[X64] [MihaZupan] Add SearchValues<char> implementation for two sets of 128 chars #464
Comments
Top method regressions15 (3.78 % of base) - System.Buffers.IndexOfAnyAsciiSearcher:ComputeAnyByteState(System.ReadOnlySpan`1[ubyte],byref) ; Assembly listing for method System.Buffers.IndexOfAnyAsciiSearcher:ComputeAnyByteState(System.ReadOnlySpan`1[ubyte],byref) (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; fully interruptible
; No PGO data
-; 0 inlinees with PGO data; 12 single block inlinees; 4 inlinees without PGO data
+; 0 inlinees with PGO data; 14 single block inlinees; 4 inlinees without PGO data
; Final local variable assignments
;
-; V00 arg0 [V00,T11] ( 3, 3 ) struct (16) [rbp-0x118] do-not-enreg[SA] multireg-arg single-def ptr <System.ReadOnlySpan`1[ubyte]>
-; V01 arg1 [V01,T10] ( 3, 3 ) byref -> rdx single-def
+; V00 arg0 [V00,T10] ( 3, 3 ) struct (16) [rbp-0x118] do-not-enreg[SA] multireg-arg single-def ptr <System.ReadOnlySpan`1[ubyte]>
+; V01 arg1 [V01,T09] ( 3, 3 ) byref -> rdx single-def
; V02 loc0 [V02 ] ( 3, 4 ) simd16 -> [rbp-0x130] do-not-enreg[XS] addr-exposed ld-addr-op <System.Runtime.Intrinsics.Vector128`1[ubyte]>
; V03 loc1 [V03 ] ( 3, 4 ) simd16 -> [rbp-0x140] do-not-enreg[XS] addr-exposed ld-addr-op <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-;* V04 loc2 [V04,T17] ( 0, 0 ) long -> zero-ref ptr
-;* V05 loc3 [V05,T18] ( 0, 0 ) long -> zero-ref ptr
+;* V04 loc2 [V04,T18] ( 0, 0 ) long -> zero-ref ptr
+;* V05 loc3 [V05,T19] ( 0, 0 ) long -> zero-ref ptr
; V06 loc4 [V06 ] ( 3, 6 ) struct (32) [rbp-0x28] do-not-enreg[XS] addr-exposed ld-addr-op unsafe-buffer <System.Buffers.BitVector256>
;* V07 loc5 [V07 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.ReadOnlySpan`1[ubyte]>
-;* V08 loc6 [V08,T19] ( 0, 0 ) int -> zero-ref ptr
-; V09 loc7 [V09,T01] ( 5, 20 ) ubyte -> rsi
-; V10 loc8 [V10,T03] ( 4, 12 ) int -> r8
-; V11 loc9 [V11,T06] ( 3, 8 ) int -> rsi
-; V12 tmp0 [V12,T20] ( 1, 1 ) int -> [rbp-0x144] do-not-enreg[V] "GSCookie dummy"
-;# V13 OutArgs [V13 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
-; V14 tmp2 [V14,T04] ( 3, 12 ) long -> rsi "dup spill"
-; V15 tmp3 [V15,T05] ( 3, 12 ) long -> rsi "dup spill"
-; V16 tmp4 [V16,T14] ( 2, 4 ) struct (192) [rbp-0xE8] do-not-enreg[SF] ld-addr-op unsafe-buffer "NewObj constructor temp" <System.Buffers.IndexOfAnyAsciiSearcher+AnyByteState>
-; V17 tmp5 [V17,T07] ( 2, 8 ) int -> r8 "Inline stloc first use temp"
-; V18 tmp6 [V18,T08] ( 2, 8 ) int -> r9 "Inline stloc first use temp"
-; V19 tmp7 [V19,T00] ( 3, 24 ) byref -> r8 "dup spill"
-; V20 tmp8 [V20,T21] ( 3, 6 ) simd16 -> mm0 "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V21 tmp9 [V21,T22] ( 3, 6 ) simd16 -> mm1 "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
