[JitDiff X64] xtqqczze/dotnet-runtime/remove-uint-cast2 #826
Comments
xtqqczze
referenced
this issue
in xtqqczze/dotnet-runtime
Dec 15, 2024
Top method regressions31 (30.39 % of base) - System.Collections.Generic.ValueListBuilder`1[double]:AppendSpan(int):System.Span`1[double]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[double]:AppendSpan(int):System.Span`1[double]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[double]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[double]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[double]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rax "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r14 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[double]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[double]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 16
+; Lcl frame size = 24
G_M9862_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 16
- lea rbp, [rsp+0x20]
+ sub rsp, 24
+ lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x20], rax
- ;; size=19 bbWeight=1 PerfScore 5.00
+ mov qword ptr [rbp-0x28], rax
+ ;; size=21 bbWeight=1 PerfScore 6.00
G_M9862_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov edx, esi
- add rcx, rdx
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, esi
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M9862_IG05
- ;; size=22 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M9862_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[double]:AppendSpanWithGrow(int):System.Span`1[double]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[double]:AppendSpanWithGrow(int):System.Span`1[double]:this
- mov bword ptr [rbp-0x20], rax
- mov qword ptr [rbp-0x18], rdx
- mov rax, bword ptr [rbp-0x20]
- mov rdx, qword ptr [rbp-0x18]
+ mov bword ptr [rbp-0x28], rax
+ mov qword ptr [rbp-0x20], rdx
+ mov rax, bword ptr [rbp-0x28]
+ mov rdx, qword ptr [rbp-0x20]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M9862_IG04:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M9862_IG05:
- lea eax, [rbx+rsi]
- mov dword ptr [rdi+0x08], eax
+ lea edx, [rbx+rsi]
+ mov dword ptr [rdi+0x08], edx
+ mov edx, r14d
+ cmp rax, rdx
+ ja SHORT G_M9862_IG07
mov eax, ebx
lea rax, bword ptr [r15+8*rax]
mov rdx, rsi
- ;; size=15 bbWeight=0.50 PerfScore 1.25
+ ;; size=23 bbWeight=0.50 PerfScore 2.00
G_M9862_IG06:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M9862_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 102, prolog size 19, PerfScore 20.62, instruction count 36, allocated bytes for code 102 (MethodHash=b075d979) for method System.Collections.Generic.ValueListBuilder`1[double]:AppendSpan(int):System.Span`1[double]:this (FullOpts)
+; Total bytes of code 133, prolog size 21, PerfScore 23.12, instruction count 46, allocated bytes for code 133 (MethodHash=b075d979) for method System.Collections.Generic.ValueListBuilder`1[double]:AppendSpan(int):System.Span`1[double]:this (FullOpts)
; ============================================================31 (30.39 % of base) - System.Collections.Generic.ValueListBuilder`1[int]:AppendSpan(int):System.Span`1[int]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[int]:AppendSpan(int):System.Span`1[int]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[int]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[int]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[int]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rax "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r14 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[int]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[int]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 16
+; Lcl frame size = 24
G_M13542_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 16
- lea rbp, [rsp+0x20]
+ sub rsp, 24
+ lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x20], rax
- ;; size=19 bbWeight=1 PerfScore 5.00
+ mov qword ptr [rbp-0x28], rax
+ ;; size=21 bbWeight=1 PerfScore 6.00
G_M13542_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov edx, esi
- add rcx, rdx
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, esi
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M13542_IG05
- ;; size=22 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M13542_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[int]:AppendSpanWithGrow(int):System.Span`1[int]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[int]:AppendSpanWithGrow(int):System.Span`1[int]:this
- mov bword ptr [rbp-0x20], rax
- mov qword ptr [rbp-0x18], rdx
- mov rax, bword ptr [rbp-0x20]
- mov rdx, qword ptr [rbp-0x18]
+ mov bword ptr [rbp-0x28], rax
+ mov qword ptr [rbp-0x20], rdx
+ mov rax, bword ptr [rbp-0x28]
+ mov rdx, qword ptr [rbp-0x20]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M13542_IG04:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M13542_IG05:
- lea eax, [rbx+rsi]
- mov dword ptr [rdi+0x08], eax
+ lea edx, [rbx+rsi]
+ mov dword ptr [rdi+0x08], edx
+ mov edx, r14d
+ cmp rax, rdx
+ ja SHORT G_M13542_IG07
mov eax, ebx
lea rax, bword ptr [r15+4*rax]
mov rdx, rsi
- ;; size=15 bbWeight=0.50 PerfScore 1.25
+ ;; size=23 bbWeight=0.50 PerfScore 2.00
G_M13542_IG06:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M13542_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 102, prolog size 19, PerfScore 20.62, instruction count 36, allocated bytes for code 102 (MethodHash=b0bbcb19) for method System.Collections.Generic.ValueListBuilder`1[int]:AppendSpan(int):System.Span`1[int]:this (FullOpts)
+; Total bytes of code 133, prolog size 21, PerfScore 23.12, instruction count 46, allocated bytes for code 133 (MethodHash=b0bbcb19) for method System.Collections.Generic.ValueListBuilder`1[int]:AppendSpan(int):System.Span`1[int]:this (FullOpts)
; ============================================================31 (30.39 % of base) - System.Collections.Generic.ValueListBuilder`1[long]:AppendSpan(int):System.Span`1[long]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[long]:AppendSpan(int):System.Span`1[long]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[long]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[long]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[long]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rax "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r14 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[long]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[long]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 16
+; Lcl frame size = 24
G_M62022_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 16
- lea rbp, [rsp+0x20]
+ sub rsp, 24
+ lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x20], rax
- ;; size=19 bbWeight=1 PerfScore 5.00
+ mov qword ptr [rbp-0x28], rax
+ ;; size=21 bbWeight=1 PerfScore 6.00
G_M62022_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov edx, esi
- add rcx, rdx
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, esi
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M62022_IG05
- ;; size=22 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M62022_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[long]:AppendSpanWithGrow(int):System.Span`1[long]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[long]:AppendSpanWithGrow(int):System.Span`1[long]:this
