4C8B45D0             mov      r8, bword ptr [rbp-30H]
       450FB64008           movzx    r8, byte  ptr [r8+8]
       488B4DD0             mov      rcx, bword ptr [rbp-30H]
       488B09               mov      rcx, gword ptr [rcx]
       488B55E8             mov      rdx, gword ptr [rbp-18H]
       E8EB258859           call     TaskAwaiter:UnsafeOnCompletedInternal(ref,ref,bool)
       90                   nop      

G_M5080_IG05:
       488D65F0             lea      rsp, [rbp-10H]
       5E                   pop      rsi
       5F                   pop      rdi
       5D                   pop      rbp
       C3                   ret      

G_M5080_IG06:
       33D2                 xor      rdx, rdx
       488955E0             mov      qword ptr [rbp-20H], rdx
       488D55E0             lea      rdx, bword ptr [rbp-20H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E889E7785F           call     CORINFO_HELP_BOX
       4885C0               test     rax, rax
       7479                 je       SHORT G_M5080_IG08
       488B5518             mov      rdx, bword ptr [rbp+18H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E871E7785F           call     CORINFO_HELP_BOX
       48894598             mov      gword ptr [rbp-68H], rax
       488B5598             mov      rdx, gword ptr [rbp-68H]
       48B980042EA0FA7F0000 mov      rcx, 0x7FFAA02E0480
       E8BAD1785F           call     CORINFO_HELP_ISINSTANCEOFINTERFACE
       4885C0               test     rax, rax
       744A                 je       SHORT G_M5080_IG08

G_M5080_IG07:
       488B5518             mov      rdx, bword ptr [rbp+18H]
       48B9701C31A0FA7F0000 mov      rcx, 0x7FFAA0311C70
       E842E7785F           call     CORINFO_HELP_BOX
       48894590             mov      gword ptr [rbp-70H], rax
       488B5590             mov      rdx, gword ptr [rbp-70H]
       48B980042EA0FA7F0000 mov      rcx, 0x7FFAA02E0480
       E84BD2785F           call     CORINFO_HELP_CHKCASTINTERFACE
       48894588             mov      gword ptr [rbp-78H], rax
       488B4D88             mov      rcx, gword ptr [rbp-78H]
       488B55E8             mov      rdx, gword ptr [rbp-18H]
       49BB70113246FA7F0000 mov      r11, 0x7FFA46321170
       3909                 cmp      dword ptr [rcx], ecx
       FF156D5EB8FF         call     [IStateMachineBoxAwareAwaiter:AwaitUnsafeOnCompleted(ref):this]
       EB29                 jmp      SHORT G_M5080_IG09

G_M5080_IG08:
       488B4DE8             mov      rcx, gword ptr [rbp-18H]
       49BB68113246FA7F0000 mov      r11, 0x7FFA46321168
       3909                 cmp      dword ptr [rcx], ecx
       FF154D5EB8FF         call     [IAsyncStateMachineBox:get_MoveNextAction():ref:this]
       48894580             mov      gword ptr [rbp-80H], rax
       488B5580             mov      rdx, gword ptr [rbp-80H]
       488B4D18             mov      rcx, bword ptr [rbp+18H]
       E854238859           call     TaskAwaiter:UnsafeOnCompleted(ref):this
       EB00                 jmp      SHORT G_M5080_IG09

Regression of dotnet/coreclr#14698?

/cc @AndyAyersMS @stephentoub

[benaadams]

Might be tiered jit?

[benaadams]

WIndows Update just broke my WSL so I can't retest by switching off tiering 😢

[benaadams]

Return; after making several requests this is definitely tiering as the asm output again; later but without boxing etc

[stephentoub]

It's tiered JIT. All of the boxing-related optimizations are disabled in tier 0. There's an open issues to consider enabling them and others that would actually help the JIT to be faster.

[danmoseley]

Cc @kouvel

[AndyAyersMS]

Putting in 3.0 for now ... I will look into enable boxing-related opts in Tier0.

[AndyAyersMS]

Most of the interesting discussion on this is happening over in dotnet/coreclr#22984.