-; V22 tmp10 [V22,T15] ( 2, 4 ) struct (32) [rbp-0x108] do-not-enreg[S] unsafe-buffer "Inlining Arg" <System.Buffers.BitVector256>
-; V23 tmp11 [V23,T23] ( 3, 6 ) simd32 -> mm0 "spilled call-like call argument"
-;* V24 tmp12 [V24 ] ( 0, 0 ) simd32 -> zero-ref "Inline return value spill temp" <System.Runtime.Intrinsics.Vector256`1[ubyte]>
-; V25 tmp13 [V25,T27] ( 1, 1 ) simd64 -> [rbp-0x1B0] ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector512`1[ubyte]>
-; V26 tmp14 [V26,T24] ( 3, 6 ) simd32 -> mm1 "spilled call-like call argument"
-;* V27 tmp15 [V27 ] ( 0, 0 ) simd32 -> zero-ref "Inline return value spill temp" <System.Runtime.Intrinsics.Vector256`1[ubyte]>
-; V28 tmp16 [V28,T28] ( 1, 1 ) simd64 -> [rbp-0x1F0] ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector512`1[ubyte]>
-; V29 tmp17 [V29,T12] ( 2, 5 ) byref -> rax single-def "field V07._reference (fldOffset=0x0)" P-INDEP
-; V30 tmp18 [V30,T09] ( 3, 6 ) int -> rcx "field V07._length (fldOffset=0x8)" P-INDEP
-; V31 tmp19 [V31,T25] ( 2, 2 ) simd64 -> mm0 "V16.[000..064)"
-; V32 tmp20 [V32,T26] ( 2, 2 ) simd64 -> mm1 "V16.[064..128)"
-; V33 GsCookie [V33 ] ( 1, 1 ) long -> [rbp-0x08] do-not-enreg[X] addr-exposed "GSSecurityCookie"
-; V34 tmp22 [V34,T16] ( 3, 3 ) struct (16) [rbp-0x200] do-not-enreg[SA] multireg-arg must-init ptr "shadowVar" <System.ReadOnlySpan`1[ubyte]>
-; V35 tmp23 [V35,T13] ( 4, 4 ) byref -> rdx single-def "shadowVar"
-; V36 rat0 [V36,T02] ( 5, 17 ) long -> rdi "Widened IV V08"
+;* V08 loc6 [V08,T20] ( 0, 0 ) int -> zero-ref ptr
+; V09 loc7 [V09,T01] ( 7, 22 ) ubyte -> rsi
+; V10 tmp0 [V10,T21] ( 1, 1 ) int -> [rbp-0x144] do-not-enreg[V] "GSCookie dummy"
+;# V11 OutArgs [V11 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
+; V12 tmp2 [V12,T13] ( 2, 4 ) struct (192) [rbp-0xE8] do-not-enreg[SF] ld-addr-op unsafe-buffer "NewObj constructor temp" <System.Buffers.IndexOfAnyAsciiSearcher+AnyByteState>
+; V13 tmp3 [V13,T06] ( 2, 8 ) int -> r8 "Inline stloc first use temp"
+; V14 tmp4 [V14,T07] ( 2, 8 ) int -> r9 "Inline stloc first use temp"
+; V15 tmp5 [V15,T00] ( 3, 24 ) byref -> r8 "dup spill"
+; V16 tmp6 [V16,T14] ( 2, 4 ) int -> r8 "Inline stloc first use temp"
+;* V17 tmp7 [V17 ] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V18 tmp8 [V18,T03] ( 3, 12 ) long -> rsi "dup spill"
+; V19 tmp9 [V19,T04] ( 3, 12 ) int -> rsi "Inlining Arg"
+; V20 tmp10 [V20,T15] ( 2, 4 ) int -> r8 "Inline stloc first use temp"
+;* V21 tmp11 [V21 ] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
+; V22 tmp12 [V22,T05] ( 3, 12 ) long -> rsi "dup spill"