- mov bword ptr [rbp-0x20], rax
- mov qword ptr [rbp-0x18], rdx
- mov rax, bword ptr [rbp-0x20]
- mov rdx, qword ptr [rbp-0x18]
+ mov bword ptr [rbp-0x28], rax
+ mov qword ptr [rbp-0x20], rdx
+ mov rax, bword ptr [rbp-0x28]
+ mov rdx, qword ptr [rbp-0x20]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M62022_IG04:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M62022_IG05:
- lea eax, [rbx+rsi]
- mov dword ptr [rdi+0x08], eax
+ lea edx, [rbx+rsi]
+ mov dword ptr [rdi+0x08], edx
+ mov edx, r14d
+ cmp rax, rdx
+ ja SHORT G_M62022_IG07
mov eax, ebx
lea rax, bword ptr [r15+8*rax]
mov rdx, rsi
- ;; size=15 bbWeight=0.50 PerfScore 1.25
+ ;; size=23 bbWeight=0.50 PerfScore 2.00
G_M62022_IG06:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M62022_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 102, prolog size 19, PerfScore 20.62, instruction count 36, allocated bytes for code 102 (MethodHash=5c0b0db9) for method System.Collections.Generic.ValueListBuilder`1[long]:AppendSpan(int):System.Span`1[long]:this (FullOpts)
+; Total bytes of code 133, prolog size 21, PerfScore 23.12, instruction count 46, allocated bytes for code 133 (MethodHash=5c0b0db9) for method System.Collections.Generic.ValueListBuilder`1[long]:AppendSpan(int):System.Span`1[long]:this (FullOpts)
; ============================================================31 (30.39 % of base) - System.Collections.Generic.ValueListBuilder`1[short]:AppendSpan(int):System.Span`1[short]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[short]:AppendSpan(int):System.Span`1[short]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[short]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[short]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[short]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rax "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r14 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[short]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[short]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 16
+; Lcl frame size = 24
G_M56006_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 16
- lea rbp, [rsp+0x20]
+ sub rsp, 24
+ lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x20], rax
- ;; size=19 bbWeight=1 PerfScore 5.00
+ mov qword ptr [rbp-0x28], rax
+ ;; size=21 bbWeight=1 PerfScore 6.00
G_M56006_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov edx, esi
- add rcx, rdx
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, esi
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M56006_IG05
- ;; size=22 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M56006_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[short]:AppendSpanWithGrow(int):System.Span`1[short]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[short]:AppendSpanWithGrow(int):System.Span`1[short]:this
- mov bword ptr [rbp-0x20], rax
- mov qword ptr [rbp-0x18], rdx
- mov rax, bword ptr [rbp-0x20]
- mov rdx, qword ptr [rbp-0x18]
+ mov bword ptr [rbp-0x28], rax
+ mov qword ptr [rbp-0x20], rdx
+ mov rax, bword ptr [rbp-0x28]
+ mov rdx, qword ptr [rbp-0x20]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M56006_IG04:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M56006_IG05:
- lea eax, [rbx+rsi]
- mov dword ptr [rdi+0x08], eax
+ lea edx, [rbx+rsi]
+ mov dword ptr [rdi+0x08], edx
+ mov edx, r14d
+ cmp rax, rdx
+ ja SHORT G_M56006_IG07
mov eax, ebx
lea rax, bword ptr [r15+2*rax]
mov rdx, rsi
- ;; size=15 bbWeight=0.50 PerfScore 1.25
+ ;; size=23 bbWeight=0.50 PerfScore 2.00
G_M56006_IG06:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M56006_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 102, prolog size 19, PerfScore 20.62, instruction count 36, allocated bytes for code 102 (MethodHash=48f22539) for method System.Collections.Generic.ValueListBuilder`1[short]:AppendSpan(int):System.Span`1[short]:this (FullOpts)
+; Total bytes of code 133, prolog size 21, PerfScore 23.12, instruction count 46, allocated bytes for code 133 (MethodHash=48f22539) for method System.Collections.Generic.ValueListBuilder`1[short]:AppendSpan(int):System.Span`1[short]:this (FullOpts)
; ============================================================31 (30.39 % of base) - System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpan(int):System.Span`1[System.Nullable`1[int]]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpan(int):System.Span`1[System.Nullable`1[int]]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[System.Nullable`1[int]]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[System.Nullable`1[int]]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[System.Nullable`1[int]]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rax "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r14 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.Nullable`1[int]]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.Nullable`1[int]]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 16
+; Lcl frame size = 24
G_M4742_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 16
- lea rbp, [rsp+0x20]
+ sub rsp, 24
+ lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x20], rax
- ;; size=19 bbWeight=1 PerfScore 5.00
+ mov qword ptr [rbp-0x28], rax
+ ;; size=21 bbWeight=1 PerfScore 6.00
G_M4742_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov edx, esi
- add rcx, rdx
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, esi
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M4742_IG05
- ;; size=22 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M4742_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpanWithGrow(int):System.Span`1[System.Nullable`1[int]]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpanWithGrow(int):System.Span`1[System.Nullable`1[int]]:this
- mov bword ptr [rbp-0x20], rax
- mov qword ptr [rbp-0x18], rdx
- mov rax, bword ptr [rbp-0x20]
- mov rdx, qword ptr [rbp-0x18]
+ mov bword ptr [rbp-0x28], rax
+ mov qword ptr [rbp-0x20], rdx
+ mov rax, bword ptr [rbp-0x28]
+ mov rdx, qword ptr [rbp-0x20]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M4742_IG04:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M4742_IG05:
- lea eax, [rbx+rsi]
- mov dword ptr [rdi+0x08], eax
+ lea edx, [rbx+rsi]
+ mov dword ptr [rdi+0x08], edx
+ mov edx, r14d
+ cmp rax, rdx
+ ja SHORT G_M4742_IG07
mov eax, ebx
lea rax, bword ptr [r15+8*rax]
mov rdx, rsi
- ;; size=15 bbWeight=0.50 PerfScore 1.25
+ ;; size=23 bbWeight=0.50 PerfScore 2.00
G_M4742_IG06:
- add rsp, 16
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M4742_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 102, prolog size 19, PerfScore 20.62, instruction count 36, allocated bytes for code 102 (MethodHash=a57ced79) for method System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpan(int):System.Span`1[System.Nullable`1[int]]:this (FullOpts)
+; Total bytes of code 133, prolog size 21, PerfScore 23.12, instruction count 46, allocated bytes for code 133 (MethodHash=a57ced79) for method System.Collections.Generic.ValueListBuilder`1[System.Nullable`1[int]]:AppendSpan(int):System.Span`1[System.Nullable`1[int]]:this (FullOpts)
; ============================================================30 (28.85 % of base) - System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpan(int):System.Span`1[System.__Canon]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpan(int):System.Span`1[System.__Canon]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 TypeCtx [V01,T02] ( 4, 3.50) long -> rsi single-def
; V02 arg1 [V02,T01] ( 6, 4.50) int -> rdx single-def
; V03 loc0 [V03,T03] ( 4, 3 ) int -> rbx
;* V04 loc1 [V04 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[System.__Canon]>
;# V05 OutArgs [V05 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V06 tmp1 [V06 ] ( 0, 0 ) long -> zero-ref "spilling helperCall"
;* V07 tmp2 [V07 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[System.__Canon]>
;* V08 tmp3 [V08 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[System.__Canon]>
;* V09 tmp4 [V09 ] ( 0, 0 ) long -> zero-ref "Inlining Arg"
;* V10 tmp5 [V10 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V11 tmp6 [V11,T05] ( 2, 1.50) byref -> r15 single-def "field V04._reference (fldOffset=0x0)" P-INDEP
-; V12 tmp7 [V12,T04] ( 2, 2 ) int -> rax "field V04._length (fldOffset=0x8)" P-INDEP
+; V11 tmp6 [V11,T06] ( 2, 1.50) byref -> r15 single-def "field V04._reference (fldOffset=0x0)" P-INDEP
+; V12 tmp7 [V12,T04] ( 3, 2.50) int -> r14 "field V04._length (fldOffset=0x8)" P-INDEP
;* V13 tmp8 [V13 ] ( 0, 0 ) byref -> zero-ref "field V07._reference (fldOffset=0x0)" P-DEP
;* V14 tmp9 [V14 ] ( 0, 0 ) int -> zero-ref "field V07._length (fldOffset=0x8)" P-DEP
-; V15 tmp10 [V15,T06] ( 2, 1 ) byref -> rax single-def "field V08._reference (fldOffset=0x0)" P-INDEP
-; V16 tmp11 [V16,T07] ( 2, 1 ) int -> rdx "field V08._length (fldOffset=0x8)" P-INDEP
-; V17 tmp12 [V17,T08] ( 2, 1 ) struct (16) [rbp-0x28] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.__Canon]>
+; V15 tmp10 [V15,T07] ( 2, 1 ) byref -> rax single-def "field V08._reference (fldOffset=0x0)" P-INDEP
+; V16 tmp11 [V16,T08] ( 2, 1 ) int -> rdx "field V08._length (fldOffset=0x8)" P-INDEP
+; V17 tmp12 [V17,T09] ( 2, 1 ) struct (16) [rbp-0x30] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.__Canon]>
+; V18 cse0 [V18,T05] ( 3, 2.50) long -> rax "CSE #03: aggressive"
;
-; Lcl frame size = 32
+; Lcl frame size = 24
G_M39622_IG01:
push rbp
push r15
+ push r14
push rbx
- sub rsp, 32
+ sub rsp, 24
lea rbp, [rsp+0x30]
xor eax, eax
- mov qword ptr [rbp-0x28], rax
- mov qword ptr [rbp-0x18], rsi
- ;; size=23 bbWeight=1 PerfScore 6.00
+ mov qword ptr [rbp-0x30], rax
+ mov qword ptr [rbp-0x20], rsi
+ ;; size=25 bbWeight=1 PerfScore 7.00
G_M39622_IG02:
mov ebx, dword ptr [rdi+0x08]
mov r15, bword ptr [rdi+0x10]
- mov eax, dword ptr [rdi+0x18]
- mov ecx, ebx
- mov r8d, edx
- add rcx, r8
- cmp rcx, rax
+ mov r14d, dword ptr [rdi+0x18]
+ mov eax, ebx
+ mov ecx, edx
+ add rax, rcx
+ mov ecx, r14d
+ cmp rax, rcx
jbe SHORT G_M39622_IG05
- ;; size=23 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M39622_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpanWithGrow(int):System.Span`1[System.__Canon]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpanWithGrow(int):System.Span`1[System.__Canon]:this
- mov bword ptr [rbp-0x28], rax
- mov qword ptr [rbp-0x20], rdx
- mov rax, bword ptr [rbp-0x28]
- mov rdx, qword ptr [rbp-0x20]
+ mov bword ptr [rbp-0x30], rax
+ mov qword ptr [rbp-0x28], rdx
+ mov rax, bword ptr [rbp-0x30]
+ mov rdx, qword ptr [rbp-0x28]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M39622_IG04:
- add rsp, 32
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
G_M39622_IG05:
- lea eax, [rbx+rdx]
- mov dword ptr [rdi+0x08], eax
+ lea ecx, [rbx+rdx]
+ mov dword ptr [rdi+0x08], ecx
+ mov ecx, r14d
+ cmp rax, rcx
+ ja SHORT G_M39622_IG07
mov eax, ebx
lea rax, bword ptr [r15+8*rax]
- ;; size=12 bbWeight=0.50 PerfScore 1.12
+ ;; size=20 bbWeight=0.50 PerfScore 1.88
G_M39622_IG06:
- add rsp, 32
+ add rsp, 24
pop rbx
+ pop r14
pop r15
pop rbp
ret
- ;; size=9 bbWeight=0.50 PerfScore 1.38
+ ;; size=11 bbWeight=0.50 PerfScore 1.62
+G_M39622_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 104, prolog size 23, PerfScore 21.50, instruction count 36, allocated bytes for code 104 (MethodHash=372b6539) for method System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpan(int):System.Span`1[System.__Canon]:this (FullOpts)
+; Total bytes of code 134, prolog size 25, PerfScore 24.00, instruction count 46, allocated bytes for code 134 (MethodHash=372b6539) for method System.Collections.Generic.ValueListBuilder`1[System.__Canon]:AppendSpan(int):System.Span`1[System.__Canon]:this (FullOpts)
; ============================================================30 (30.00 % of base) - System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpan(int):System.Span`1[System.Numerics.Vector`1[float]]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpan(int):System.Span`1[System.Numerics.Vector`1[float]]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[System.Numerics.Vector`1[float]]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[System.Numerics.Vector`1[float]]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[System.Numerics.Vector`1[float]]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> rax single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rcx "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> rax single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r15 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x18] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.Numerics.Vector`1[float]]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[System.Numerics.Vector`1[float]]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rcx "CSE #03: aggressive"
;
-; Lcl frame size = 24
+; Lcl frame size = 16
G_M25766_IG01:
push rbp
+ push r15
push rbx
- sub rsp, 24
+ sub rsp, 16
lea rbp, [rsp+0x20]
xor eax, eax
- mov qword ptr [rbp-0x18], rax
- ;; size=17 bbWeight=1 PerfScore 4.00
+ mov qword ptr [rbp-0x20], rax
+ ;; size=19 bbWeight=1 PerfScore 5.00
G_M25766_IG02:
mov ebx, dword ptr [rdi+0x08]
mov rax, bword ptr [rdi+0x10]
- mov ecx, dword ptr [rdi+0x18]
- mov edx, ebx
- mov r8d, esi
- add rdx, r8
- cmp rdx, rcx
+ mov r15d, dword ptr [rdi+0x18]
+ mov ecx, ebx
+ mov edx, esi
+ add rcx, rdx
+ mov edx, r15d
+ cmp rcx, rdx
jbe SHORT G_M25766_IG05
- ;; size=23 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M25766_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpanWithGrow(int):System.Span`1[System.Numerics.Vector`1[float]]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpanWithGrow(int):System.Span`1[System.Numerics.Vector`1[float]]:this
- mov bword ptr [rbp-0x18], rax
- mov qword ptr [rbp-0x10], rdx
- mov rax, bword ptr [rbp-0x18]
- mov rdx, qword ptr [rbp-0x10]
+ mov bword ptr [rbp-0x20], rax
+ mov qword ptr [rbp-0x18], rdx
+ mov rax, bword ptr [rbp-0x20]
+ mov rdx, qword ptr [rbp-0x18]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M25766_IG04:
- add rsp, 24
+ add rsp, 16
pop rbx
+ pop r15
pop rbp
ret
- ;; size=7 bbWeight=0.50 PerfScore 1.12
+ ;; size=9 bbWeight=0.50 PerfScore 1.38
G_M25766_IG05:
lea edx, [rbx+rsi]
mov dword ptr [rdi+0x08], edx
+ mov edx, r15d
+ cmp rcx, rdx
+ ja SHORT G_M25766_IG07
mov edx, ebx
shl rdx, 5
add rax, rdx
mov rdx, rsi
- ;; size=18 bbWeight=0.50 PerfScore 1.38
+ ;; size=26 bbWeight=0.50 PerfScore 2.12
G_M25766_IG06:
- add rsp, 24
+ add rsp, 16
pop rbx
+ pop r15
pop rbp
ret
- ;; size=7 bbWeight=0.50 PerfScore 1.12
+ ;; size=9 bbWeight=0.50 PerfScore 1.38
+G_M25766_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 100, prolog size 17, PerfScore 19.25, instruction count 34, allocated bytes for code 100 (MethodHash=cacc9b59) for method System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpan(int):System.Span`1[System.Numerics.Vector`1[float]]:this (FullOpts)
+; Total bytes of code 130, prolog size 19, PerfScore 21.75, instruction count 44, allocated bytes for code 130 (MethodHash=cacc9b59) for method System.Collections.Generic.ValueListBuilder`1[System.Numerics.Vector`1[float]]:AppendSpan(int):System.Span`1[System.Numerics.Vector`1[float]]:this (FullOpts)
; ============================================================30 (31.25 % of base) - System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpan(int):System.Span`1[ubyte]:this ; Assembly listing for method System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpan(int):System.Span`1[ubyte]:this (FullOpts)
; Emitting BLENDED_CODE for X64 with AVX512 - Unix
; FullOpts code
; optimized code
; rbp based frame
; partially interruptible
; No PGO data
; 0 inlinees with PGO data; 1 single block inlinees; 1 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T00] ( 7, 6 ) byref -> rdi this single-def
; V01 arg1 [V01,T01] ( 6, 4.50) int -> rsi single-def
; V02 loc0 [V02,T02] ( 4, 3 ) int -> rbx
;* V03 loc1 [V03 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op <System.Span`1[ubyte]>
;# V04 OutArgs [V04 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V05 tmp1 [V05 ] ( 0, 0 ) struct (16) zero-ref do-not-enreg[SBR] multireg-ret "Return value temp for multireg return" <System.Span`1[ubyte]>
;* V06 tmp2 [V06 ] ( 0, 0 ) struct (16) zero-ref multireg-ret ld-addr-op "NewObj constructor temp" <System.Span`1[ubyte]>
;* V07 tmp3 [V07 ] ( 0, 0 ) byref -> zero-ref "Inlining Arg"
-; V08 tmp4 [V08,T04] ( 2, 1.50) byref -> rax single-def "field V03._reference (fldOffset=0x0)" P-INDEP
-; V09 tmp5 [V09,T03] ( 2, 2 ) int -> rcx "field V03._length (fldOffset=0x8)" P-INDEP
+; V08 tmp4 [V08,T05] ( 2, 1.50) byref -> rax single-def "field V03._reference (fldOffset=0x0)" P-INDEP
+; V09 tmp5 [V09,T03] ( 3, 2.50) int -> r15 "field V03._length (fldOffset=0x8)" P-INDEP
;* V10 tmp6 [V10 ] ( 0, 0 ) byref -> zero-ref "field V05._reference (fldOffset=0x0)" P-DEP
;* V11 tmp7 [V11 ] ( 0, 0 ) int -> zero-ref "field V05._length (fldOffset=0x8)" P-DEP
-; V12 tmp8 [V12,T05] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
-; V13 tmp9 [V13,T06] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
-; V14 tmp10 [V14,T07] ( 2, 1 ) struct (16) [rbp-0x18] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[ubyte]>
+; V12 tmp8 [V12,T06] ( 2, 1 ) byref -> rax single-def "field V06._reference (fldOffset=0x0)" P-INDEP
+; V13 tmp9 [V13,T07] ( 2, 1 ) int -> rsi "field V06._length (fldOffset=0x8)" P-INDEP
+; V14 tmp10 [V14,T08] ( 2, 1 ) struct (16) [rbp-0x20] do-not-enreg[SR] multireg-ret must-init "Return value temp for multi-reg return (rejected tail call)." <System.Span`1[ubyte]>
+; V15 cse0 [V15,T04] ( 3, 2.50) long -> rcx "CSE #03: aggressive"
;
-; Lcl frame size = 24
+; Lcl frame size = 16
G_M1894_IG01:
push rbp
+ push r15
push rbx
- sub rsp, 24
+ sub rsp, 16
lea rbp, [rsp+0x20]
xor eax, eax
- mov qword ptr [rbp-0x18], rax
- ;; size=17 bbWeight=1 PerfScore 4.00
+ mov qword ptr [rbp-0x20], rax
+ ;; size=19 bbWeight=1 PerfScore 5.00
G_M1894_IG02:
mov ebx, dword ptr [rdi+0x08]
mov rax, bword ptr [rdi+0x10]
- mov ecx, dword ptr [rdi+0x18]
- mov edx, ebx
- mov r8d, esi
- add rdx, r8
- cmp rdx, rcx
+ mov r15d, dword ptr [rdi+0x18]
+ mov ecx, ebx
+ mov edx, esi
+ add rcx, rdx
+ mov edx, r15d
+ cmp rcx, rdx
jbe SHORT G_M1894_IG05
- ;; size=23 bbWeight=1 PerfScore 8.00
+ ;; size=26 bbWeight=1 PerfScore 8.25
G_M1894_IG03:
mov rax, 0xD1FFAB1E ; code for System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpanWithGrow(int):System.Span`1[ubyte]:this
call [rax]System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpanWithGrow(int):System.Span`1[ubyte]:this
- mov bword ptr [rbp-0x18], rax
- mov qword ptr [rbp-0x10], rdx
- mov rax, bword ptr [rbp-0x18]
- mov rdx, qword ptr [rbp-0x10]
+ mov bword ptr [rbp-0x20], rax
+ mov qword ptr [rbp-0x18], rdx
+ mov rax, bword ptr [rbp-0x20]
+ mov rdx, qword ptr [rbp-0x18]
;; size=28 bbWeight=0.50 PerfScore 3.62
G_M1894_IG04:
- add rsp, 24
+ add rsp, 16
pop rbx
+ pop r15
pop rbp
ret
- ;; size=7 bbWeight=0.50 PerfScore 1.12
+ ;; size=9 bbWeight=0.50 PerfScore 1.38
G_M1894_IG05:
lea edx, [rbx+rsi]
mov dword ptr [rdi+0x08], edx
+ mov edx, r15d
+ cmp rcx, rdx
+ ja SHORT G_M1894_IG07
mov edx, ebx
add rax, rdx
mov rdx, rsi
- ;; size=14 bbWeight=0.50 PerfScore 1.12
+ ;; size=22 bbWeight=0.50 PerfScore 1.88
G_M1894_IG06:
- add rsp, 24
+ add rsp, 16
pop rbx
+ pop r15
pop rbp
ret
- ;; size=7 bbWeight=0.50 PerfScore 1.12
+ ;; size=9 bbWeight=0.50 PerfScore 1.38
+G_M1894_IG07:
+ mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
+ int3
+ ;; size=13 bbWeight=0 PerfScore 0.00
-; Total bytes of code 96, prolog size 17, PerfScore 19.00, instruction count 33, allocated bytes for code 96 (MethodHash=6fdff899) for method System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpan(int):System.Span`1[ubyte]:this (FullOpts)
+; Total bytes of code 126, prolog size 19, PerfScore 21.50, instruction count 43, allocated bytes for code 126 (MethodHash=6fdff899) for method System.Collections.Generic.ValueListBuilder`1[ubyte]:AppendSpan(int):System.Span`1[ubyte]:this (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/e4005e339ff846d8e51fce4c8e40ad9e |
Top method improvements-1 (-0.11 % of base) - System.Numerics.Tensors.ReadOnlyTensorSpan`1[double]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this ; Assembly listing for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[double]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (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; 20 single block inlinees; 10 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T11] ( 5, 3.50) byref -> [rbp-0xB8] this single-def
; V01 arg1 [V01,T06] ( 14, 8.50) ref -> r14 class-hnd single-def <System.Array>
;* V02 arg2 [V02 ] ( 0, 0 ) struct (16) zero-ref multireg-arg ld-addr-op single-def <System.ReadOnlySpan`1[System.Buffers.NIndex]>
;* V03 arg3 [V03 ] ( 0, 0 ) struct (16) zero-ref multireg-arg multireg-ret ld-addr-op <System.ReadOnlySpan`1[long]>
;* V04 arg4 [V04 ] ( 0, 0 ) struct (16) zero-ref multireg-arg ld-addr-op single-def <System.ReadOnlySpan`1[long]>
;* V05 loc0 [V05 ] ( 0, 0 ) long -> zero-ref single-def
; V06 loc1 [V06,T21] ( 4, 2 ) long -> [rbp-0x30] spill-single-def
;* V07 loc2 [V07 ] ( 0, 0 ) long -> zero-ref single-def
;* V08 loc3 [V08 ] ( 0, 0 ) struct (16) zero-ref multireg-arg <System.Span`1[long]>
;* V09 loc4 [V09 ] ( 0, 0 ) int -> zero-ref single-def
;# V10 OutArgs [V10 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V11 tmp1 [V11 ] ( 0, 0 ) struct (16) zero-ref multireg-arg <System.ReadOnlySpan`1[long]>
; V12 tmp2 [V12 ] ( 2, 2 ) struct (104) [rbp-0x98] do-not-enreg[XS] must-init addr-exposed ld-addr-op "NewObj constructor temp" <System.Numerics.Tensors.TensorShape>
; V13 tmp3 [V13,T16] ( 3, 3 ) ref -> rax class-hnd exact single-def "NewArr temp" <long[]>
;* V14 tmp4 [V14 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.Span`1[long]>
; V15 tmp5 [V15,T41] ( 2, 1 ) int -> rdi "Inline return value spill temp"
;* V16 tmp6 [V16 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V17 tmp7 [V17,T18] ( 3, 3 ) int -> rdi "impAppendStmt"
; V18 tmp8 [V18,T26] ( 2, 2 ) byref -> rsi single-def "Inlining Arg"
; V19 tmp9 [V19,T22] ( 4, 2 ) int -> rsi "Inline return value spill temp"
;* V20 tmp10 [V20 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V21 tmp11 [V21,T19] ( 3, 3 ) int -> rsi "impAppendStmt"
;* V22 tmp12 [V22,T45] ( 0, 0 ) byref -> zero-ref single-def "Inlining Arg"
; V23 tmp13 [V23,T28] ( 2, 2 ) long -> rdi "Inlining Arg"
; V24 tmp14 [V24,T42] ( 2, 1 ) int -> rsi "Inline return value spill temp"
;* V25 tmp15 [V25 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V26 tmp16 [V26,T29] ( 2, 2 ) int -> rsi "impAppendStmt"
;* V27 tmp17 [V27,T46] ( 0, 0 ) byref -> zero-ref single-def "Inlining Arg"
;* V28 tmp18 [V28 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.Span`1[long]>
; V29 tmp19 [V29,T23] ( 4, 2 ) long -> [rbp-0xA0] "Inline return value spill temp"
;* V30 tmp20 [V30 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
; V31 tmp21 [V31,T07] ( 5, 9.25) long -> rdi "Inline stloc first use temp"
;* V32 tmp22 [V32,T52] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
; V33 tmp23 [V33,T17] ( 3, 3 ) ref -> [rbp-0xC0] class-hnd exact spill-single-def "Inlining Arg" <System.RuntimeType>
; V34 tmp24 [V34,T38] ( 2, 1 ) ref -> rax class-hnd "Inline return value spill temp" <System.RuntimeType>
; V35 tmp25 [V35,T24] ( 4, 2 ) long -> [rbp-0xA8] spill-single-def "Inline stloc first use temp"
;* V36 tmp26 [V36 ] ( 0, 0 ) ref -> zero-ref class-hnd exact single-def "impAppendStmt" <System.RuntimeType>
; V37 tmp27 [V37,T27] ( 2, 2 ) ref -> rax class-hnd exact single-def "dup spill" <System.RuntimeType>
; V38 tmp28 [V38,T20] ( 4, 2 ) ref -> rax
; V39 tmp29 [V39,T12] ( 4, 4 ) ref -> rax class-hnd single-def "Inlining Arg" <long[]>
;* V40 tmp30 [V40 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.ReadOnlySpan`1[long]>
; V41 tmp31 [V41,T08] ( 4, 9 ) long -> rsi "Inline stloc first use temp"
; V42 tmp32 [V42,T00] ( 6, 20.50) int -> rdx "Inline stloc first use temp"
;* V43 tmp33 [V43 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
;* V44 tmp34 [V44 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
; V45 tmp35 [V45,T14] ( 2, 1.50) byref -> [rbp-0xC8] single-def "field V02._reference (fldOffset=0x0)" P-INDEP
; V46 tmp36 [V46,T13] ( 3, 2 ) int -> [rbp-0xAC] single-def "field V02._length (fldOffset=0x8)" P-INDEP
; V47 tmp37 [V47,T03] ( 7, 11 ) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
; V48 tmp38 [V48,T02] ( 10, 13.50) int -> rbx single-def "field V03._length (fldOffset=0x8)" P-INDEP
; V49 tmp39 [V49,T10] ( 5, 6 ) byref -> r12 "field V04._reference (fldOffset=0x0)" P-INDEP
; V50 tmp40 [V50,T09] ( 6, 6.50) int -> r13 "field V04._length (fldOffset=0x8)" P-INDEP
; V51 tmp41 [V51,T31] ( 3, 1.50) byref -> rdi "field V08._reference (fldOffset=0x0)" P-INDEP
; V52 tmp42 [V52,T34] ( 3, 1.50) int -> rsi "field V08._length (fldOffset=0x8)" P-INDEP
; V53 tmp43 [V53,T32] ( 3, 1.50) byref -> rdi "field V11._reference (fldOffset=0x0)" P-INDEP
; V54 tmp44 [V54,T35] ( 3, 1.50) int -> rsi "field V11._length (fldOffset=0x8)" P-INDEP
; V55 tmp45 [V55,T39] ( 2, 1 ) byref -> rdi single-def "field V14._reference (fldOffset=0x0)" P-INDEP
; V56 tmp46 [V56,T43] ( 2, 1 ) int -> rsi single-def "field V14._length (fldOffset=0x8)" P-INDEP
; V57 tmp47 [V57,T40] ( 2, 1 ) byref -> rdi "field V28._reference (fldOffset=0x0)" P-INDEP
; V58 tmp48 [V58,T44] ( 2, 1 ) int -> rsi "field V28._length (fldOffset=0x8)" P-INDEP
;* V59 tmp49 [V59,T47] ( 0, 0 ) byref -> zero-ref single-def "field V30._reference (fldOffset=0x0)" P-INDEP
; V60 tmp50 [V60,T30] ( 2, 2 ) int -> rdi single-def "field V30._length (fldOffset=0x8)" P-INDEP
; V61 tmp51 [V61,T33] ( 3, 1.50) byref -> rdi "field V40._reference (fldOffset=0x0)" P-INDEP
; V62 tmp52 [V62,T36] ( 3, 1.50) int -> rsi "field V40._length (fldOffset=0x8)" P-INDEP
;* V63 tmp53 [V63,T48] ( 0, 0 ) byref -> zero-ref single-def "field V43._reference (fldOffset=0x0)" P-INDEP
;* V64 tmp54 [V64,T50] ( 0, 0 ) int -> zero-ref single-def "field V43._length (fldOffset=0x8)" P-INDEP
;* V65 tmp55 [V65,T49] ( 0, 0 ) byref -> zero-ref single-def "field V44._reference (fldOffset=0x0)" P-INDEP
;* V66 tmp56 [V66,T51] ( 0, 0 ) int -> zero-ref single-def "field V44._length (fldOffset=0x8)" P-INDEP
; V67 GsCookie [V67 ] ( 1, 1 ) long -> [rbp-0xD0] do-not-enreg[X] addr-exposed "GSSecurityCookie"
; V68 cse0 [V68,T25] ( 4, 2 ) int -> rsi "CSE #04: conservative"
; V69 cse1 [V69,T04] ( 3, 12 ) long -> r8 "CSE #07: aggressive"
-; V70 cse2 [V70,T37] ( 3, 1.50) int -> rdx "CSE #10: conservative"
+; V70 cse2 [V70,T37] ( 3, 1.50) long -> rdx "CSE #11: conservative"
; V71 cse3 [V71,T15] ( 4, 2 ) long -> rsi multi-def "CSE #01: moderate"
; V72 cse4 [V72,T05] ( 3, 12 ) long -> rcx "CSE #08: aggressive"
; V73 rat0 [V73,T01] ( 5, 16.25) long -> rsi "Widened IV V32"
;
; Lcl frame size = 168
G_M55583_IG01:
push rbp
push r15
push r14
push r13
push r12
push rbx
sub rsp, 168
lea rbp, [rsp+0xD0]
xor eax, eax
mov qword ptr [rbp-0x98], rax
vxorps xmm8, xmm8, xmm8
vmovdqu32 zmmword ptr [rbp-0x90], zmm8
vmovdqa xmmword ptr [rbp-0x50], xmm8
vmovdqa xmmword ptr [rbp-0x40], xmm8
mov qword ptr [rbp-0xD0], 0xD1FFAB1E
mov bword ptr [rbp-0xB8], rdi
mov bword ptr [rbp-0xC8], rdx
mov dword ptr [rbp-0xAC], ecx
mov r14, rsi
mov r15, r8
mov ebx, r9d
mov r12, bword ptr [rbp+0x10]
mov r13d, dword ptr [rbp+0x18]
;; size=107 bbWeight=1 PerfScore 21.08
G_M55583_IG02:
test ebx, ebx
jne G_M55583_IG08
;; size=8 bbWeight=1 PerfScore 1.25
G_M55583_IG03:
test r14, r14
je G_M55583_IG08
mov rsi, qword ptr [r14]
mov esi, dword ptr [rsi+0x04]
add esi, -24
shr esi, 3
mov edi, esi
test edi, edi
je SHORT G_M55583_IG04
cmp edi, 8
jle SHORT G_M55583_IG04
movsxd rsi, esi
mov rdi, 0xD1FFAB1E ; long[]
call CORINFO_HELP_NEWARR_1_VC
lea rdi, bword ptr [rax+0x10]
mov esi, dword ptr [rax+0x08]
jmp SHORT G_M55583_IG07
align [14 bytes for IG10]
;; size=73 bbWeight=0.50 PerfScore 7.38
G_M55583_IG04:
mov edi, 1
test esi, esi
cmove esi, edi
mov edi, esi
mov r8d, 8
mov rax, rdi
mul rdx:rax, r8
jb G_M55583_IG30
mov rdi, rax
test rdi, rdi
je SHORT G_M55583_IG06
add rdi, 15
shr rdi, 4
;; size=46 bbWeight=0.50 PerfScore 3.88
G_M55583_IG05:
push 0
push 0
dec rdi
jne SHORT G_M55583_IG05
lea rdi, [rsp]
;; size=13 bbWeight=0.50 PerfScore 1.88
G_M55583_IG06:
test esi, esi
jl G_M55583_IG28
;; size=8 bbWeight=0.50 PerfScore 0.62
G_M55583_IG07:
mov rdx, r14
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:FillLengths(System.Span`1[long],System.Array):System.ReadOnlySpan`1[long]
call [rax]System.Numerics.Tensors.TensorSpanHelpers:FillLengths(System.Span`1[long],System.Array):System.ReadOnlySpan`1[long]
mov r15, rax
mov ebx, edx
;; size=20 bbWeight=0.50 PerfScore 2.00
G_M55583_IG08:
mov edi, ebx
test edi, edi
je SHORT G_M55583_IG12
;; size=6 bbWeight=1 PerfScore 1.50
G_M55583_IG09:
mov edi, 1
xor esi, esi
;; size=7 bbWeight=0.25 PerfScore 0.12
G_M55583_IG10:
mov r8, qword ptr [r15+8*rsi]
test r8, r8
jl G_M55583_IG27
imul rdi, r8
inc esi
cmp esi, ebx
jl SHORT G_M55583_IG10
;; size=23 bbWeight=4 PerfScore 27.00
G_M55583_IG11:
test rdi, rdi
jl G_M55583_IG27
mov rsi, rdi
mov qword ptr [rbp-0xA0], rsi
jmp SHORT G_M55583_IG13
align [0 bytes for IG20]
;; size=21 bbWeight=0.50 PerfScore 2.25
G_M55583_IG12:
xor esi, esi
mov qword ptr [rbp-0xA0], rsi
;; size=9 bbWeight=0.50 PerfScore 0.62
G_M55583_IG13:
test r14, r14
je G_M55583_IG22
;; size=9 bbWeight=1 PerfScore 1.25
G_M55583_IG14:
mov rdi, r14
mov r8, 0xD1FFAB1E ; code for System.Object:GetType():System.Type:this
call [r8]System.Object:GetType():System.Type:this
mov gword ptr [rbp-0xC0], rax
mov rdi, qword ptr [rax+0x18]
call System.RuntimeTypeHandle:GetElementTypeHandle(long):long
mov qword ptr [rbp-0xA8], rax
test rax, rax
je G_M55583_IG26
mov rdi, rax
call System.RuntimeTypeHandle:GetRuntimeTypeFromHandleIfExists(long):System.RuntimeType
test rax, rax
jne SHORT G_M55583_IG15
mov rdi, qword ptr [rbp-0xA8]
mov rax, 0xD1FFAB1E ; code for System.RuntimeTypeHandle:GetRuntimeTypeFromHandleSlow(long):System.RuntimeType
call [rax]System.RuntimeTypeHandle:GetRuntimeTypeFromHandleSlow(long):System.RuntimeType
;; size=80 bbWeight=0.50 PerfScore 8.25
G_M55583_IG15:
mov rdi, 0xD1FFAB1E ; 'System.Double'
cmp rax, rdi
jne G_M55583_IG26
test r13d, r13d
jne SHORT G_M55583_IG18
mov rdi, r15
mov esi, ebx
mov rdx, qword ptr [rbp-0xA0]
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:CalculateStrides(System.ReadOnlySpan`1[long],long):long[]
call [rax]System.Numerics.Tensors.TensorSpanHelpers:CalculateStrides(System.ReadOnlySpan`1[long],long):long[]
test rax, rax
je SHORT G_M55583_IG16
lea rdi, bword ptr [rax+0x10]
mov esi, dword ptr [rax+0x08]
jmp SHORT G_M55583_IG17
;; size=62 bbWeight=0.50 PerfScore 6.62
G_M55583_IG16:
xor rdi, rdi
xor esi, esi
;; size=4 bbWeight=0.50 PerfScore 0.25
G_M55583_IG17:
jmp SHORT G_M55583_IG19
;; size=2 bbWeight=0.50 PerfScore 1.00
G_M55583_IG18:
mov rdi, r12
mov esi, r13d
;; size=6 bbWeight=0.50 PerfScore 0.25
G_M55583_IG19:
mov r12, rdi
mov r13d, esi
mov rdi, r12
mov esi, r13d
mov rdx, r15
mov ecx, ebx
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:ValidateStrides(System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long])
call [rax]System.Numerics.Tensors.TensorSpanHelpers:ValidateStrides(System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long])
mov rsi, bword ptr [rbp-0xC8]
mov edx, dword ptr [rbp-0xAC]
mov rdi, r14
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:ComputeStartOffsetSystemArray(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex]):long
call [rax]System.Numerics.Tensors.TensorSpanHelpers:ComputeStartOffsetSystemArray(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex]):long
mov qword ptr [rbp-0x30], rax
xor esi, esi
xor edx, edx
test ebx, ebx
jle SHORT G_M55583_IG21
;; size=69 bbWeight=0.50 PerfScore 6.50
G_M55583_IG20:
cmp edx, r13d
jae G_M55583_IG29
mov ecx, edx
mov r8, qword ptr [r12+8*rcx]
mov rcx, qword ptr [r15+8*rcx]
dec rcx
imul rcx, r8
add rsi, rcx
inc edx
cmp edx, ebx
jl SHORT G_M55583_IG20
;; size=35 bbWeight=4 PerfScore 38.00
G_M55583_IG21:
mov edx, eax
mov esi, esi
add rsi, rdx
mov edx, dword ptr [r14+0x08]
test edx, edx
jl G_M55583_IG30
- mov ecx, edx
- cmp rsi, rcx
+ cmp rsi, rdx
ja G_M55583_IG27
- mov esi, edx
+ mov rsi, rdx
sub rsi, rax
mov rdx, r15
mov ecx, ebx
mov r8, r12
mov r9d, r13d
lea rdi, [rbp-0x98]
mov r10, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorShape:.ctor(long,System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this
call [r10]System.Numerics.Tensors.TensorShape:.ctor(long,System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this
mov rbx, bword ptr [rbp-0xB8]
lea rdi, bword ptr [rbx+0x08]
lea rsi, bword ptr [rbp-0x98]
call CORINFO_HELP_ASSIGN_BYREF
mov ecx, 12
rep movsq
lea rax, bword ptr [r14+0x08]
mov rsi, qword ptr [r14]
mov ecx, dword ptr [rsi+0x04]
add rcx, -16
add rax, rcx
mov ecx, dword ptr [rbp-0x30]
shl rcx, 3
add rax, rcx
mov bword ptr [rbx], rax
jmp SHORT G_M55583_IG23
- ;; size=129 bbWeight=0.50 PerfScore 24.38
+ ;; size=128 bbWeight=0.50 PerfScore 24.25
G_M55583_IG22:
cmp dword ptr [rbp-0xAC], 0
jne SHORT G_M55583_IG27
cmp qword ptr [rbp-0xA0], 0
jne SHORT G_M55583_IG27
xor ecx, ecx
mov rax, bword ptr [rbp-0xB8]
mov qword ptr [rax], rcx
mov qword ptr [rax+0x08], rcx
vxorps xmm0, xmm0, xmm0
vmovdqu32 zmmword ptr [rax+0x10], zmm0
vmovdqu ymmword ptr [rax+0x50], ymm0
;; size=54 bbWeight=0.50 PerfScore 6.79
G_M55583_IG23:
cmp qword ptr [rbp-0xD0], 0xD1FFAB1E
je SHORT G_M55583_IG24
call CORINFO_HELP_FAIL_FAST
;; size=18 bbWeight=1 PerfScore 4.00
G_M55583_IG24:
nop
;; size=1 bbWeight=1 PerfScore 0.25
G_M55583_IG25:
lea rsp, [rbp-0x28]
pop rbx
pop r12
pop r13
pop r14
pop r15
pop rbp
ret
;; size=15 bbWeight=1 PerfScore 4.50
G_M55583_IG26:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArrayTypeMismatchException()
call [rax]System.ThrowHelper:ThrowArrayTypeMismatchException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M55583_IG27:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M55583_IG28:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M55583_IG29:
call CORINFO_HELP_RNGCHKFAIL
int3
;; size=6 bbWeight=0 PerfScore 0.00
G_M55583_IG30:
call CORINFO_HELP_OVERFLOW
int3
;; size=6 bbWeight=0 PerfScore 0.00
-; Total bytes of code 876, prolog size 107, PerfScore 171.63, instruction count 225, allocated bytes for code 876 (MethodHash=a96926e0) for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[double]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (FullOpts)
+; Total bytes of code 875, prolog size 107, PerfScore 171.50, instruction count 224, allocated bytes for code 875 (MethodHash=a96926e0) for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[double]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (FullOpts)
; ============================================================-1 (-0.11 % of base) - System.Numerics.Tensors.ReadOnlyTensorSpan`1[int]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this ; Assembly listing for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[int]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (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; 20 single block inlinees; 10 inlinees without PGO data
; Final local variable assignments
;
; V00 this [V00,T11] ( 5, 3.50) byref -> [rbp-0xB8] this single-def
; V01 arg1 [V01,T06] ( 14, 8.50) ref -> r14 class-hnd single-def <System.Array>
;* V02 arg2 [V02 ] ( 0, 0 ) struct (16) zero-ref multireg-arg ld-addr-op single-def <System.ReadOnlySpan`1[System.Buffers.NIndex]>
;* V03 arg3 [V03 ] ( 0, 0 ) struct (16) zero-ref multireg-arg multireg-ret ld-addr-op <System.ReadOnlySpan`1[long]>
;* V04 arg4 [V04 ] ( 0, 0 ) struct (16) zero-ref multireg-arg ld-addr-op single-def <System.ReadOnlySpan`1[long]>
;* V05 loc0 [V05 ] ( 0, 0 ) long -> zero-ref single-def
; V06 loc1 [V06,T21] ( 4, 2 ) long -> [rbp-0x30] spill-single-def
;* V07 loc2 [V07 ] ( 0, 0 ) long -> zero-ref single-def
;* V08 loc3 [V08 ] ( 0, 0 ) struct (16) zero-ref multireg-arg <System.Span`1[long]>
;* V09 loc4 [V09 ] ( 0, 0 ) int -> zero-ref single-def
;# V10 OutArgs [V10 ] ( 1, 1 ) struct ( 0) [rsp+0x00] do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;* V11 tmp1 [V11 ] ( 0, 0 ) struct (16) zero-ref multireg-arg <System.ReadOnlySpan`1[long]>
; V12 tmp2 [V12 ] ( 2, 2 ) struct (104) [rbp-0x98] do-not-enreg[XS] must-init addr-exposed ld-addr-op "NewObj constructor temp" <System.Numerics.Tensors.TensorShape>
; V13 tmp3 [V13,T16] ( 3, 3 ) ref -> rax class-hnd exact single-def "NewArr temp" <long[]>
;* V14 tmp4 [V14 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.Span`1[long]>
; V15 tmp5 [V15,T41] ( 2, 1 ) int -> rdi "Inline return value spill temp"
;* V16 tmp6 [V16 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V17 tmp7 [V17,T18] ( 3, 3 ) int -> rdi "impAppendStmt"
; V18 tmp8 [V18,T26] ( 2, 2 ) byref -> rsi single-def "Inlining Arg"
; V19 tmp9 [V19,T22] ( 4, 2 ) int -> rsi "Inline return value spill temp"
;* V20 tmp10 [V20 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V21 tmp11 [V21,T19] ( 3, 3 ) int -> rsi "impAppendStmt"
;* V22 tmp12 [V22,T45] ( 0, 0 ) byref -> zero-ref single-def "Inlining Arg"
; V23 tmp13 [V23,T28] ( 2, 2 ) long -> rdi "Inlining Arg"
; V24 tmp14 [V24,T42] ( 2, 1 ) int -> rsi "Inline return value spill temp"
;* V25 tmp15 [V25 ] ( 0, 0 ) int -> zero-ref single-def "Inline stloc first use temp"
; V26 tmp16 [V26,T29] ( 2, 2 ) int -> rsi "impAppendStmt"
;* V27 tmp17 [V27,T46] ( 0, 0 ) byref -> zero-ref single-def "Inlining Arg"
;* V28 tmp18 [V28 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.Span`1[long]>
; V29 tmp19 [V29,T23] ( 4, 2 ) long -> [rbp-0xA0] "Inline return value spill temp"
;* V30 tmp20 [V30 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
; V31 tmp21 [V31,T07] ( 5, 9.25) long -> rdi "Inline stloc first use temp"
;* V32 tmp22 [V32,T52] ( 0, 0 ) int -> zero-ref "Inline stloc first use temp"
; V33 tmp23 [V33,T17] ( 3, 3 ) ref -> [rbp-0xC0] class-hnd exact spill-single-def "Inlining Arg" <System.RuntimeType>
; V34 tmp24 [V34,T38] ( 2, 1 ) ref -> rax class-hnd "Inline return value spill temp" <System.RuntimeType>
; V35 tmp25 [V35,T24] ( 4, 2 ) long -> [rbp-0xA8] spill-single-def "Inline stloc first use temp"
;* V36 tmp26 [V36 ] ( 0, 0 ) ref -> zero-ref class-hnd exact single-def "impAppendStmt" <System.RuntimeType>
; V37 tmp27 [V37,T27] ( 2, 2 ) ref -> rax class-hnd exact single-def "dup spill" <System.RuntimeType>
; V38 tmp28 [V38,T20] ( 4, 2 ) ref -> rax
; V39 tmp29 [V39,T12] ( 4, 4 ) ref -> rax class-hnd single-def "Inlining Arg" <long[]>
;* V40 tmp30 [V40 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "NewObj constructor temp" <System.ReadOnlySpan`1[long]>
; V41 tmp31 [V41,T08] ( 4, 9 ) long -> rsi "Inline stloc first use temp"
; V42 tmp32 [V42,T00] ( 6, 20.50) int -> rdx "Inline stloc first use temp"
;* V43 tmp33 [V43 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
;* V44 tmp34 [V44 ] ( 0, 0 ) struct (16) zero-ref ld-addr-op "Inlining Arg" <System.ReadOnlySpan`1[long]>
; V45 tmp35 [V45,T14] ( 2, 1.50) byref -> [rbp-0xC8] single-def "field V02._reference (fldOffset=0x0)" P-INDEP
; V46 tmp36 [V46,T13] ( 3, 2 ) int -> [rbp-0xAC] single-def "field V02._length (fldOffset=0x8)" P-INDEP
; V47 tmp37 [V47,T03] ( 7, 11 ) byref -> r15 single-def "field V03._reference (fldOffset=0x0)" P-INDEP
; V48 tmp38 [V48,T02] ( 10, 13.50) int -> rbx single-def "field V03._length (fldOffset=0x8)" P-INDEP
; V49 tmp39 [V49,T10] ( 5, 6 ) byref -> r12 "field V04._reference (fldOffset=0x0)" P-INDEP
; V50 tmp40 [V50,T09] ( 6, 6.50) int -> r13 "field V04._length (fldOffset=0x8)" P-INDEP
; V51 tmp41 [V51,T31] ( 3, 1.50) byref -> rdi "field V08._reference (fldOffset=0x0)" P-INDEP
; V52 tmp42 [V52,T34] ( 3, 1.50) int -> rsi "field V08._length (fldOffset=0x8)" P-INDEP
; V53 tmp43 [V53,T32] ( 3, 1.50) byref -> rdi "field V11._reference (fldOffset=0x0)" P-INDEP
; V54 tmp44 [V54,T35] ( 3, 1.50) int -> rsi "field V11._length (fldOffset=0x8)" P-INDEP
; V55 tmp45 [V55,T39] ( 2, 1 ) byref -> rdi single-def "field V14._reference (fldOffset=0x0)" P-INDEP
; V56 tmp46 [V56,T43] ( 2, 1 ) int -> rsi single-def "field V14._length (fldOffset=0x8)" P-INDEP
; V57 tmp47 [V57,T40] ( 2, 1 ) byref -> rdi "field V28._reference (fldOffset=0x0)" P-INDEP
; V58 tmp48 [V58,T44] ( 2, 1 ) int -> rsi "field V28._length (fldOffset=0x8)" P-INDEP
;* V59 tmp49 [V59,T47] ( 0, 0 ) byref -> zero-ref single-def "field V30._reference (fldOffset=0x0)" P-INDEP
; V60 tmp50 [V60,T30] ( 2, 2 ) int -> rdi single-def "field V30._length (fldOffset=0x8)" P-INDEP
; V61 tmp51 [V61,T33] ( 3, 1.50) byref -> rdi "field V40._reference (fldOffset=0x0)" P-INDEP
; V62 tmp52 [V62,T36] ( 3, 1.50) int -> rsi "field V40._length (fldOffset=0x8)" P-INDEP
;* V63 tmp53 [V63,T48] ( 0, 0 ) byref -> zero-ref single-def "field V43._reference (fldOffset=0x0)" P-INDEP
;* V64 tmp54 [V64,T50] ( 0, 0 ) int -> zero-ref single-def "field V43._length (fldOffset=0x8)" P-INDEP
;* V65 tmp55 [V65,T49] ( 0, 0 ) byref -> zero-ref single-def "field V44._reference (fldOffset=0x0)" P-INDEP
;* V66 tmp56 [V66,T51] ( 0, 0 ) int -> zero-ref single-def "field V44._length (fldOffset=0x8)" P-INDEP
; V67 GsCookie [V67 ] ( 1, 1 ) long -> [rbp-0xD0] do-not-enreg[X] addr-exposed "GSSecurityCookie"
; V68 cse0 [V68,T25] ( 4, 2 ) int -> rsi "CSE #04: conservative"
; V69 cse1 [V69,T04] ( 3, 12 ) long -> r8 "CSE #07: aggressive"
-; V70 cse2 [V70,T37] ( 3, 1.50) int -> rdx "CSE #10: conservative"
+; V70 cse2 [V70,T37] ( 3, 1.50) long -> rdx "CSE #11: conservative"
; V71 cse3 [V71,T15] ( 4, 2 ) long -> rsi multi-def "CSE #01: moderate"
; V72 cse4 [V72,T05] ( 3, 12 ) long -> rcx "CSE #08: aggressive"
; V73 rat0 [V73,T01] ( 5, 16.25) long -> rsi "Widened IV V32"
;
; Lcl frame size = 168
G_M51193_IG01:
push rbp
push r15
push r14
push r13
push r12
push rbx
sub rsp, 168
lea rbp, [rsp+0xD0]
xor eax, eax
mov qword ptr [rbp-0x98], rax
vxorps xmm8, xmm8, xmm8
vmovdqu32 zmmword ptr [rbp-0x90], zmm8
vmovdqa xmmword ptr [rbp-0x50], xmm8
vmovdqa xmmword ptr [rbp-0x40], xmm8
mov qword ptr [rbp-0xD0], 0xD1FFAB1E
mov bword ptr [rbp-0xB8], rdi
mov bword ptr [rbp-0xC8], rdx
mov dword ptr [rbp-0xAC], ecx
mov r14, rsi
mov r15, r8
mov ebx, r9d
mov r12, bword ptr [rbp+0x10]
mov r13d, dword ptr [rbp+0x18]
;; size=107 bbWeight=1 PerfScore 21.08
G_M51193_IG02:
test ebx, ebx
jne G_M51193_IG08
;; size=8 bbWeight=1 PerfScore 1.25
G_M51193_IG03:
test r14, r14
je G_M51193_IG08
mov rsi, qword ptr [r14]
mov esi, dword ptr [rsi+0x04]
add esi, -24
shr esi, 3
mov edi, esi
test edi, edi
je SHORT G_M51193_IG04
cmp edi, 8
jle SHORT G_M51193_IG04
movsxd rsi, esi
mov rdi, 0xD1FFAB1E ; long[]
call CORINFO_HELP_NEWARR_1_VC
lea rdi, bword ptr [rax+0x10]
mov esi, dword ptr [rax+0x08]
jmp SHORT G_M51193_IG07
align [14 bytes for IG10]
;; size=73 bbWeight=0.50 PerfScore 7.38
G_M51193_IG04:
mov edi, 1
test esi, esi
cmove esi, edi
mov edi, esi
mov r8d, 8
mov rax, rdi
mul rdx:rax, r8
jb G_M51193_IG30
mov rdi, rax
test rdi, rdi
je SHORT G_M51193_IG06
add rdi, 15
shr rdi, 4
;; size=46 bbWeight=0.50 PerfScore 3.88
G_M51193_IG05:
push 0
push 0
dec rdi
jne SHORT G_M51193_IG05
lea rdi, [rsp]
;; size=13 bbWeight=0.50 PerfScore 1.88
G_M51193_IG06:
test esi, esi
jl G_M51193_IG28
;; size=8 bbWeight=0.50 PerfScore 0.62
G_M51193_IG07:
mov rdx, r14
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:FillLengths(System.Span`1[long],System.Array):System.ReadOnlySpan`1[long]
call [rax]System.Numerics.Tensors.TensorSpanHelpers:FillLengths(System.Span`1[long],System.Array):System.ReadOnlySpan`1[long]
mov r15, rax
mov ebx, edx
;; size=20 bbWeight=0.50 PerfScore 2.00
G_M51193_IG08:
mov edi, ebx
test edi, edi
je SHORT G_M51193_IG12
;; size=6 bbWeight=1 PerfScore 1.50
G_M51193_IG09:
mov edi, 1
xor esi, esi
;; size=7 bbWeight=0.25 PerfScore 0.12
G_M51193_IG10:
mov r8, qword ptr [r15+8*rsi]
test r8, r8
jl G_M51193_IG27
imul rdi, r8
inc esi
cmp esi, ebx
jl SHORT G_M51193_IG10
;; size=23 bbWeight=4 PerfScore 27.00
G_M51193_IG11:
test rdi, rdi
jl G_M51193_IG27
mov rsi, rdi
mov qword ptr [rbp-0xA0], rsi
jmp SHORT G_M51193_IG13
align [0 bytes for IG20]
;; size=21 bbWeight=0.50 PerfScore 2.25
G_M51193_IG12:
xor esi, esi
mov qword ptr [rbp-0xA0], rsi
;; size=9 bbWeight=0.50 PerfScore 0.62
G_M51193_IG13:
test r14, r14
je G_M51193_IG22
;; size=9 bbWeight=1 PerfScore 1.25
G_M51193_IG14:
mov rdi, r14
mov r8, 0xD1FFAB1E ; code for System.Object:GetType():System.Type:this
call [r8]System.Object:GetType():System.Type:this
mov gword ptr [rbp-0xC0], rax
mov rdi, qword ptr [rax+0x18]
call System.RuntimeTypeHandle:GetElementTypeHandle(long):long
mov qword ptr [rbp-0xA8], rax
test rax, rax
je G_M51193_IG26
mov rdi, rax
call System.RuntimeTypeHandle:GetRuntimeTypeFromHandleIfExists(long):System.RuntimeType
test rax, rax
jne SHORT G_M51193_IG15
mov rdi, qword ptr [rbp-0xA8]
mov rax, 0xD1FFAB1E ; code for System.RuntimeTypeHandle:GetRuntimeTypeFromHandleSlow(long):System.RuntimeType
call [rax]System.RuntimeTypeHandle:GetRuntimeTypeFromHandleSlow(long):System.RuntimeType
;; size=80 bbWeight=0.50 PerfScore 8.25
G_M51193_IG15:
mov rdi, 0xD1FFAB1E ; 'System.Int32'
cmp rax, rdi
jne G_M51193_IG26
test r13d, r13d
jne SHORT G_M51193_IG18
mov rdi, r15
mov esi, ebx
mov rdx, qword ptr [rbp-0xA0]
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:CalculateStrides(System.ReadOnlySpan`1[long],long):long[]
call [rax]System.Numerics.Tensors.TensorSpanHelpers:CalculateStrides(System.ReadOnlySpan`1[long],long):long[]
test rax, rax
je SHORT G_M51193_IG16
lea rdi, bword ptr [rax+0x10]
mov esi, dword ptr [rax+0x08]
jmp SHORT G_M51193_IG17
;; size=62 bbWeight=0.50 PerfScore 6.62
G_M51193_IG16:
xor rdi, rdi
xor esi, esi
;; size=4 bbWeight=0.50 PerfScore 0.25
G_M51193_IG17:
jmp SHORT G_M51193_IG19
;; size=2 bbWeight=0.50 PerfScore 1.00
G_M51193_IG18:
mov rdi, r12
mov esi, r13d
;; size=6 bbWeight=0.50 PerfScore 0.25
G_M51193_IG19:
mov r12, rdi
mov r13d, esi
mov rdi, r12
mov esi, r13d
mov rdx, r15
mov ecx, ebx
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:ValidateStrides(System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long])
call [rax]System.Numerics.Tensors.TensorSpanHelpers:ValidateStrides(System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long])
mov rsi, bword ptr [rbp-0xC8]
mov edx, dword ptr [rbp-0xAC]
mov rdi, r14
mov rax, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorSpanHelpers:ComputeStartOffsetSystemArray(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex]):long
call [rax]System.Numerics.Tensors.TensorSpanHelpers:ComputeStartOffsetSystemArray(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex]):long
mov qword ptr [rbp-0x30], rax
xor esi, esi
xor edx, edx
test ebx, ebx
jle SHORT G_M51193_IG21
;; size=69 bbWeight=0.50 PerfScore 6.50
G_M51193_IG20:
cmp edx, r13d
jae G_M51193_IG29
mov ecx, edx
mov r8, qword ptr [r12+8*rcx]
mov rcx, qword ptr [r15+8*rcx]
dec rcx
imul rcx, r8
add rsi, rcx
inc edx
cmp edx, ebx
jl SHORT G_M51193_IG20
;; size=35 bbWeight=4 PerfScore 38.00
G_M51193_IG21:
mov edx, eax
mov esi, esi
add rsi, rdx
mov edx, dword ptr [r14+0x08]
test edx, edx
jl G_M51193_IG30
- mov ecx, edx
- cmp rsi, rcx
+ cmp rsi, rdx
ja G_M51193_IG27
- mov esi, edx
+ mov rsi, rdx
sub rsi, rax
mov rdx, r15
mov ecx, ebx
mov r8, r12
mov r9d, r13d
lea rdi, [rbp-0x98]
mov r10, 0xD1FFAB1E ; code for System.Numerics.Tensors.TensorShape:.ctor(long,System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this
call [r10]System.Numerics.Tensors.TensorShape:.ctor(long,System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this
mov rbx, bword ptr [rbp-0xB8]
lea rdi, bword ptr [rbx+0x08]
lea rsi, bword ptr [rbp-0x98]
call CORINFO_HELP_ASSIGN_BYREF
mov ecx, 12
rep movsq
lea rax, bword ptr [r14+0x08]
mov rsi, qword ptr [r14]
mov ecx, dword ptr [rsi+0x04]
add rcx, -16
add rax, rcx
mov ecx, dword ptr [rbp-0x30]
shl rcx, 2
add rax, rcx
mov bword ptr [rbx], rax
jmp SHORT G_M51193_IG23
- ;; size=129 bbWeight=0.50 PerfScore 24.38
+ ;; size=128 bbWeight=0.50 PerfScore 24.25
G_M51193_IG22:
cmp dword ptr [rbp-0xAC], 0
jne SHORT G_M51193_IG27
cmp qword ptr [rbp-0xA0], 0
jne SHORT G_M51193_IG27
xor ecx, ecx
mov rax, bword ptr [rbp-0xB8]
mov qword ptr [rax], rcx
mov qword ptr [rax+0x08], rcx
vxorps xmm0, xmm0, xmm0
vmovdqu32 zmmword ptr [rax+0x10], zmm0
vmovdqu ymmword ptr [rax+0x50], ymm0
;; size=54 bbWeight=0.50 PerfScore 6.79
G_M51193_IG23:
cmp qword ptr [rbp-0xD0], 0xD1FFAB1E
je SHORT G_M51193_IG24
call CORINFO_HELP_FAIL_FAST
;; size=18 bbWeight=1 PerfScore 4.00
G_M51193_IG24:
nop
;; size=1 bbWeight=1 PerfScore 0.25
G_M51193_IG25:
lea rsp, [rbp-0x28]
pop rbx
pop r12
pop r13
pop r14
pop r15
pop rbp
ret
;; size=15 bbWeight=1 PerfScore 4.50
G_M51193_IG26:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArrayTypeMismatchException()
call [rax]System.ThrowHelper:ThrowArrayTypeMismatchException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M51193_IG27:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M51193_IG28:
mov rax, 0xD1FFAB1E ; code for System.ThrowHelper:ThrowArgumentOutOfRangeException()
call [rax]System.ThrowHelper:ThrowArgumentOutOfRangeException()
int3
;; size=13 bbWeight=0 PerfScore 0.00
G_M51193_IG29:
call CORINFO_HELP_RNGCHKFAIL
int3
;; size=6 bbWeight=0 PerfScore 0.00
G_M51193_IG30:
call CORINFO_HELP_OVERFLOW
int3
;; size=6 bbWeight=0 PerfScore 0.00
-; Total bytes of code 876, prolog size 107, PerfScore 171.63, instruction count 225, allocated bytes for code 876 (MethodHash=f2973806) for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[int]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (FullOpts)
+; Total bytes of code 875, prolog size 107, PerfScore 171.50, instruction count 224, allocated bytes for code 875 (MethodHash=f2973806) for method System.Numerics.Tensors.ReadOnlyTensorSpan`1[int]:.ctor(System.Array,System.ReadOnlySpan`1[System.Buffers.NIndex],System.ReadOnlySpan`1[long],System.ReadOnlySpan`1[long]):this (FullOpts)
; ============================================================Larger list of diffs: https://gist.github.com/MihuBot/74a157011b60a951ecff4c32d41a59a3 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Job completed in 14 minutes 12 seconds.
Diffs
Diffs
Artifacts:
The text was updated successfully, but these errors were encountered: