Avoid boxing in Enum.HasFlag in Tier0

[EgorBo]

We see a lot of cases when methods stay in Tier0 despite being called a lot in large apps (the "tier1 promotion queue" problem) so we'd better remove allocations in Tier0 too where possible, I gave up on this opt in #77357 but I still think it's useful:

bool Test(MyEnum e) => e.HasFlag(MyEnum.B);

Main (Tier0):

; Assembly listing for method Program:Test(int):bool (Tier0)	
       push     rbp	
       sub      rsp, 48	
       lea      rbp, [rsp+30H]	
       xor      eax, eax	
       mov      qword ptr [rbp-08H], rax	
       mov      qword ptr [rbp-10H], rax	
       mov      dword ptr [rbp+10H], ecx	
       mov      rcx, 0x7FFAE2C3B208	
       call     CORINFO_HELP_NEWSFAST  ;; <--- Allocation
       mov      gword ptr [rbp-08H], rax	
       mov      rcx, gword ptr [rbp-08H]	
       mov      eax, dword ptr [rbp+10H]	
       mov      dword ptr [rcx+08H], eax	
       mov      rcx, 0x7FFAE2C3B208
       call     CORINFO_HELP_NEWSFAST  ;; <--- Allocation
       mov      gword ptr [rbp-10H], rax	
       mov      rcx, gword ptr [rbp-10H]	
       mov      dword ptr [rcx+08H], 1	
       mov      rcx, gword ptr [rbp-08H]	
       mov      rdx, gword ptr [rbp-10H]	
       call     [System.Enum:HasFlag(System.Enum):bool:this]	
       nop	
       add      rsp, 48	
       pop      rbp	
       ret	
; Total bytes of code 103

PR (Tier0):

; Assembly listing for method Program:Test(int):bool (Tier0)
       push     rbp
       sub      rsp, 32
       lea      rbp, [rsp+20H]
       xor      eax, eax
       mov      qword ptr [rbp-08H], rax
       mov      qword ptr [rbp-10H], rax
       mov      dword ptr [rbp+10H], ecx
       mov      eax, dword ptr [rbp+10H]
       mov      dword ptr [rbp-14H], eax
       mov      eax, dword ptr [rbp-14H]
       and      eax, 1
       cmp      eax, 1
       sete     al
       movzx    rax, al
       add      rsp, 32
       pop      rbp
       ret
; Total bytes of code 50

[ghost]

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

Issue Details

---

We see a lot of cases when methods stay in Tier0 despite being called a lot in large apps (the "tier1 promotion queue" problem) so we'd better remove allocations in Tier0 too where possible, I gave up on this opt in #77357 but I still think it's useful:

bool Test(MyEnum e) => e.HasFlag(MyEnum.B);

Main:

; Assembly listing for method Program:Test(int):bool (Tier0)	
       push     rbp	
       sub      rsp, 48	
       lea      rbp, [rsp+30H]	
       xor      eax, eax	
       mov      qword ptr [rbp-08H], rax	
       mov      qword ptr [rbp-10H], rax	
       mov      dword ptr [rbp+10H], ecx	
       mov      rcx, 0x7FFAE2C3B208	
       call     CORINFO_HELP_NEWSFAST  ;; <--- Allocation
       mov      gword ptr [rbp-08H], rax	
       mov      rcx, gword ptr [rbp-08H]	
       mov      eax, dword ptr [rbp+10H]	
       mov      dword ptr [rcx+08H], eax	
       mov      rcx, 0x7FFAE2C3B208
       call     CORINFO_HELP_NEWSFAST  ;; <--- Allocation
       mov      gword ptr [rbp-10H], rax	
       mov      rcx, gword ptr [rbp-10H]	
       mov      dword ptr [rcx+08H], 1	
       mov      rcx, gword ptr [rbp-08H]	
       mov      rdx, gword ptr [rbp-10H]	
       call     [System.Enum:HasFlag(System.Enum):bool:this]	
       nop	
       add      rsp, 48	
       pop      rbp	
       ret	
; Total bytes of code 103

PR:

; Assembly listing for method Program:Test(int):bool (Tier0)
       push     rbp
       sub      rsp, 32
       lea      rbp, [rsp+20H]
       xor      eax, eax
       mov      qword ptr [rbp-08H], rax
       mov      qword ptr [rbp-10H], rax
       mov      dword ptr [rbp+10H], ecx
       mov      eax, dword ptr [rbp+10H]
       mov      dword ptr [rbp-14H], eax
       mov      eax, dword ptr [rbp-14H]
       and      eax, 1
       cmp      eax, 1
       sete     al
       movzx    rax, al
       add      rsp, 32
       pop      rbp
       ret
; Total bytes of code 50

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

[EgorBo]

@MihuBot -tier0

[EgorBo]

@dotnet/jit-contrib PTAL, simple change - no SPMI diffs (new contexts), jit-diffs in tier0: Total bytes of delta: -2158 (-0.00 % of base)

[EgorBo]

@jakobbotsch @dotnet/jit-contrib PTAL, simple change

Should fix the concerns raised in AvaloniaUI/Avalonia#4873 (comment)

[EgorBo]

Acutally, I don't think we need to guard this under debug/minopts - it won't decrease quality of debug. Maybe we shouldn't share them for debug/minopts at all

[jakobbotsch]

Acutally, I don't think we need to guard this under debug/minopts - it won't decrease quality of debug. Maybe we shouldn't share them for debug/minopts at all

Presumably there is a reason the code exists in the first place. It would be good to verify the original scenarios that caused the code to be added do not get regressed with these changes.

Also cc @AndyAyersMS

[jakobbotsch]

Looks good to me with the above note -- it would be nice to double check whether there were some particular programs that caused this sharing to be added in the first place.

Would also like to hear @AndyAyersMS's opinion.

[jakobbotsch]

/azp run runtime-coreclr superpmi-diffs

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jakobbotsch]

Seems like there are build failures.

[EgorBo]

Seems like there are build failures.

Oops, fixed.

it would be nice to double check whether there were some particular programs that caused this sharing to be added in the first place.

I wonder if sharing actually makes debug experience worse.

[AndyAyersMS]

I don't believe there is any impact on debugging as the box temp is always a (evaluation) stack object.

I think the sharing bit is from way back; I always thought it was there to avoid some code size / frame bloat. If you disable sharing entirely presumably you might see some of this in SPMI.