+; V23 tmp13 [V23,T22] ( 3, 6 ) simd16 -> mm0 "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V24 tmp14 [V24,T23] ( 3, 6 ) simd16 -> mm1 "Inlining Arg" <System.Runtime.Intrinsics.Vector128`1[ubyte]>
+; V25 tmp15 [V25,T16] ( 2, 4 ) struct (32) [rbp-0x108] do-not-enreg[S] unsafe-buffer "Inlining Arg" <System.Buffers.BitVector256>
+; V26 tmp16 [V26,T24] ( 3, 6 ) simd32 -> mm0 "spilled call-like call argument"
+;* V27 tmp17 [V27 ] ( 0, 0 ) simd32 -> zero-ref "Inline return value spill temp" <System.Runtime.Intrinsics.Vector256`1[ubyte]>
+; V28 tmp18 [V28,T28] ( 1, 1 ) simd64 -> [rbp-0x1B0] ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector512`1[ubyte]>
+; V29 tmp19 [V29,T25] ( 3, 6 ) simd32 -> mm1 "spilled call-like call argument"
+;* V30 tmp20 [V30 ] ( 0, 0 ) simd32 -> zero-ref "Inline return value spill temp" <System.Runtime.Intrinsics.Vector256`1[ubyte]>
+; V31 tmp21 [V31,T29] ( 1, 1 ) simd64 -> [rbp-0x1F0] ld-addr-op "Inline ldloca(s) first use temp" <System.Runtime.Intrinsics.Vector512`1[ubyte]>
+; V32 tmp22 [V32,T11] ( 2, 5 ) byref -> rax single-def "field V07._reference (fldOffset=0x0)" P-INDEP
+; V33 tmp23 [V33,T08] ( 3, 6 ) int -> rcx "field V07._length (fldOffset=0x8)" P-INDEP
+; V34 tmp24 [V34,T26] ( 2, 2 ) simd64 -> mm0 "V12.[000..064)"
+; V35 tmp25 [V35,T27] ( 2, 2 ) simd64 -> mm1 "V12.[064..128)"
+; V36 GsCookie [V36 ] ( 1, 1 ) long -> [rbp-0x08] do-not-enreg[X] addr-exposed "GSSecurityCookie"
+; V37 tmp27 [V37,T17] ( 3, 3 ) struct (16) [rbp-0x200] do-not-enreg[SA] multireg-arg must-init ptr "shadowVar" <System.ReadOnlySpan`1[ubyte]>
+; V38 tmp28 [V38,T12] ( 4, 4 ) byref -> rdx single-def "shadowVar"
+; V39 rat0 [V39,T02] ( 5, 17 ) long -> rdi "Widened IV V08"
;
; Lcl frame size = 512
G_M64303_IG01:
push rbp
sub rsp, 512
lea rbp, [rsp+0x200]
xor eax, eax
mov qword ptr [rbp-0x200], rax
mov qword ptr [rbp-0x08], 0xD1FFAB1E
mov bword ptr [rbp-0x118], rdi
mov qword ptr [rbp-0x110], rsi
;; size=47 bbWeight=1 PerfScore 6.00
G_M64303_IG02:
vmovdqu xmm0, xmmword ptr [rbp-0x118]
vmovdqu xmmword ptr [rbp-0x200], xmm0
;; size=16 bbWeight=1 PerfScore 4.00
G_M64303_IG03:
vxorps xmm0, xmm0, xmm0
vmovaps xmmword ptr [rbp-0x130], xmm0
vmovaps xmmword ptr [rbp-0x140], xmm0
vxorps ymm0, ymm0, ymm0
vmovdqu ymmword ptr [rbp-0x28], ymm0
mov rax, bword ptr [rbp-0x200]
mov ecx, dword ptr [rbp-0x1F8]
xor edi, edi
test ecx, ecx
- jle SHORT G_M64303_IG08
+ jle G_M64303_IG08
align [0 bytes for IG04]
- ;; size=48 bbWeight=1 PerfScore 7.17
+ ;; size=52 bbWeight=1 PerfScore 7.17
G_M64303_IG04:
movzx rsi, byte ptr [rax+rdi]
mov r8d, esi
sar r8d, 5
mov r9d, 1
shlx r9d, r9d, esi
lea r10, bword ptr [rbp-0x28]
lea r8, bword ptr [r10+4*r8]
or dword ptr [r8], r9d
- mov r8d, esi
- sar r8d, 4
- and esi, 15
- cmp r8d, 8
+ cmp esi, 128
jge SHORT G_M64303_IG06
- ;; size=50 bbWeight=4 PerfScore 39.00
+ ;; size=42 bbWeight=4 PerfScore 35.00
G_M64303_IG05:
+ mov r8d, esi
+ sar r8d, 4
lea r9, [rbp-0x130]
- mov esi, esi
+ and esi, 15
add rsi, r9
mov r9d, 1
shlx r8d, r9d, r8d
movzx r8, r8b
or byte ptr [rsi], r8b
jmp SHORT G_M64303_IG07
- ;; size=32 bbWeight=2 PerfScore 14.00
+ ;; size=40 bbWeight=2 PerfScore 15.50
G_M64303_IG06:
+ add esi, -128
+ mov r8d, esi
+ sar r8d, 4
lea r9, [rbp-0x140]
- mov esi, esi
+ and esi, 15
add rsi, r9
- add r8d, -8
mov r9d, 1
shlx r8d, r9d, r8d
movzx r8, r8b
or byte ptr [rsi], r8b
- ;; size=34 bbWeight=2 PerfScore 10.50
+ ;; size=41 bbWeight=2 PerfScore 12.00
G_M64303_IG07:
inc edi
cmp edi, ecx
- jl SHORT G_M64303_IG04
- ;; size=6 bbWeight=4 PerfScore 6.00
+ jl G_M64303_IG04
+ ;; size=10 bbWeight=4 PerfScore 6.00
G_M64303_IG08:
vmovaps xmm0, xmmword ptr [rbp-0x130]
vmovaps xmm1, xmmword ptr [rbp-0x140]
vmovdqu ymm2, ymmword ptr [rbp-0x28]
vmovdqu ymmword ptr [rbp-0x108], ymm2
vmovaps ymm2, ymm0
vinserti128 ymm0, ymm2, xmm0, 1
vmovups zmm2, zmmword ptr [rbp-0x1B0]
vinsertf64x4 zmm2, zmm2, ymm0, 0
vinsertf64x4 zmm0, zmm2, ymm0, 1
vmovaps ymm2, ymm1
vinserti128 ymm1, ymm2, xmm1, 1
vmovups zmm2, zmmword ptr [rbp-0x1F0]
vinsertf64x4 zmm2, zmm2, ymm1, 0
vinsertf64x4 zmm1, zmm2, ymm1, 1
vmovdqu ymm2, ymmword ptr [rbp-0x108]
vmovdqu ymmword ptr [rbp-0x68], ymm2
vmovups zmmword ptr [rdx], zmm0
vmovups zmmword ptr [rdx+0x40], zmm1
vmovups ymm0, ymmword ptr [rbp-0x68]
vmovups ymmword ptr [rdx+0x80], ymm0
cmp qword ptr [rbp-0x08], 0xD1FFAB1E
je SHORT G_M64303_IG09
call CORINFO_HELP_FAIL_FAST
;; size=151 bbWeight=1 PerfScore 48.50
G_M64303_IG09:
nop
;; size=1 bbWeight=1 PerfScore 0.25
G_M64303_IG10:
vzeroupper
add rsp, 512
pop rbp
ret
;; size=12 bbWeight=1 PerfScore 2.75
-; Total bytes of code 397, prolog size 47, PerfScore 138.17, instruction count 81, allocated bytes for code 401 (MethodHash=97c904d0) for method System.Buffers.IndexOfAnyAsciiSearcher:ComputeAnyByteState(System.ReadOnlySpan`1[ubyte],byref) (FullOpts)
+; Total bytes of code 412, prolog size 47, PerfScore 137.17, instruction count 82, allocated bytes for code 412 (MethodHash=97c904d0) for method System.Buffers.IndexOfAnyAsciiSearcher:ComputeAnyByteState(System.ReadOnlySpan`1[ubyte],byref) (FullOpts)Note: some changes were skipped as they were too large to fit into a comment. Larger list of diffs: https://gist.github.com/MihuBot/4d0dd4484f7c8523a4e0a64b9031560c |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Job completed in 14 minutes.
dotnet/runtime#103216
Diffs
Artifacts:
The text was updated successfully, but these errors were encountered: