Avoid WriteBarrier assign for nulling struct #7659
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Changes from this more expensive asm complete with G_M56063_IG01:
57 push rdi
56 push rsi
55 push rbp
53 push rbx
4881ECA8000000 sub rsp, 168
C5F877 vzeroupper
488BF1 mov rsi, rcx
488D7C2428 lea rdi, [rsp+28H]
B920000000 mov ecx, 32
33C0 xor rax, rax
F3AB rep stosd
488BCE mov rcx, rsi
488BD9 mov rbx, rcx
488BEA mov rbp, rdx
G_M56063_IG02:
48896C2468 mov gword ptr [rsp+68H], rbp
488D4C2478 lea rcx, bword ptr [rsp+78H]
C5F857C0 vxorps xmm0, xmm0
C5FA7F01 vmovdqu qword ptr [rcx], xmm0
C5FA7F4110 vmovdqu qword ptr [rcx+16], xmm0
C5FA7F4120 vmovdqu qword ptr [rcx+32], xmm0
488BCD mov rcx, rbp
49BBD811C8D9F87F0000 mov r11, 0x7FF8D9C811D8
3909 cmp dword ptr [rcx], ecx
FF15CF52A3FF call [IFeatureCollection:get_Revision():int:this]
89442470 mov dword ptr [rsp+70H], eax
488D7B38 lea rdi, bword ptr [rbx+56]
488D742468 lea rsi, bword ptr [rsp+68H]
E8E5FB675F call CORINFO_HELP_ASSIGN_BYREF
48A5 movsq
E8DEFB675F call CORINFO_HELP_ASSIGN_BYREF
E8D9FB675F call CORINFO_HELP_ASSIGN_BYREF
E8D4FB675F call CORINFO_HELP_ASSIGN_BYREF
E8CFFB675F call CORINFO_HELP_ASSIGN_BYREF
E8CAFB675F call CORINFO_HELP_ASSIGN_BYREF
E8C5FB675F call CORINFO_HELP_ASSIGN_BYREFTo this more svelte number G_M56075_IG01:
57 push rdi
56 push rsi
55 push rbp
53 push rbx
4883EC28 sub rsp, 40
488BF1 mov rsi, rcx
488BFA mov rdi, rdx
G_M56075_IG02:
3936 cmp dword ptr [rsi], esi
488D5E38 lea rbx, bword ptr [rsi+56]
488BCF mov rcx, rdi
49BBD81109D7F87F0000 mov r11, 0x7FF8D70911D8
3909 cmp dword ptr [rcx], ecx
FF15DF4FA4FF call [IFeatureCollection:get_Revision():int:this]
894308 mov dword ptr [rbx+8], eax
488BCB mov rcx, rbx
488BD7 mov rdx, rdi
E829F8685F call CORINFO_HELP_CHECKED_ASSIGN_REF |
benaadams
force-pushed
the
Work-around-WriteBarrier-assign-for-nulling-struct
branch
from
b58ef2c to
e815690
Compare
|
And finally down to one interface call to ; Assembly listing for method DefaultHttpContext:Initialize(ref):this
; Lcl frame size = 32
G_M56075_IG01:
57 push rdi
56 push rsi
53 push rbx
4883EC20 sub rsp, 32
488BF1 mov rsi, rcx
488BFA mov rdi, rdx
G_M56075_IG02:
488BCF mov rcx, rdi
49BBD8113AD7F87F0000 mov r11, 0x7FF8D73A11D8
3909 cmp dword ptr [rcx], ecx
FF157654A4FF call [IFeatureCollection:get_Revision():int:this]
8BD8 mov ebx, eax
3936 cmp dword ptr [rsi], esi
488D4E38 lea rcx, bword ptr [rsi+56]
895908 mov dword ptr [rcx+8], ebx
488BD7 mov rdx, rdi
E8BBFC695F call CORINFO_HELP_CHECKED_ASSIGN_REF
488B5608 mov rdx, gword ptr [rsi+8]
8B0A mov ecx, dword ptr [rdx]
488D4A18 lea rcx, bword ptr [rdx+24]
488B5210 mov rdx, gword ptr [rdx+16]
8B02 mov eax, dword ptr [rdx]
4883C238 add rdx, 56
488B12 mov rdx, gword ptr [rdx]
895908 mov dword ptr [rcx+8], ebx
E89CFC695F call CORINFO_HELP_CHECKED_ASSIGN_REF
488B5610 mov rdx, gword ptr [rsi+16]
8B0A mov ecx, dword ptr [rdx]
488D4A10 lea rcx, bword ptr [rdx+16]
488B5208 mov rdx, gword ptr [rdx+8]
8B02 mov eax, dword ptr [rdx]
4883C238 add rdx, 56
488B12 mov rdx, gword ptr [rdx]
895908 mov dword ptr [rcx+8], ebx
E87DFC695F call CORINFO_HELP_CHECKED_ASSIGN_REF
488B4E18 mov rcx, gword ptr [rsi+24]
4885C9 test rcx, rcx
740F je SHORT G_M56075_IG04
G_M56075_IG03:
4883C108 add rcx, 8
895908 mov dword ptr [rcx+8], ebx
488BD7 mov rdx, rdi
E865FC695F call CORINFO_HELP_CHECKED_ASSIGN_REF
G_M56075_IG04:
488B4E20 mov rcx, gword ptr [rsi+32]
4885C9 test rcx, rcx
7508 jne SHORT G_M56075_IG06
G_M56075_IG05:
4883C420 add rsp, 32
5B pop rbx
5E pop rsi
5F pop rdi
C3 ret
G_M56075_IG06:
4883C108 add rcx, 8
895908 mov dword ptr [rcx+8], ebx
488BD7 mov rdx, rdi
E845FC695F call CORINFO_HELP_CHECKED_ASSIGN_REF
90 nop
G_M56075_IG07:
4883C420 add rsp, 32
5B pop rbx
5E pop rsi
5F pop rdi
C3 ret
; Total bytes of code 180, prolog size 7 for method DefaultHttpContext:Initialize(ref):this |
|
Pre
Post
Pre (7 Revision calls per request)
Post (5 Revision calls per request)
|
|
Should take a big chunk out of the 3.7% time now rather than just the 2% time of the WriteBarrier
|
davidfowl
approved these changes
Feb 17, 2019
benaadams
force-pushed
the
Work-around-WriteBarrier-assign-for-nulling-struct
branch
from
e9b578f to
e815690
Compare
benaadams
force-pushed
the
Work-around-WriteBarrier-assign-for-nulling-struct
branch
from
e815690 to
be3a6ba
Compare
|
Squashed, so it was always a perfect change 😉 |
jkotalik
approved these changes
Feb 17, 2019
benaadams
deleted the
Work-around-WriteBarrier-assign-for-nulling-struct
branch
benaadams
changed the title
amcasey
added
area-networking
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Its
defaultin the .ctors so only needs the two fields set;Uninitalizesets it todefaultsoInitalizeagain only needs to set the two fields and doesn't need to blank theCachefield.Workaround for https://github.com/dotnet/coreclr/issues/22661 to skip 5
Jit_ByRefWriteBarriers (and 10 more in called methods + conditionally 4 more; so eliminates up to 19CORINFO_HELP_ASSIGN_BYREFper request)Jit_ByRefWriteBarriershows up at around 2% of total run time in the TE benchmarks from this entry point:+5% in TE Plaintext benchmarks without the stack clear and the additional skips of the interface lookup for revision in this PR.
/cc @davidfowl
/cc @kkokosa as I read about these in a book recently :)