Skip to content

Instantly share code, notes, and snippets.

@kunalspathak
Created June 18, 2024 06:48
Show Gist options
  • Save kunalspathak/797640e95a3f3edfd67ef107c19389ae to your computer and use it in GitHub Desktop.
Save kunalspathak/797640e95a3f3edfd67ef107c19389ae to your computer and use it in GitHub Desktop.
Unit tests
; Assembly listing for method System.Text.Unicode.Utf16Utility:GetPointerToFirstInvalidChar(ulong,int,byref,byref):ulong (FullOpts)
; Emitting BLENDED_CODE for generic ARM64 - Windows
; FullOpts code
; optimized code
; optimized using Static PGO
; fp based frame
; fully interruptible
; with Static PGO: fgCalledCount is 100
; PGO data available, but IL did not match
; 0 inlinees with PGO data; 13 single block inlinees; 5 inlinees without PGO data
; Final local variable assignments
;
;  V00 arg0         [V00,T00] ( 24,  9.92)    long  ->  x19        
;  V01 arg1         [V01,T01] ( 11,  9   )     int  ->  x20        
;  V02 arg2         [V02,T02] (  4,  3   )   byref  ->  x21         single-def
;  V03 arg3         [V03,T03] (  4,  3   )   byref  ->  x22         single-def
;  V04 loc0         [V04,T04] (  5,  4.50)     int  ->  x23        
;  V05 loc1         [V05,T05] ( 12,  1.93)    long  ->  x23        
;  V06 loc2         [V06,T11] (  6,  1.11)     int  ->  x24        
;  V07 loc3         [V07,T08] (  4,  1.23)    long  ->  x25        
;* V08 loc4         [V08    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V09 loc5         [V09    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V10 loc6         [V10    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V11 loc7         [V11    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V12 loc8         [V12    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V13 loc9         [V13    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V14 loc10        [V14    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V15 loc11        [V15    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V16 loc12        [V16    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V17 loc13        [V17    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V18 loc14        [V18    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V19 loc15        [V19    ] (  0,  0   )     ref  ->  zero-ref    class-hnd <System.Object>
;* V20 loc16        [V20,T29] (  0,  0   )  simd16  ->  zero-ref    HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;* V21 loc17        [V21,T30] (  0,  0   )  simd16  ->  zero-ref    HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;  V22 loc18        [V22,T26] (  3,  0.50)  simd16  ->   d8         HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;* V23 loc19        [V23,T31] (  0,  0   )  simd16  ->  zero-ref    HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;  V24 loc20        [V24,T13] (  3,  0.38)    long  ->  x20        
;  V25 loc21        [V25,T23] (  7,  1.22)  simd16  ->  d16         HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;* V26 loc22        [V26    ] (  0,  0   )  simd16  ->  zero-ref    HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;* V27 loc23        [V27    ] (  0,  0   )  simd16  ->  zero-ref    HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;  V28 loc24        [V28,T25] (  3,  0.56)  simd16  ->  d17         HFA(simd16)  ld-addr-op <System.Runtime.Intrinsics.Vector128`1[ulong]>
;  V29 loc25        [V29,T10] (  5,  0.94)    long  ->   x0        
;  V30 loc26        [V30,T12] (  7,  0.91)     int  ->   x0        
;  V31 loc27        [V31,T24] (  4,  0.66)  simd16  ->  d17         HFA(simd16)  <System.Runtime.Intrinsics.Vector128`1[ushort]>
;* V32 loc28        [V32,T16] (  0,  0   )     int  ->  zero-ref   
;  V33 loc29        [V33,T27] (  4,  0.25)  simd16  ->  d16         HFA(simd16)  ld-addr-op <System.Runtime.Intrinsics.Vector128`1[ushort]>
;  V34 loc30        [V34,T28] (  3,  0.22)  simd16  ->  d17         HFA(simd16)  ld-addr-op <System.Runtime.Intrinsics.Vector128`1[ushort]>
;  V35 loc31        [V35,T17] (  4,  0.14)  ushort  ->   x1        
;  V36 loc32        [V36,T18] (  4,  0.12)    long  ->   x1        
;  V37 loc33        [V37,T15] (  5,  0.23)     int  ->   x2        
;  V38 loc34        [V38,T09] (  4,  1.14)     int  ->   x0        
;# V39 OutArgs      [V39    ] (  1,  1   )  struct ( 0) [sp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;  V40 tmp1         [V40,T06] (  2,  1   )     int  ->   x0        
;  V41 tmp2         [V41,T07] (  2,  1   )     int  ->   x0        
;* V42 tmp3         [V42    ] (  0,  0   )     int  ->  zero-ref   
;  V43 tmp4         [V43,T20] (  2,  0.12)     int  ->   x3         "impAppendStmt"
;* V44 tmp5         [V44    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V45 tmp6         [V45    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V46 tmp7         [V46    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V47 tmp8         [V47    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V48 tmp9         [V48    ] (  0,  0   )    long  ->  zero-ref    "Inline return value spill temp"
;* V49 tmp10        [V49    ] (  0,  0   )    long  ->  zero-ref    "Inlining Arg"
;* V50 tmp11        [V50    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V51 tmp12        [V51    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V52 tmp13        [V52    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;* V53 tmp14        [V53    ] (  0,  0   )   ubyte  ->  zero-ref    "Inlining Arg"
;  V54 tmp15        [V54,T21] (  2,  0.12)     int  ->   x4         "Inlining Arg"
;  V55 cse0         [V55,T22] (  3,  0.09)     int  ->   x3         "CSE #04: conservative"
;  V56 cse1         [V56,T14] (  2,  0.38)    long  ->   x0         "CSE #02: conservative"
;  V57 cse2         [V57,T19] (  3,  0.12)     int  ->   x2         "CSE #05: conservative"
;  V58 rat0         [V58    ] (  1,  1   )  simd16  ->  [fp+0x10]  "SIMDInitTempVar"
;
; Lcl frame size = 24

G_M65128_IG01:  ;; offset=0x0000
            stp     fp, lr, [sp, #-0x70]!
            stp     d8, d9, [sp, #0x28]
            stp     x19, x20, [sp, #0x38]
            stp     x21, x22, [sp, #0x48]
            stp     x23, x24, [sp, #0x58]
            str     x25, [sp, #0x68]
            mov     fp, sp
            mov     x19, x0
            mov     w20, w1
            mov     x21, x2
            mov     x22, x3
						;; size=44 bbWeight=1 PerfScore 8.50
G_M65128_IG02:  ;; offset=0x002C
            tbz     w20, #31, G_M65128_IG03
            movz    x0, #0x828
            movk    x0, #0x4B76 LSL #16
            movk    x0, #478 LSL #32
            movz    x1, #8
            movk    x1, #0x4B76 LSL #16
            movk    x1, #478 LSL #32
            movz    x2, #0xF6C0      // code for System.Diagnostics.Debug:Fail(System.String,System.String)
            movk    x2, #0xF55 LSL #16
            movk    x2, #0x7FFB LSL #32
            ldr     x2, [x2]
            blr     x2
						;; size=48 bbWeight=1 PerfScore 9.50
G_M65128_IG03:  ;; offset=0x005C
            cbnz    x19, G_M65128_IG06
						;; size=4 bbWeight=1 PerfScore 1.00
G_M65128_IG04:  ;; offset=0x0060
            cmp     w20, #0
            cset    x0, eq
            tst     w0, #255
            bne     G_M65128_IG06
						;; size=16 bbWeight=0.50 PerfScore 1.25
G_M65128_IG05:  ;; offset=0x0070
            movz    x0, #0x888
            movk    x0, #0x4B76 LSL #16
            movk    x0, #478 LSL #32
            movz    x1, #8
            movk    x1, #0x4B76 LSL #16
            movk    x1, #478 LSL #32
            movz    x2, #0xF6C0      // code for System.Diagnostics.Debug:Fail(System.String,System.String)
            movk    x2, #0xF55 LSL #16
            movk    x2, #0x7FFB LSL #32
            ldr     x2, [x2]
            blr     x2
						;; size=44 bbWeight=1 PerfScore 8.50
G_M65128_IG06:  ;; offset=0x009C
            mov     w1, w20
            mov     x0, x19
            movz    x2, #0xDE60      // code for System.Text.Ascii:GetIndexOfFirstNonAsciiChar_Intrinsified(ulong,ulong):ulong
            movk    x2, #0xF82 LSL #16
            movk    x2, #0x7FFB LSL #32
            ldr     x2, [x2]
            blr     x2
            mov     w23, w0
            tbnz    w23, #31, G_M65128_IG08
						;; size=36 bbWeight=1 PerfScore 8.00
G_M65128_IG07:  ;; offset=0x00C0
            cmp     w23, w20
            cset    x0, le
            tst     w0, #255
            bne     G_M65128_IG09
						;; size=16 bbWeight=0.50 PerfScore 1.25
G_M65128_IG08:  ;; offset=0x00D0
            movz    x0, #8
            movk    x0, #0x4B76 LSL #16
            movk    x0, #478 LSL #32
            movz    x1, #8
            movk    x1, #0x4B76 LSL #16
            movk    x1, #478 LSL #32
            movz    x2, #0xF6C0      // code for System.Diagnostics.Debug:Fail(System.String,System.String)
            movk    x2, #0xF55 LSL #16
            movk    x2, #0x7FFB LSL #32
            ldr     x2, [x2]
            blr     x2
						;; size=44 bbWeight=1 PerfScore 8.50
G_M65128_IG09:  ;; offset=0x00FC
            ubfiz   x0, x23, #1, #32
            add     x19, x19, x0
            sub     w20, w20, w23
            cbnz    w20, G_M65128_IG12
						;; size=16 bbWeight=1 PerfScore 3.00
G_M65128_IG10:  ;; offset=0x010C
            str     xzr, [x21]
            str     wzr, [x22]
            mov     x0, x19
						;; size=12 bbWeight=0.50 PerfScore 1.25
G_M65128_IG11:  ;; offset=0x0118
            ldr     x25, [sp, #0x68]
            ldp     x23, x24, [sp, #0x58]
            ldp     x21, x22, [sp, #0x48]
            ldp     x19, x20, [sp, #0x38]
            ldp     d8, d9, [sp, #0x28]
            ldp     fp, lr, [sp], #0x70
            ret     lr
						;; size=28 bbWeight=0.50 PerfScore 4.00
G_M65128_IG12:  ;; offset=0x0134
            mov     x23, xzr
            mov     w24, wzr
            ubfiz   x0, x20, #1, #32
            add     x25, x19, x0
            cmp     w20, #8
            blt     G_M65128_IG24
						;; size=24 bbWeight=0.50 PerfScore 2.00
G_M65128_IG13:  ;; offset=0x014C
            movi    v8.8h, #0x08,  LSL #8
            sub     x20, x25, #16
            cmp     x20, x19
            bhs     G_M65128_IG14
            movz    x0, #8
            movk    x0, #0x4B76 LSL #16
            movk    x0, #478 LSL #32
            movz    x1, #8
            movk    x1, #0x4B76 LSL #16
            movk    x1, #478 LSL #32
            movz    x2, #0xF6C0      // code for System.Diagnostics.Debug:Fail(System.String,System.String)
            movk    x2, #0xF55 LSL #16
            movk    x2, #0x7FFB LSL #32
            ldr     x2, [x2]
            mov     v9.d[0], v8.d[1]
            blr     x2
            mov     v8.d[1], v9.d[0]
						;; size=68 bbWeight=0.12 PerfScore 1.62
G_M65128_IG14:  ;; offset=0x0190
            ldr     q16, [x19]
            cmhs    v17.8h, v16.8h, v8.8h
            movi    v18.8h, #0x80
            cmhs    v18.8h, v16.8h, v18.8h
            movi    v19.4s, #0
            sub     v18.8h, v19.8h, v18.8h
            sub     v17.8h, v18.8h, v17.8h
            mov     w0, wzr
            umov    x0, v17.d[0]
            mov     w1, #1
            umov    x1, v17.d[1]
            add     x0, x1, x0
            lsr     x1, x0, #32
            add     w0, w0, w1
            uxth    w1, w0
            add     w0, w1, w0,  LSR #16
            movi    v17.8h, #0xD8,  LSL #8
            sub     v16.8h, v16.8h, v17.8h
            cmhi    v17.8h, v8.8h, v16.8h
            umaxp   v18.4s, v17.4s, v17.4s
            umov    x1, v18.d[0]
            cmp     x1, #0
            beq     G_M65128_IG22
						;; size=92 bbWeight=0.19 PerfScore 3.84
G_M65128_IG15:  ;; offset=0x01EC
            movi    v18.8h, #0x04,  LSL #8
            cmhi    v16.8h, v18.8h, v16.8h
            bic     v17.8h, v17.8h, v16.8h
            umov    w1, v17.h[0]
            cbnz    w1, G_M65128_IG25
						;; size=20 bbWeight=0.09 PerfScore 0.37
G_M65128_IG16:  ;; offset=0x0200
            mov     w1, wzr
            mov     w2, wzr
						;; size=8 bbWeight=0.05 PerfScore 0.05
G_M65128_IG17:  ;; offset=0x0208
            add     x3, fp, #16	// [V58 rat0]
            str     q16, [x3]
            ldrh    w3, [x3, x2, LSL #1]
            sub     w1, w1, w3
            uxth    w1, w1
            add     w2, w2, #1
            sxtw    w4, w2
            add     x5, fp, #16	// [V58 rat0]
            str     q17, [x5]
            ldrh    w5, [x5, x4, LSL #1]
            cmp     w5, w3
            bne     G_M65128_IG24
						;; size=48 bbWeight=0.03 PerfScore 0.39
G_M65128_IG18:  ;; offset=0x0238
            cmp     w2, #7
            blt     G_M65128_IG17
						;; size=8 bbWeight=0.06 PerfScore 0.09
G_M65128_IG19:  ;; offset=0x0240
            mov     w2, #7
            umov    w2, v16.h[7]
            cbz     w2, G_M65128_IG21
						;; size=12 bbWeight=0.03 PerfScore 0.08
G_M65128_IG20:  ;; offset=0x024C
            sub     x19, x19, #2
            sub     w0, w0, #2
						;; size=8 bbWeight=0.02 PerfScore 0.02
G_M65128_IG21:  ;; offset=0x0254
            mov     w1, w1
            sub     w24, w24, w1
            sub     x2, x23, x1
            sub     x23, x2, x1
						;; size=16 bbWeight=0.03 PerfScore 0.06
G_M65128_IG22:  ;; offset=0x0264
            add     x23, x23, w0, UXTW
            add     x19, x19, #16
            cmp     x19, x20
            bls     G_M65128_IG14
            b       G_M65128_IG24
						;; size=20 bbWeight=0.12 PerfScore 0.50
G_M65128_IG23:  ;; offset=0x0278
            add     x19, x19, #2
						;; size=4 bbWeight=0.31 PerfScore 0.16
G_M65128_IG24:  ;; offset=0x027C
            cmp     x19, x25
            blo     G_M65128_IG27
						;; size=8 bbWeight=0.51 PerfScore 0.77
G_M65128_IG25:  ;; offset=0x0284
            str     x23, [x21]
            str     w24, [x22]
            mov     x0, x19
						;; size=12 bbWeight=0.50 PerfScore 1.25
G_M65128_IG26:  ;; offset=0x0290
            ldr     x25, [sp, #0x68]
            ldp     x23, x24, [sp, #0x58]
            ldp     x21, x22, [sp, #0x48]
            ldp     x19, x20, [sp, #0x38]
            ldp     d8, d9, [sp, #0x28]
            ldp     fp, lr, [sp], #0x70
            ret     lr
						;; size=28 bbWeight=0.50 PerfScore 4.00
G_M65128_IG27:  ;; offset=0x02AC
            ldrh    w0, [x19]
            cmp     w0, #127
            bls     G_M65128_IG23
						;; size=12 bbWeight=0.38 PerfScore 1.72
G_M65128_IG28:  ;; offset=0x02B8
            mov     w1, #0x1F800
            add     w1, w0, w1
            lsr     w1, w1, #16
            add     x23, x23, w1, UXTW
            movn    w1, #0xD7FF
            add     w0, w0, w1
            cmp     w0, #0x7FF
            bhi     G_M65128_IG23
						;; size=32 bbWeight=0.19 PerfScore 1.05
G_M65128_IG29:  ;; offset=0x02D8
            sub     x23, x23, #2
            sub     x0, x25, x19
            cmp     x0, #4
            blo     G_M65128_IG25
						;; size=16 bbWeight=0.10 PerfScore 0.24
G_M65128_IG30:  ;; offset=0x02E8
            ldr     w0, [x19]
            movz    w1, #0x2800
            movk    w1, #0x23FF LSL #16
            add     w0, w0, w1
            tst     w0, #0xFC00FC00
            bne     G_M65128_IG25
						;; size=24 bbWeight=0.05 PerfScore 0.29
G_M65128_IG31:  ;; offset=0x0300
            sub     w24, w24, #1
            add     x23, x23, #2
            add     x19, x19, #2
            b       G_M65128_IG41
            nop     
						;; size=20 bbWeight=0.02 PerfScore 0.07
G_M65128_IG32:  ;; offset=0x0314
            and     z0.b, p1/m, z0.b, z2.b
            bic     z3.h, p4/m, z3.h, z5.h
            eor     z14.s, p5/m, z14.s, z16.s
            orr     z29.d, p7/m, z29.d, z31.d
            add     z5.b, p6/m, z5.b, z7.b
            sub     z15.h, p7/m, z15.h, z29.h
            subr    z2.s, p0/m, z2.s, z13.s
            smax    z24.b, p0/m, z24.b, z2.b
            smin    z9.h, p1/m, z9.h, z27.h
            sabd    z5.b, p2/m, z5.b, z6.b
            uabd    z23.s, p3/m, z23.s, z9.s
            umax    z15.s, p4/m, z15.s, z2.s
            umin    z12.d, p7/m, z12.d, z0.d
            mul     z5.d, p1/m, z5.d, z3.d
            smulh   z17.s, p5/m, z17.s, z5.s
            umulh   z12.b, p2/m, z12.b, z24.b
            asr     z5.s, p0/m, z5.s, z21.s
            asrr    z1.b, p7/m, z1.b, z20.b
            lsl     z0.h, p2/m, z0.h, z0.h
            lslr    z27.d, p6/m, z27.d, z31.d
            lsr     z5.b, p5/m, z5.b, z6.b
            lsrr    z15.s, p4/m, z15.s, z17.s
            shadd   z15.b, p0/m, z15.b, z10.b
            shsub   z16.h, p1/m, z16.h, z11.h
            shsubr  z17.s, p2/m, z17.s, z12.s
            srhadd  z18.d, p3/m, z18.d, z13.d
            uhadd   z19.b, p4/m, z19.b, z14.b
            uhsub   z20.h, p5/m, z20.h, z15.h
            uhsubr  z21.s, p6/m, z21.s, z16.s
            urhadd  z22.d, p7/m, z22.d, z17.d
            addp    z23.b, p6/m, z23.b, z18.b
            smaxp   z24.h, p5/m, z24.h, z19.h
            sminp   z25.s, p4/m, z25.s, z20.s
            umaxp   z26.d, p3/m, z26.d, z21.d
            uminp   z27.b, p2/m, z27.b, z22.b
            sqadd   z28.b, p1/m, z28.b, z23.b
            sqsub   z29.h, p0/m, z29.h, z24.h
            sqsubr  z30.h, p1/m, z30.h, z25.h
            suqadd  z31.b, p2/m, z31.b, z26.b
            uqadd   z0.s, p3/m, z0.s, z27.s
            uqsub   z1.d, p4/m, z1.d, z28.d
            uqsubr  z2.b, p5/m, z2.b, z29.b
            usqadd  z3.b, p6/m, z3.b, z30.b
            sqrshl  z4.b, p7/m, z4.b, z31.b
            sqrshlr z5.h, p0/m, z5.h, z30.h
            sqshl   z6.s, p1/m, z6.s, z29.s
            sqshlr  z7.d, p2/m, z7.d, z28.d
            srshl   z8.b, p3/m, z8.b, z27.b
            srshlr  z9.h, p4/m, z9.h, z26.h
            uqrshl  z10.s, p5/m, z10.s, z25.s
            uqrshlr z11.d, p6/m, z11.d, z24.d
            uqshl   z12.b, p7/m, z12.b, z23.b
            uqshlr  z13.h, p0/m, z13.h, z22.h
            urshl   z14.s, p1/m, z14.s, z21.s
            urshlr  z15.d, p2/m, z15.d, z20.d
            sdiv    z3.s, p2/m, z3.s, z9.s
            sdivr   z31.d, p3/m, z31.d, z29.d
            udiv    z1.s, p0/m, z1.s, z0.s
            udivr   z13.d, p7/m, z13.d, z15.d
            movprfx z13.b, p0/z, z31.b
            movprfx z14.h, p1/z, z0.h
            movprfx z13.s, p2/z, z1.s
            movprfx z15.d, p3/z, z2.d
            movprfx z16.b, p4/m, z3.b
            movprfx z17.h, p5/m, z12.h
            movprfx z0.s, p6/m, z13.s
            movprfx z31.d, p7/m, z22.d
            asr     z0.b, p0/m, z0.b, #1
            asr     z1.b, p1/m, z1.b, #2
            asr     z2.b, p2/m, z2.b, #3
            asr     z3.b, p3/m, z3.b, #7
            asr     z4.b, p7/m, z4.b, #8
            asrd    z5.h, p0/m, z5.h, #1
            asrd    z6.h, p4/m, z6.h, #8
            asrd    z7.h, p7/m, z7.h, #15
            asrd    z31.h, p6/m, z31.h, #16
            lsr     z8.s, p5/m, z8.s, #1
            lsr     z9.s, p6/m, z9.s, #5
            lsr     z31.s, p7/m, z31.s, #17
            lsr     z0.s, p0/m, z0.s, #31
            lsr     z5.s, p1/m, z5.s, #32
            srshr   z6.d, p3/m, z6.d, #1
            srshr   z7.d, p4/m, z7.d, #3
            srshr   z21.d, p5/m, z21.d, #21
            srshr   z22.d, p6/m, z22.d, #63
            urshr   z31.d, p7/m, z31.d, #64
            lsl     z1.b, p0/m, z1.b, #0
            lsl     z3.b, p1/m, z3.b, #1
            lsl     z5.b, p2/m, z5.b, #7
            sqshl   z7.h, p3/m, z7.h, #0
            sqshl   z9.h, p4/m, z9.h, #7
            sqshl   z11.h, p0/m, z11.h, #9
            sqshl   z13.h, p7/m, z13.h, #15
            sqshlu  z14.s, p6/m, z14.s, #0
            sqshlu  z15.s, p0/m, z15.s, #12
            sqshlu  z16.s, p1/m, z16.s, #15
            sqshlu  z0.s, p2/m, z0.s, #31
            uqshl   z17.d, p7/m, z17.d, #0
            uqshl   z18.d, p0/m, z18.d, #18
            uqshl   z19.d, p3/m, z19.d, #32
            uqshl   z0.d, p2/m, z0.d, #63
            asr     z4.b, p3/m, z4.b, z24.d
            lsl     z19.h, p7/m, z19.h, z3.d
            lsr     z0.s, p0/m, z0.s, z0.d
            pmov    p2.b, z12
            pmov    p7.h, z2[0]
            pmov    p15.d, z7[7]
            pmov    p7.d, z16[0]
            pmov    p0.h, z31[1]
            pmov    p1.h, z1[0]
            pmov    p3.s, z9[3]
            pmov    p10.s, z4[0]
            pmov    z11, p12.b
            pmov    z2[0], p7.s
            pmov    z6[7], p8.d
            pmov    z9[0], p7.d
            pmov    z8[1], p4.h
            pmov    z5[0], p9.h
            pmov    z14[3], p2.s
            pmov    z3[0], p15.s
            rev     p1.b, p2.b
            rev     p4.h, p5.h
            rev     p3.s, p7.s
            rev     p0.d, p6.d
            punpkhi p1.h, p7.b
            punpklo p5.h, p3.b
            clasta  z31.b, p7, z31.b, z31.b
            clastb  z30.d, p6, z30.d, z30.d
            clasta  z12, p1, z12, z15.h
            clastb  z13, p2, z13, z16.s
            clastb  z14, p0, z14, z17.d
            clasta  w0, p0, w0, z0.b
            clasta  w1, p2, w1, z3.h
            clastb  w23, p5, w23, z12.s
            clastb  x3, p6, x3, z9.d
            revd    z1.q, p0/m, z6.q
            revd    z5.q, p7/m, z12.q
            cmpeq   p15.b, p0/z, z0.b, z10.b
            cmpge   p14.h, p1/z, z10.h, z23.h
            cmpgt   p13.s, p2/z, z20.s, z24.s
            cmphi   p12.d, p3/z, z30.d, z25.d
            cmphs   p11.b, p4/z, z1.b, z26.b
            cmpge   p10.h, p5/z, z27.h, z11.h
            cmphi   p11.s, p6/z, z28.s, z12.s
            cmphs   p0.d, p7/z, z29.d, z13.d
            cmpgt   p2.b, p0/z, z30.b, z14.b
            cmpne   p3.h, p1/z, z15.h, z20.h
            cmpeq   p15.b, p7/z, z31.b, z3.d
            cmpge   p14.h, p6/z, z21.h, z13.d
            cmpgt   p13.s, p5/z, z11.s, z23.d
            cmphi   p12.b, p4/z, z1.b, z31.d
            cmphs   p11.h, p3/z, z0.h, z30.d
            cmpge   p4.s, p2/z, z0.s, z10.d
            cmphi   p3.b, p1/z, z1.b, z20.d
            cmphs   p2.h, p0/z, z2.h, z30.d
            cmpgt   p1.s, p7/z, z8.s, z24.d
            cmpne   p0.b, p0/z, z14.b, z28.d
            cmpeq   p15.b, p0/z, z31.b, #8
            cmpge   p11.h, p7/z, z21.h, #1
            cmpgt   p10.s, p1/z, z18.s, #4
            cmple   p8.d, p6/z, z11.d, #15
            cmplt   p7.b, p2/z, z8.b, #-16
            cmpne   p0.h, p5/z, z0.h, #-14
            cmphi   p15.b, p7/z, z19.b, #0
            cmphs   p11.h, p1/z, z0.h, #0x24
            cmplo   p8.s, p5/z, z21.s, #0x40
            cmpls   p0.d, p3/z, z9.d, #0x7F
            and     p0.b, p3/z, p14.b, p15.b
            ands    p2.b, p6/z, p8.b, p13.b
            bic     p4.b, p9/z, p12.b, p11.b
            bics    p6.b, p12/z, p0.b, p9.b
            eor     p8.b, p15/z, p11.b, p7.b
            eors    p10.b, p0/z, p5.b, p3.b
            mov     p12.b, p1/z, p10.b
            movs    p14.b, p4/z, p0.b
            nand    p1.b, p7/z, p2.b, p15.b
            nands   p3.b, p0/z, p6.b, p9.b
            nor     p5.b, p13/z, p10.b, p7.b
            nors    p7.b, p2/z, p14.b, p5.b
            not     p9.b, p5/z, p15.b
            nots    p11.b, p8/z, p0.b
            orn     p13.b, p3/z, p1.b, p11.b
            orns    p15.b, p4/z, p1.b, p8.b
            orr     p0.b, p0/z, p15.b, p12.b
            orrs    p2.b, p3/z, p9.b, p14.b
            sel     p4.b, p6, p13.b, p10.b
            movs    p0.b, p15.b
            mov     p0.b, p8/m, p15.b
            mov     p0.b, p15.b
            brkpa   p0.b, p1/z, p10.b, p15.b
            brkpas  p3.b, p4/z, p15.b, p0.b
            brkpb   p7.b, p8/z, p11.b, p13.b
            brkpbs  p11.b, p9/z, p1.b, p3.b
            brka    p0.b, p8/z, p15.b
            brkb    p2.b, p9/z, p14.b
            brka    p0.b, p8/m, p15.b
            brkb    p2.b, p9/m, p14.b
            brkas   p0.b, p14/z, p11.b
            brkbs   p15.b, p1/z, p13.b
            brkn    p0.b, p8/z, p15.b, p0.b
            brkns   p1.b, p7/z, p13.b, p1.b
            pfirst  p0.b, p15, p0.b
            ptrue   p0.b, pow2
            ptrue   p7.h, mul3
            ptrues  p8.s
            ptrues  p15.d
            pnext   p0.b, p15, p0.b
            rdffr   p10.b, p15/z
            rdffrs  p7.b, p14/z
            rdffr   p8.b
            pfalse  p13.b
            ptest   p2, p14.b
            cntp    fp, p0, p15.d
            match   p15.b, p0/z, z21.b, z0.b
            nmatch  p0.h, p7/z, z11.h, z31.h
            bfcvtnt z3.h, p0/m, z4.s
            fcvtlt  z0.d, p7/m, z1.s
            fcvtlt  z14.s, p7/m, z20.h
            fcvtnt  z18.h, p3/m, z9.s
            fcvtnt  z12.s, p3/m, z5.d
						;; size=880 bbWeight=0.02 PerfScore 71.02
G_M65128_IG33:  ;; offset=0x0684
            fcvtxnt z1.s, p2/m, z3.d
            faddp   z16.h, p3/m, z16.h, z19.h
            fmaxnmp z17.s, p4/m, z17.s, z18.s
            fmaxp   z18.d, p5/m, z18.d, z17.d
            fminnmp z19.s, p6/m, z19.s, z16.s
            fminp   z20.h, p7/m, z20.h, z15.h
            fadda   z21, p6, z21, z14.h
            fadda   z22, p5, z22, z13.s
            fadda   z23, p4, z23, z12.d
            fabd    z24.h, p3/m, z24.h, z11.h
            fadd    z25.s, p2/m, z25.s, z10.s
            fdiv    z28.s, p0/m, z28.s, z7.s
            fdivr   z29.d, p1/m, z29.d, z6.d
            fmax    z30.h, p2/m, z30.h, z5.h
            fmaxnm  z31.s, p3/m, z31.s, z4.s
            fmin    z0.d, p4/m, z0.d, z3.d
            fminnm  z1.h, p5/m, z1.h, z2.h
            fmul    z2.s, p6/m, z2.s, z1.s
            fmulx   z3.d, p7/m, z3.d, z0.d
            fscale  z4.h, p6/m, z4.h, z31.h
            fsub    z5.s, p5/m, z5.s, z30.s
            fsubr   z6.d, p4/m, z6.d, z29.d
            bfadd   z0.h, p0/m, z0.h, z1.h
            bfmax   z2.h, p1/m, z2.h, z3.h
            bfmaxnm z4.h, p2/m, z4.h, z5.h
            bfmin   z6.h, p3/m, z6.h, z7.h
            bfminnm z8.h, p4/m, z8.h, z9.h
            bfmul   z10.h, p5/m, z10.h, z11.h
            bfsub   z12.h, p6/m, z12.h, z13.h
            bfcvt   z3.h, p2/m, z9.s
            fcvt    z7.d, p7/m, z1.s
            fcvt    z29.s, p3/m, z12.d
            fcvt    z0.h, p4/m, z13.d
            fcvt    z1.d, p5/m, z14.h
            fcvt    z2.h, p6/m, z15.s
            fcvt    z3.s, p7/m, z16.h
            fcvtx   z2.s, p0/m, z6.d
            fcvtzs  z9.s, p1/m, z3.s
            fcvtzs  z5.d, p0/m, z24.s
            fcvtzs  z12.s, p3/m, z6.d
            fcvtzs  z2.d, p1/m, z17.d
            fcvtzs  z3.h, p2/m, z18.h
            fcvtzs  z4.s, p3/m, z19.h
            fcvtzs  z5.d, p4/m, z20.h
            fcvtzu  z3.s, p2/m, z10.s
            fcvtzu  z10.d, p7/m, z1.s
            fcvtzu  z4.s, p3/m, z13.d
            fcvtzu  z22.d, p6/m, z4.d
            fcvtzu  z23.h, p7/m, z5.h
            fcvtzu  z24.s, p0/m, z6.h
            fcvtzu  z25.d, p1/m, z7.h
            scvtf   z19.s, p2/m, z8.s
            scvtf   z1.d, p5/m, z19.s
            scvtf   z4.s, p0/m, z14.d
            scvtf   z0.d, p0/m, z0.d
            scvtf   z12.h, p5/m, z14.h
            scvtf   z14.h, p7/m, z16.s
            scvtf   z16.h, p1/m, z18.d
            ucvtf   z17.s, p6/m, z11.s
            ucvtf   z3.d, p3/m, z20.s
            ucvtf   z8.s, p1/m, z7.d
            ucvtf   z8.d, p4/m, z9.d
            ucvtf   z13.h, p6/m, z15.h
            ucvtf   z15.h, p0/m, z17.s
            ucvtf   z17.h, p2/m, z19.d
            facge   p0.h, p0/z, z10.h, z31.h
            facgt   p15.s, p1/z, z20.s, z21.s
            facge   p1.d, p2/z, z11.d, z0.d
            facgt   p14.h, p3/z, z1.h, z30.h
            fcmeq   p2.s, p4/z, z28.s, z8.s
            fcmge   p13.d, p5/z, z8.d, z18.d
            fcmgt   p3.h, p6/z, z18.h, z28.h
            fcmge   p12.s, p7/z, z30.s, z1.s
            fcmgt   p4.d, p0/z, z0.d, z11.d
            fcmne   p11.h, p1/z, z21.h, z10.h
            fcmuo   p5.s, p2/z, z31.s, z20.s
            fmla    z0.h, p0/m, z1.h, z2.h
            fmls    z3.s, p2/m, z4.s, z5.s
            fnmla   z6.d, p4/m, z7.d, z8.d
            fnmls   z9.h, p6/m, z10.h, z11.h
            andv    z0, p0, z0.b
            eorv    z1, p1, z1.h
            orv     z2, p2, z2.s
            orv     z3, p3, z3.d
            saddv   d1, p4, z2.b
            saddv   d2, p5, z3.h
            uaddv   d3, p6, z4.s
            smaxv   z15, p7, z4.d
            sminv   z16, p6, z14.s
            umaxv   z17, p5, z24.h
            uminv   z18, p4, z31.b
            cls     z31.b, p0/m, z0.b
            clz     z30.h, p1/m, z1.h
            cnot    z29.s, p2/m, z2.s
            cnt     z28.d, p3/m, z3.d
            fabs    z27.h, p4/m, z4.h
            fneg    z26.s, p5/m, z5.s
            not     z25.b, p6/m, z6.b
            abs     z24.b, p7/m, z7.b
            neg     z23.s, p0/m, z8.s
            sxtb    z22.h, p1/m, z9.h
            sxtb    z22.s, p1/m, z9.s
            sxtb    z22.d, p1/m, z9.d
            sxth    z21.s, p2/m, z10.s
            sxth    z21.d, p2/m, z10.d
            sxtw    z20.d, p3/m, z11.d
            uxtb    z19.h, p4/m, z12.h
            uxtb    z19.s, p4/m, z12.s
            uxtb    z19.d, p4/m, z12.d
            uxth    z18.s, p5/m, z13.s
            uxth    z18.d, p5/m, z13.d
            uxtw    z17.d, p6/m, z14.d
            mla     z0.b, p0/m, z2.b, z19.b
            mls     z2.h, p1/m, z31.h, z31.h
            mad     z4.s, p2/m, z23.s, z12.s
            msb     z8.d, p7/m, z12.d, z22.d
            add     z0.b, z0.b, z0.b
            sqadd   z3.h, z31.h, z12.h
            sqsub   z7.s, z0.s, z31.s
            sub     z19.d, z7.d, z13.d
            uqadd   z23.b, z28.b, z29.b
            uqsub   z31.h, z31.h, z31.h
            mul     z5.b, z0.b, z31.b
            smulh   z0.h, z31.h, z5.h
            umulh   z31.d, z5.d, z0.d
            sqdmulh z7.b, z28.b, z0.b
            sqrdmulh z23.h, z3.h, z31.h
            ftssel  z17.d, z16.d, z15.d
            trn1    z0.b, z1.b, z2.b
            trn1    z3.h, z4.h, z5.h
            trn2    z6.s, z7.s, z8.s
            trn2    z9.d, z10.d, z11.d
            uzp1    z12.b, z13.b, z14.b
            uzp1    z15.h, z16.h, z17.h
            uzp2    z18.s, z19.s, z20.s
            uzp2    z21.d, z22.d, z23.d
            zip1    z24.b, z25.b, z26.b
            zip1    z27.h, z28.h, z29.h
            zip2    z30.s, z31.s, z0.s
            zip2    z1.d, z2.d, z3.d
            tbxq    z0.b, z1.b, z2.b
            tbxq    z3.h, z4.h, z5.h
            tbxq    z6.s, z7.s, z8.s
            tbxq    z9.d, z10.d, z11.d
            sclamp  z0.b, z1.b, z2.b
            sclamp  z3.h, z4.h, z5.h
            uclamp  z6.s, z7.s, z8.s
            uclamp  z9.d, z10.d, z11.d
            fclamp  z0.h, z1.h, z2.h
            fclamp  z3.s, z4.s, z5.s
            fclamp  z6.d, z7.d, z8.d
            eorbt   z0.b, z1.b, z2.b
            eorbt   z3.h, z4.h, z5.h
            eortb   z6.s, z7.s, z8.s
            eortb   z9.d, z10.d, z11.d
            bdep    z0.b, z1.b, z2.b
            bext    z3.h, z4.h, z5.h
            bgrp    z6.s, z7.s, z8.s
            bgrp    z9.d, z10.d, z11.d
            fadd    z0.h, z1.h, z2.h
            fmul    z3.s, z4.s, z5.s
            frecps  z6.d, z7.d, z8.d
            frsqrts z9.h, z10.h, z11.h
            fsub    z12.s, z13.s, z14.s
            ftsmul  z15.d, z16.d, z17.d
            index   z24.b, wzr, w9
            index   z12.d, x15, x0
            pmul    z0.b, z1.b, z2.b
            asr     z9.b, z31.b, z2.d
            lsl     z19.h, z0.h, z12.d
            lsr     z29.s, z10.s, z22.d
            adr     z4.d, [z2.d, z0.d]
            adr     z29.s, [z1.s, z10.s, lsl #1]
            adr     z9.d, [z7.d, z9.d, sxtw]
            adr     z12.d, [z3.d, z5.d, sxtw #2]
            adr     z9.d, [z10.d, z14.d, uxtw]
            adr     z3.d, [z15.d, z11.d, uxtw #3]
            trn1    z0.q, z1.q, z2.q
            trn2    z3.q, z4.q, z5.q
            uzp1    z6.q, z7.q, z8.q
            uzp2    z9.q, z10.q, z11.q
            zip1    z12.q, z13.q, z14.q
            zip2    z15.q, z16.q, z17.q
            and     z0.b, z0.b, #170
            and     z1.b, z1.b, #170
            and     z2.h, z2.h, #0xFF00
            and     z3.h, z3.h, #0xFF00
            eor     z4.s, z4.s, #0x3FFC000
            eor     z5.s, z5.s, #0x3FFC000
            orr     z6.d, z6.d, #0xfffff000000000
            orr     z7.d, z7.d, #0xfffff000000000
            dupm    z0.b, #112
            mov     z1.h, #0x3FFC
            dupm    z2.s, #0x7000
            mov     z3.d, #-0x10000
            dupm    z4.b, #63
						;; size=784 bbWeight=0.02 PerfScore 10.95
G_M65128_IG34:  ;; offset=0x0994
            dupm    z5.h, #0x700
            mov     z6.s, #0xFFFFF0
            mov     z7.d, #0xfffff00000ffffff
            mov     z15.b, p5/z, #0
            mov     z5.b, p15/z, #27
            mov     z31.b, p0/z, #-128
            mov     z0.b, p5/z, #127
            mov     z15.h, p5/z, #0
            mov     z23.s, p12/z, #10
            mov     z4.d, p0/z, #-128
            mov     z19.h, p15/z, #127
            mov     z1.s, p0/z, #1, LSL #8
            mov     z1.d, p0/z, #12, LSL #8
            mov     z1.h, p0/z, #-12, LSL #8
            mov     z1.s, p0/z, #-128, LSL #8
            mov     z0.d, p0/z, #127, LSL #8
            mov     z1.b, p12/m, #5
            mov     z27.h, p13/m, #22, LSL #8
            mov     z27.h, p13/m, #-22, LSL #8
            mov     z0.h, p1/m, #0
            mov     z2.s, p3/m, #0
            mov     z4.d, p5/m, #0
            mov     z4.b, z12.b[63]
            mov     z8.h, z9.h[31]
            mov     z11.s, z28.s[15]
            mov     z21.d, z12.d[7]
            mov     z14.q, z7.q[3]
            mov     z13.b, b8
            mov     z2.h, h0
            mov     z15.s, s31
            mov     z23.d, d27
            mov     z4.q, q3
            mov     z1.b, z16.b[63]
            mov     z17.h, z18.h[31]
            mov     z9.s, z11.s[15]
            mov     z2.d, z3.d[7]
            mov     z3.q, z8.q[3]
            mov     z13.b, b9
            mov     z12.h, h6
            mov     z2.s, s7
            mov     z0.d, d0
            mov     z10.q, q20
            tbl     z0.b, { z1.b }, z2.b
            tbl     z3.h, { z4.h }, z5.h
            tbx     z6.s, z7.s, z8.s
            tbx     z9.d, z10.d, z11.d
            tbl     z0.b, { z1.b, z2.b }, z2.b
            tbl     z3.h, { z4.h, z5.h }, z5.h
            tbl     z6.s, { z7.s, z8.s }, z8.s
            tbl     z9.d, { z10.d, z11.d }, z11.d
            sdot    z0.s, z1.b, z2.b
            sdot    z3.d, z4.h, z5.h
            udot    z6.s, z7.b, z8.b
            udot    z9.d, z10.h, z11.h
            smlalb  z0.h, z1.b, z2.b
            smlalt  z3.s, z4.h, z5.h
            smlslb  z6.d, z7.s, z8.s
            smlslt  z9.h, z10.b, z11.b
            umlalb  z12.s, z13.h, z14.h
            umlalt  z15.d, z16.s, z17.s
            umlslb  z18.h, z19.b, z20.b
            umlslt  z21.s, z22.h, z23.h
            sqdmlalbt z0.h, z1.b, z2.b
            sqdmlslbt z3.s, z4.h, z5.h
            sqdmlslbt z6.d, z7.s, z8.s
            sqdmlalb z0.h, z1.b, z2.b
            sqdmlalt z3.s, z4.h, z5.h
            sqdmlslb z6.d, z7.s, z8.s
            sqdmlslt z9.h, z10.b, z11.b
            sabalb  z0.h, z1.b, z2.b
            sabalt  z3.s, z4.h, z5.h
            uabalb  z6.d, z7.s, z8.s
            uabalt  z9.h, z10.b, z11.b
            sqrdmlah z0.b, z1.b, z2.b
            sqrdmlah z3.h, z4.h, z5.h
            sqrdmlsh z6.s, z7.s, z8.s
            sqrdmlsh z9.d, z10.d, z11.d
            tblq    z0.b, { z1.b }, z2.b
            uzpq1   z3.h, z4.h, z5.h
            uzpq2   z6.s, z7.s, z8.s
            zipq1   z9.d, z10.d, z11.d
            zipq2   z12.b, z13.b, z14.b
            sabdlb  z0.h, z1.b, z2.b
            sabdlt  z3.s, z4.h, z5.h
            saddlb  z6.d, z7.s, z8.s
            saddlt  z9.h, z10.b, z11.b
            ssublb  z12.s, z13.h, z14.h
            ssublt  z15.d, z16.s, z17.s
            uabdlb  z18.h, z19.b, z20.b
            uabdlt  z21.s, z22.h, z24.h
            uaddlb  z24.d, z25.s, z26.s
            uaddlt  z27.h, z28.b, z29.b
            usublb  z30.s, z31.h, z0.h
            usublt  z1.d, z2.s, z3.s
            pmullb  z0.h, z1.b, z2.b
            pmullt  z3.d, z4.s, z5.s
            smullb  z6.h, z7.b, z8.b
            smullt  z9.d, z10.s, z11.s
            sqdmullb z12.h, z13.b, z14.b
            sqdmullt z15.d, z16.s, z17.s
            umullb  z18.h, z19.b, z20.b
            umullt  z21.d, z22.s, z23.s
            saddlbt z0.h, z1.b, z2.b
            ssublbt z3.s, z4.h, z5.h
            ssubltb z6.d, z7.s, z8.s
            saddwb  z0.h, z1.h, z2.b
            saddwt  z3.s, z4.s, z5.h
            ssubwb  z6.d, z7.d, z8.s
            ssubwt  z9.h, z10.h, z11.b
            uaddwb  z12.s, z13.s, z14.h
            uaddwt  z15.d, z16.d, z17.s
            usubwb  z18.h, z19.h, z20.b
            usubwt  z21.s, z22.s, z23.h
            pmullb  z0.q, z1.d, z2.d
            pmullt  z3.q, z4.d, z5.d
            smmla   z0.s, z1.b, z2.b
            ummla   z3.s, z4.b, z5.b
            usmmla  z6.s, z7.b, z8.b
            saba    z0.b, z1.b, z2.b
            saba    z3.h, z4.h, z5.h
            uaba    z6.s, z7.s, z8.s
            uaba    z9.d, z10.d, z11.d
            histseg z0.b, z1.b, z2.b
            bfclamp z0.h, z1.h, z2.h
            bfadd   z0.h, z1.h, z2.h
            bfmul   z3.h, z4.h, z5.h
            bfsub   z6.h, z7.h, z8.h
            and     z0.d, z1.d, z2.d
            bic     z3.d, z4.d, z5.d
            eor     z6.d, z7.d, z8.d
            orr     z9.d, z10.d, z11.d
            orr     z12.d, z13.d, z14.d
            bcax    z0.d, z0.d, z1.d, z2.d
            bsl     z3.d, z3.d, z4.d, z5.d
            bsl1n   z6.d, z6.d, z7.d, z8.d
            bsl2n   z9.d, z9.d, z10.d, z11.d
            eor3    z12.d, z12.d, z13.d, z14.d
            nbsl    z15.d, z15.d, z16.d, z17.d
            xar     z0.b, z0.b, z1.b, #1
            xar     z2.b, z2.b, z3.b, #8
            xar     z4.h, z4.h, z5.h, #2
            xar     z6.h, z6.h, z7.h, #16
            xar     z8.s, z8.s, z9.s, #3
            xar     z10.s, z10.s, z11.s, #32
            xar     z12.d, z12.d, z13.d, #4
            xar     z14.d, z14.d, z15.d, #64
            index   z0.b, #-16, #15
            index   z1.h, #15, #-16
            index   z2.s, #0, #0
            index   z3.d, #-5, #5
            index   z0.b, #-16, w0
            index   z1.h, #0, w1
            index   z2.s, #5, w2
            index   z3.d, #10, x3
            index   z4.b, #-16, wzr
            index   z5.d, #15, xzr
            index   z0.b, w0, #-16
            index   z1.h, w1, #0
            index   z2.s, w2, #5
            index   z3.d, x3, #10
            index   z4.b, wzr, #-16
            index   z5.d, xzr, #15
            addpl   x0, x1, #-32
            addpl   x2, sp, #0
            addvl   x3, x4, #5
            addvl   sp, x5, #31
            addvl   sp, sp, #0
            rdvl    x0, #-32
            rdvl    x5, #0
            rdvl    x10, #5
            rdvl    x15, #31
            cntb    x0, pow2
            cntd    lr, vl1, mul #16
            cnth    x12, vl7, mul #5
            cntw    x23, vl256, mul #7
            cntb    x21, mul4, mul #8
            cntd    x15, mul3, mul #10
            cnth    x5, all, mul #13
            decb    x0, pow2
            decd    x1, vl16, mul #3
            dech    x2, vl32, mul #5
            decw    x3, vl64, mul #7
            incb    x4, vl128, mul #9
            incd    x5, mul3, mul #10
            inch    x6, mul4, mul #13
            incw    x7, all, mul #16
            decd    z0.d, pow2
            dech    z1.h, vl2, mul #2
            decw    z2.s, vl3, mul #4
            incd    z3.d, vl4, mul #8
            inch    z4.h, vl5, mul #12
            incw    z5.s, vl6, mul #16
            sqdecb  x0, w0, pow2
            sqdecd  x1, vl1, mul #2
            sqdech  x2, w2, vl2, mul #3
            sqdecw  x3, vl3, mul #4
            sqincb  x4, w4, vl4, mul #5
            sqincd  x5, vl5, mul #6
            sqinch  x6, w6, vl6, mul #7
            sqincw  x7, vl7, mul #8
            uqdecb  w8, vl8, mul #9
            uqdecd  x9, vl16, mul #10
            uqdech  w10, vl32, mul #11
            uqdecw  x11, vl64, mul #12
            uqincb  w12, vl128, mul #13
            uqincd  x13, vl256, mul #14
            uqinch  w14, mul4, mul #15
            uqincw  x15, all, mul #16
            sqdecd  z0.d, vl1
            sqdech  z1.h, vl2, mul #2
            sqdecw  z2.s, vl3, mul #3
            sqincd  z3.d, vl4, mul #4
            sqinch  z4.h, vl5, mul #5
            sqincw  z5.s, vl6, mul #6
						;; size=856 bbWeight=0.02 PerfScore 12.54
G_M65128_IG35:  ;; offset=0x0CEC
            uqdecd  z6.d, vl7, mul #7
            uqdech  z7.h, vl8, mul #8
            uqdecw  z8.s, vl16, mul #9
            uqincd  z9.d, vl32, mul #10
            uqinch  z10.h, pow2, mul #11
            uqincw  z11.s, all, mul #16
            ext     z0.b, {v1.b, v2.b}, #0
            ext     z2.b, {v3.b, v4.b}, #5
            ext     z4.b, {v5.b, v6.b}, #128
            ext     z6.b, {v31.b, v0.b}, #255
            ext     z0.b, z0.b, z1.b, #0
            ext     z2.b, z2.b, z3.b, #31
            ext     z4.b, z4.b, z5.b, #64
            ext     z6.b, z6.b, z7.b, #255
            fmov    z0.h, p1/m, #2.0000
            fmov    z2.s, p3/m, #1.0000
            fmov    z4.d, p5/m, #-10.0000
            fmov    z6.h, p7/m, #-0.1250
            fmov    z8.s, p9/m, #31.0000
            fmov    z10.d, p11/m, #0.5000
            insr    z0.b, b13
            insr    z29.h, h0
            insr    z4.s, s15
            insr    z8.d, d2
            insr    z4.b, w23
            insr    z11.h, w1
            insr    z14.s, w9
            insr    z19.d, x0
            insr    z29.d, xzr
            trn1    p1.b, p3.b, p4.b
            trn2    p5.h, p2.h, p7.h
            uzp1    p0.s, p0.s, p0.s
            uzp2    p0.d, p0.d, p0.d
            zip1    p0.b, p0.b, p0.b
            zip2    p0.h, p0.h, p0.h
            compact z16.s, p7, z13.s
            compact z15.d, p0, z12.d
            mov     z14.b, p1/m, z11
            mov     z13.s, p2/m, z10
            mov     z12.h, p3/m, z9
            mov     z11.d, p4/m, z8
            mov     z10.d, p5/m, sp
            mov     z9.h, p6/m, w30
            mov     z8.s, p7/m, w29
            mov     z7.b, p0/m, w28
            lasta   z6, p1, z27.b
            lasta   z5, p2, z26.h
            lastb   z4, p3, z25.s
            lastb   z3, p4, z24.d
            lasta   w1, p5, z23.b
            lasta   w0, p6, z22.s
            lastb   w30, p7, z21.h
            lastb   fp, p0, z20.d
            rbit    z28.h, p1/m, z19.h
            rbit    z28.b, p1/m, z19.b
            rbit    z28.s, p1/m, z19.s
            rbit    z28.d, p1/m, z19.d
            revb    z27.h, p2/m, z18.h
            revb    z27.s, p2/m, z18.s
            revb    z27.d, p2/m, z18.d
            revh    z26.s, p3/m, z17.s
            revh    z26.d, p3/m, z17.d
            revw    z25.d, p4/m, z16.d
            splice  z0.b, p0, { z30.b, z31.b }
            splice  z3.d, p7, { z27.d, z28.d }
            splice  z1.h, p1, z1.h, z29.h
            splice  z2.s, p6, z2.s, z28.s
            mov     z0.h, p0/m, z30.h
            sel     z29.d, p15, z28.d, z4.d
            mov     z5.s, p13/m, z27.s
            sadalp  z26.h, p3/m, z8.b
            uadalp  z27.s, p2/m, z9.h
            uadalp  z28.d, p0/m, z31.s
            sqabs   z29.b, p7/m, z0.b
            sqneg   z31.h, p6/m, z1.h
            sqneg   z0.s, p5/m, z2.s
            sqneg   z1.d, p4/m, z3.d
            urecpe  z2.s, p3/m, z4.s
            ursqrte z3.s, p0/m, z5.s
            sqcvtn  z0.h, { z2.s, z3.s }
            sqcvtun z6.h, { z8.s, z9.s }
            uqcvtn  z14.h, { z16.s, z17.s }
            sqrshrn z0.h, { z0.s, z1.s }, #5
            sqrshrun z0.h, { z0.s, z1.s }, #5
            uqrshrn z0.h, { z0.s, z1.s }, #5
            sqrshrn z0.h, { z2.s, z3.s }, #16
            sqrshrun z0.h, { z4.s, z5.s }, #7
            uqrshrn z0.h, { z6.s, z7.s }, #1
            sqrshrn z30.h, { z16.s, z17.s }, #16
            sqrshrun z16.h, { z8.s, z9.s }, #7
            uqrshrn z15.h, { z12.s, z13.s }, #1
            rshrnb  z0.b, z1.h, #1
            rshrnt  z2.h, z3.s, #1
            shrnb   z4.s, z5.d, #1
            shrnt   z6.b, z7.h, #2
            sqrshrnb z8.h, z9.s, #3
            sqrshrnt z10.s, z11.d, #4
            sqrshrunb z12.b, z13.h, #5
            sqrshrunt z14.h, z15.s, #8
            sqshrnb z16.s, z17.d, #8
            sqshrnt z18.b, z19.h, #6
            sqshrunb z20.h, z21.s, #13
            sqshrunt z22.s, z23.d, #16
            uqrshrnb z24.b, z25.h, #7
            uqrshrnt z26.h, z27.s, #16
            uqshrnb z28.s, z29.d, #32
            uqshrnt z30.b, z31.h, #8
            cntp    x0, pn0.b, vlx2
            cntp    x1, pn1.b, vlx4
            cntp    x2, pn2.h, vlx2
            cntp    x3, pn3.h, vlx4
            cntp    x4, pn4.s, vlx2
            cntp    x5, pn5.s, vlx4
            cntp    x6, pn6.d, vlx2
            cntp    x7, pn7.d, vlx4
            decp    x0, p0.b
            decp    x1, p1.h
            decp    x2, p2.s
            decp    x3, p3.d
            incp    x4, p4.b
            incp    x5, p5.h
            incp    x6, p6.s
            incp    x7, p7.d
            decp    z0.h, p0.h
            decp    z1.s, p1.s
            decp    z2.d, p2.d
            incp    z3.h, p3.h
            incp    z4.s, p4.s
            incp    z5.d, p5.d
            sqdecp  x0, p0.b, w0
            sqdecp  x1, p1.h, w1
            sqdecp  x2, p2.s, w2
            sqdecp  x3, p3.d, w3
            sqdecp  x4, p4.b
            sqdecp  x5, p5.h
            sqdecp  x6, p6.s
            sqdecp  x7, p7.d
            sqincp  x0, p0.h, w0
            sqincp  x1, p1.s, w1
            sqincp  x2, p2.b, w2
            sqincp  x3, p3.d, w3
            sqincp  x4, p4.b
            sqincp  x5, p5.h
            sqincp  x6, p6.s
            sqincp  x7, p7.d
            uqdecp  w0, p0.b
            uqdecp  w1, p1.h
            uqdecp  w2, p2.s
            uqdecp  w3, p3.d
            uqdecp  x4, p4.b
            uqdecp  x5, p5.h
            uqdecp  x6, p6.s
            uqdecp  x7, p7.d
            uqincp  w0, p0.b
            uqincp  w1, p1.h
            uqincp  w2, p2.s
            uqincp  w3, p3.d
            uqincp  x4, p4.b
            uqincp  x5, p5.h
            uqincp  x6, p6.s
            uqincp  x7, p7.d
            sqdecp  z0.h, p0.h
            sqdecp  z1.s, p1.s
            sqdecp  z2.d, p2.d
            sqincp  z3.h, p3.h
            sqincp  z4.s, p4.s
            sqincp  z5.d, p5.d
            uqdecp  z6.h, p6.h
            uqdecp  z7.s, p7.s
            uqdecp  z8.d, p0.d
            uqincp  z9.h, p1.h
            uqincp  z10.s, p2.s
            uqincp  z11.d, p3.d
            setffr  
            wrffr   p0.b
            ctermeq x0, x1
            ctermeq w2, w3
            ctermne x4, x5
            ctermne w6, w7
            sqxtnb  z0.b, z5.h
            sqxtnt  z3.b, z7.h
            sqxtunb z0.b, z8.h
            sqxtunt z5.b, z3.h
            uqxtnb  z0.b, z7.h
            uqxtnt  z0.b, z1.h
            sqxtnb  z0.h, z0.s
            sqxtnt  z9.h, z9.s
            sqxtunb z0.h, z6.s
            sqxtunt z0.h, z4.s
            uqxtnb  z6.h, z2.s
            uqxtnt  z0.h, z1.s
            sqxtnb  z7.s, z7.d
            sqxtnt  z0.s, z8.d
            sqxtunb z0.s, z9.d
            sqxtunt z0.s, z4.d
            uqxtnb  z3.s, z9.d
            uqxtnt  z0.s, z3.d
            aesd    z0.b, z0.b, z0.b
            aese    z1.b, z1.b, z2.b
            sm4e    z3.s, z3.s, z5.s
            aesimc  z0.b, z0.b
            aesmc   z5.b, z5.b
            fmla    z0.s, z2.s, z1.s[0]
            fmla    z4.s, z6.s, z3.s[1]
            fmls    z8.s, z10.s, z5.s[2]
            fmls    z12.s, z14.s, z7.s[3]
            fmla    z1.d, z0.d, z0.d[0]
            fmla    z3.d, z2.d, z5.d[1]
            fmls    z5.d, z4.d, z10.d[0]
            fmls    z7.d, z6.d, z15.d[1]
            bfmla   z1.h, z2.h, z0.h[0]
            bfmla   z3.h, z4.h, z2.h[2]
            bfmls   z5.h, z6.h, z4.h[5]
            bfmls   z7.h, z8.h, z7.h[7]
            fcmla   z0.s, z1.s, z0.s[0], #0
            fcmla   z2.s, z3.s, z5.s[1], #90
            fcmla   z4.s, z5.s, z10.s[0], #180
            fcmla   z6.s, z7.s, z15.s[1], #270
            fmul    z0.s, z2.s, z1.s[0]
            fmul    z4.s, z6.s, z3.s[1]
            fmul    z8.s, z10.s, z5.s[2]
            fmul    z12.s, z14.s, z7.s[3]
            fmul    z1.d, z0.d, z0.d[0]
            fmul    z3.d, z2.d, z5.d[1]
            fmul    z5.d, z4.d, z10.d[0]
            fmul    z7.d, z6.d, z15.d[1]
            bfmul   z1.h, z2.h, z0.h[0]
            bfmul   z3.h, z4.h, z2.h[2]
            bfmul   z5.h, z6.h, z4.h[5]
            bfmul   z7.h, z8.h, z7.h[7]
            fdot    z0.s, z2.h, z1.h[0]
            fdot    z4.s, z6.h, z3.h[1]
            bfdot   z8.s, z10.h, z5.h[2]
            bfdot   z12.s, z14.h, z7.h[3]
            bfmlalb z0.s, z1.h, z0.h[0]
            bfmlalt z2.s, z3.h, z1.h[1]
            bfmlslb z4.s, z5.h, z2.h[2]
            bfmlslt z6.s, z7.h, z3.h[3]
            fmlalb  z8.s, z9.h, z4.h[4]
            fmlalt  z10.s, z11.h, z5.h[5]
						;; size=960 bbWeight=0.02 PerfScore 23.25
G_M65128_IG36:  ;; offset=0x10AC
            fmlslb  z12.s, z13.h, z6.h[6]
            fmlslt  z14.s, z15.h, z7.h[7]
            bfdot   z0.s, z1.h, z2.h
            fdot    z3.s, z4.h, z5.h
            bfmlalb z0.s, z1.h, z2.h
            bfmlalt z3.s, z4.h, z5.h
            bfmlslb z6.s, z7.h, z8.h
            bfmlslt z9.s, z10.h, z11.h
            fmlalb  z12.s, z13.h, z14.h
            fmlalt  z15.s, z16.h, z17.h
            fmlslb  z18.s, z19.h, z20.h
            fmlslt  z21.s, z22.h, z23.h
            bfmmla  z0.s, z1.h, z2.h
            faddv   z21, p7, z7.h
            fmaxnmv z22, p6, z6.h
            fmaxv   z23, p5, z5.s
            fminnmv z24, p4, z4.d
            fminv   z25, p3, z3.s
            frinta  z26.h, p7/m, z2.h
            frinti  z27.s, p6/m, z1.s
            frintm  z28.d, p5/m, z0.d
            frintn  z29.h, p4/m, z10.h
            frintp  z30.s, p3/m, z11.s
            frintx  z31.d, p2/m, z12.d
            frintz  z0.h, p0/m, z13.h
            frecpx  z5.h, p5/m, z5.h
            fsqrt   z6.s, p6/m, z6.s
            whilege p0.b, w0, w1
            whilege p1.b, x2, x3
            whilegt p2.b, w4, w5
            whilegt p3.b, x6, x7
            whilehi p4.h, w8, w9
            whilehi p5.h, x10, x11
            whilehs p6.h, w12, w13
            whilehs p7.h, x14, x15
            whilele p8.s, w0, w1
            whilele p9.s, x2, x3
            whilelo p10.s, w4, w5
            whilelo p11.s, x6, x7
            whilels p12.d, w8, w9
            whilels p13.d, x10, x11
            whilelt p14.d, w12, w13
            whilelt p15.d, x14, x15
            whilerw p0.b, x0, x1
            whilerw p1.h, x2, x3
            whilerw p2.s, x4, x5
            whilerw p3.d, x6, x7
            whilewr p4.b, x8, x9
            whilewr p5.h, x10, x11
            whilewr p6.s, x12, x13
            whilewr p7.d, x14, x15
            pext    p0.b, pn8[0]
            pext    p1.h, pn9[1]
            pext    p2.s, pn10[2]
            pext    p3.d, pn11[3]
            pext    { p8.b, p9.b }, pn12[0]
            pext    { p9.h, p10.h }, pn13[1]
            pext    { p10.s, p11.s }, pn14[0]
            pext    { p11.d, p12.d }, pn15[1]
            whilege { p0.b, p1.b }, x0, x1
            whilegt { p2.b, p3.b }, x2, x3
            whilehi { p4.h, p5.h }, x4, x5
            whilehs { p6.h, p7.h }, x6, x7
            whilele { p8.s, p9.s }, x8, x9
            whilelo { p10.s, p11.s }, x10, x11
            whilels { p12.d, p13.d }, x12, x13
            whilelt { p14.d, p15.d }, x14, x15
            whilege pn8.b, x0, x1, vlx2
            whilege pn9.b, x2, x3, vlx4
            whilegt pn10.h, x4, x5, vlx2
            whilegt pn11.h, x6, x7, vlx4
            whilehi pn12.s, x8, x9, vlx2
            whilehi pn13.s, x10, x11, vlx4
            whilehs pn14.d, x12, x13, vlx2
            whilehs pn15.d, x14, x15, vlx4
            whilele pn8.b, x0, x1, vlx2
            whilele pn9.b, x2, x3, vlx4
            whilelo pn10.h, x4, x5, vlx2
            whilelo pn11.h, x6, x7, vlx4
            whilels pn12.d, x8, x9, vlx2
            whilels pn13.d, x10, x11, vlx4
            whilelt pn14.b, x12, x13, vlx2
            whilelt pn15.b, x14, x15, vlx4
            ptrue   pn8.b
            ptrue   pn9.h
            ptrue   pn10.s
            ptrue   pn11.d
            fmov    z0.h, #2.0000
            fmov    z1.s, #1.0000
            fmov    z2.d, #2.0000
            fmov    z3.h, #-10.0000
            fmov    z4.s, #-0.1250
            fmov    z5.d, #31.0000
            mov     z0.b, #-128
            mov     z1.h, #0
            mov     z2.s, #5
            mov     z3.d, #127
            mov     z4.d, #1, LSL #8
            mov     z5.h, #-128, LSL #8
            mov     z6.s, #5, LSL #8
            mov     z7.d, #127, LSL #8
            mov     z0.b, #0
            mov     z1.h, #0
            mov     z2.s, #0
            mov     z3.d, #0
            add     z0.b, z0.b, #0
            sqadd   z1.h, z1.h, #5
            sqsub   z2.s, z2.s, #128
            sub     z3.d, z3.d, #255
            subr    z4.d, z4.d, #1, LSL #8
            uqadd   z5.h, z5.h, #5, LSL #8
            uqsub   z6.s, z6.s, #255, LSL #8
            sdot    z1.s, z2.h, z0.h[0]
            sdot    z3.s, z4.h, z1.h[1]
            udot    z5.s, z6.h, z2.h[2]
            udot    z7.s, z8.h, z3.h[3]
            cdot    z0.s, z1.b, z2.b, #0
            cdot    z3.s, z4.b, z5.b, #90
            cdot    z6.d, z7.h, z8.h, #180
            cdot    z9.d, z10.h, z11.h, #270
            cmla    z0.b, z1.b, z2.b, #0
            cmla    z3.h, z4.h, z5.h, #90
            cmla    z6.s, z7.s, z8.s, #180
            cmla    z9.d, z10.d, z11.d, #270
            sqrdcmlah z12.b, z13.b, z14.b, #0
            sqrdcmlah z15.h, z16.h, z17.h, #90
            sqrdcmlah z18.s, z19.s, z20.s, #180
            sqrdcmlah z21.d, z22.d, z23.d, #270
            sdot    z9.s, z10.s, z4.s[0]
            sdot    z11.s, z12.s, z5.s[1]
            udot    z13.s, z14.s, z6.s[2]
            udot    z15.s, z16.s, z7.s[3]
            sdot    z0.d, z1.h, z0.h[0]
            sdot    z2.d, z3.h, z5.h[1]
            udot    z4.d, z5.h, z10.h[0]
            udot    z6.d, z7.h, z15.h[1]
            sudot   z17.s, z18.b, z0.b[0]
            sudot   z19.s, z20.b, z1.b[1]
            usdot   z21.s, z22.b, z2.b[2]
            usdot   z23.s, z24.b, z3.b[3]
            mul     z0.h, z1.h, z1.h[1]
            mul     z2.h, z3.h, z3.h[3]
            mul     z4.h, z5.h, z5.h[5]
            mul     z6.h, z7.h, z7.h[7]
            mul     z8.s, z9.s, z1.s[0]
            mul     z10.s, z11.s, z3.s[1]
            mul     z12.s, z13.s, z5.s[2]
            mul     z14.s, z15.s, z7.s[3]
            mul     z16.d, z17.d, z0.d[0]
            mul     z18.d, z19.d, z5.d[1]
            mul     z20.d, z21.d, z10.d[0]
            mul     z22.d, z23.d, z15.d[1]
            smullb  z0.s, z1.h, z0.h[0]
            smullb  z2.s, z3.h, z1.h[1]
            smullt  z4.s, z5.h, z2.h[2]
            smullt  z6.s, z7.h, z3.h[3]
            umullb  z8.s, z9.h, z4.h[4]
            umullb  z10.s, z11.h, z5.h[5]
            umullt  z12.s, z13.h, z6.h[6]
            umullt  z14.s, z15.h, z7.h[7]
            smullb  z0.d, z1.s, z0.s[0]
            smullb  z2.d, z3.s, z2.s[1]
            smullt  z4.d, z5.s, z4.s[2]
            smullt  z6.d, z7.s, z6.s[3]
            umullb  z8.d, z9.s, z8.s[0]
            umullb  z10.d, z11.s, z10.s[1]
            umullt  z12.d, z13.s, z12.s[2]
            umullt  z14.d, z15.s, z14.s[3]
            smlalb  z0.s, z1.h, z0.h[0]
            smlalt  z2.s, z3.h, z1.h[1]
            smlslb  z4.s, z5.h, z2.h[2]
            smlslt  z6.s, z7.h, z3.h[3]
            umlalb  z8.s, z9.h, z4.h[4]
            umlalt  z10.s, z11.h, z5.h[5]
            umlslb  z12.s, z13.h, z6.h[6]
            umlslt  z14.s, z15.h, z7.h[7]
            smlalb  z0.d, z1.s, z0.s[0]
            smlalt  z2.d, z3.s, z2.s[1]
            smlslb  z4.d, z5.s, z4.s[2]
            smlslt  z6.d, z7.s, z6.s[3]
            umlalb  z8.d, z9.s, z8.s[0]
            umlalt  z10.d, z11.s, z10.s[1]
            umlslb  z12.d, z13.s, z12.s[2]
            umlslt  z14.d, z15.s, z14.s[3]
            sqdmullb z0.s, z2.h, z1.h[1]
            sqdmullb z4.s, z6.h, z3.h[3]
            sqdmullt z8.s, z10.h, z5.h[5]
            sqdmullt z12.s, z14.h, z7.h[7]
            sqdmullb z0.d, z2.s, z0.s[0]
            sqdmullb z4.d, z6.s, z5.s[1]
            sqdmullt z8.d, z10.s, z10.s[2]
            sqdmullt z12.d, z14.s, z15.s[3]
            sqdmulh z0.h, z1.h, z1.h[1]
            sqdmulh z2.h, z3.h, z3.h[3]
            sqrdmulh z4.h, z5.h, z5.h[5]
            sqrdmulh z6.h, z7.h, z7.h[7]
            sqdmulh z8.s, z9.s, z0.s[0]
            sqdmulh z10.s, z11.s, z2.s[1]
            sqrdmulh z12.s, z13.s, z4.s[2]
            sqrdmulh z14.s, z15.s, z6.s[3]
            sqdmulh z16.d, z17.d, z0.d[0]
            sqdmulh z18.d, z19.d, z5.d[1]
            sqrdmulh z20.d, z21.d, z10.d[0]
            sqrdmulh z22.d, z23.d, z15.d[1]
						;; size=816 bbWeight=0.02 PerfScore 12.95
G_M65128_IG37:  ;; offset=0x13DC
            sqdmlalb z0.s, z1.h, z1.h[1]
            sqdmlalt z2.s, z3.h, z3.h[3]
            sqdmlslb z4.s, z5.h, z5.h[5]
            sqdmlslt z6.s, z0.h, z7.h[7]
            sqdmlalb z8.d, z9.s, z0.s[0]
            sqdmlalt z10.d, z11.s, z5.s[1]
            sqdmlslb z12.d, z13.s, z10.s[2]
            sqdmlslt z14.d, z15.s, z15.s[3]
            mla     z0.h, z1.h, z1.h[1]
            mla     z2.h, z3.h, z3.h[3]
            mls     z4.h, z5.h, z5.h[5]
            mls     z6.h, z7.h, z7.h[7]
            mla     z8.s, z9.s, z1.s[0]
            mla     z10.s, z11.s, z3.s[1]
            mls     z12.s, z13.s, z5.s[2]
            mls     z14.s, z15.s, z7.s[3]
            mla     z16.d, z17.d, z0.d[0]
            mla     z18.d, z19.d, z5.d[1]
            mls     z20.d, z21.d, z10.d[0]
            mls     z22.d, z23.d, z15.d[1]
            sqrdmlah z0.h, z1.h, z1.h[1]
            sqrdmlah z2.h, z3.h, z3.h[3]
            sqrdmlsh z4.h, z5.h, z5.h[5]
            sqrdmlsh z6.h, z7.h, z7.h[7]
            sqrdmlah z8.s, z9.s, z0.s[0]
            sqrdmlah z10.s, z11.s, z2.s[1]
            sqrdmlsh z12.s, z13.s, z4.s[2]
            sqrdmlsh z14.s, z15.s, z6.s[3]
            sqrdmlah z16.d, z17.d, z0.d[0]
            sqrdmlah z18.d, z19.d, z5.d[1]
            sqrdmlsh z20.d, z21.d, z10.d[0]
            sqrdmlsh z22.d, z23.d, z15.d[1]
            sshllb  z0.h, z1.b, #1
            sshllt  z2.h, z3.b, #3
            ushllb  z4.h, z5.b, #5
            ushllt  z6.h, z7.b, #7
            sshllb  z8.s, z9.h, #0
            sshllt  z10.s, z11.h, #5
            ushllb  z12.s, z13.h, #10
            ushllt  z14.s, z15.h, #15
            sshllb  z16.d, z17.s, #8
            sshllt  z18.d, z19.s, #16
            ushllb  z20.d, z21.s, #24
            ushllt  z22.d, z23.s, #31
            cadd    z0.b, z0.b, z1.b, #90
            cadd    z2.h, z2.h, z3.h, #90
            cadd    z4.s, z4.s, z5.s, #270
            cadd    z6.d, z6.d, z7.d, #270
            sqcadd  z8.b, z8.b, z9.b, #270
            sqcadd  z10.h, z10.h, z11.h, #270
            sqcadd  z12.s, z12.s, z13.s, #90
            sqcadd  z14.d, z14.d, z15.d, #90
            adclb   z0.s, z1.s, z2.s
            adclb   z3.d, z4.d, z5.d
            adclt   z6.s, z7.s, z8.s
            adclt   z9.d, z10.d, z11.d
            sbclb   z12.s, z13.s, z14.s
            sbclb   z15.d, z16.d, z17.d
            sbclt   z18.s, z19.s, z20.s
            sbclt   z21.d, z22.d, z23.d
            smax    z0.b, z0.b, #-128
            smax    z1.h, z1.h, #127
            smin    z2.s, z2.s, #-128
            smin    z3.d, z3.d, #127
            umax    z4.b, z4.b, #0
            umax    z5.h, z5.h, #255
            umin    z6.s, z6.s, #0
            umin    z7.d, z7.d, #255
            mul     z0.b, z0.b, #-128
            mul     z1.h, z1.h, #0
            mul     z2.s, z2.s, #5
            mul     z3.d, z3.d, #127
            sdot    z0.s, z1.h, z2.h
            udot    z3.s, z4.h, z5.h
            usdot   z0.s, z1.b, z2.b
            cdot    z0.s, z7.b, z1.b[3], #0
            cdot    z2.s, z5.b, z3.b[2], #90
            cdot    z4.s, z3.b, z5.b[1], #180
            cdot    z6.s, z1.b, z7.b[0], #270
            cdot    z0.d, z1.h, z0.h[0], #0
            cdot    z2.d, z3.h, z5.h[1], #90
            cdot    z4.d, z5.h, z10.h[0], #180
            cdot    z6.d, z7.h, z15.h[1], #270
            cmla    z0.h, z7.h, z1.h[3], #0
            cmla    z2.h, z5.h, z3.h[2], #90
            cmla    z4.h, z3.h, z5.h[1], #180
            cmla    z6.h, z1.h, z7.h[0], #270
            cmla    z0.s, z1.s, z0.s[0], #0
            cmla    z2.s, z3.s, z5.s[1], #90
            cmla    z4.s, z5.s, z10.s[0], #180
            cmla    z6.s, z7.s, z15.s[1], #270
            sqrdcmlah z0.h, z7.h, z1.h[3], #0
            sqrdcmlah z2.h, z5.h, z3.h[2], #90
            sqrdcmlah z4.h, z3.h, z5.h[1], #180
            sqrdcmlah z6.h, z1.h, z7.h[0], #270
            sqrdcmlah z0.s, z1.s, z0.s[0], #0
            sqrdcmlah z2.s, z3.s, z5.s[1], #90
            sqrdcmlah z4.s, z5.s, z10.s[0], #180
            sqrdcmlah z6.s, z7.s, z15.s[1], #270
            ld1d    { z5.d }, p3/z, [x4]
            ld1d    { z0.q }, p2/z, [x3, #5, mul vl]
            ld1w    { z0.s }, p2/z, [x3, #3, mul vl]
            ld1w    { z0.d }, p2/z, [x3, #3, mul vl]
            ld1w    { z0.q }, p2/z, [x3, #3, mul vl]
            ld1sw   { z0.d }, p5/z, [x3, #4, mul vl]
            ld1sb   { z3.h }, p0/z, [x2, #6, mul vl]
            ld1sb   { z3.s }, p0/z, [x2, #6, mul vl]
            ld1sb   { z3.d }, p0/z, [x2, #6, mul vl]
            ld1b    { z5.b }, p1/z, [x3, #7, mul vl]
            ld1b    { z5.h }, p1/z, [x3, #7, mul vl]
            ld1b    { z5.s }, p1/z, [x3, #7, mul vl]
            ld1b    { z5.d }, p1/z, [x3, #7, mul vl]
            ld1sh   { z7.s }, p3/z, [x5, #2, mul vl]
            ld1sh   { z7.d }, p3/z, [x5, #2, mul vl]
            ld1h    { z2.h }, p1/z, [x6, #1, mul vl]
            ld1h    { z2.s }, p1/z, [x6, #1, mul vl]
            ld1h    { z2.d }, p1/z, [x6, #1, mul vl]
            ldnf1d  { z0.d }, p0/z, [x0]
            ldnf1sw { z0.d }, p0/z, [x0]
            ldnf1d  { z0.d }, p1/z, [x2, #5, mul vl]
            ldnf1sw { z0.d }, p1/z, [x2, #5, mul vl]
            ldnf1sh { z0.s }, p1/z, [x5, #5, mul vl]
            ldnf1w  { z0.s }, p2/z, [x4, #5, mul vl]
            ldnf1sh { z0.d }, p1/z, [x5, #5, mul vl]
            ldnf1w  { z0.d }, p2/z, [x4, #5, mul vl]
            ldnf1h  { z1.h }, p3/z, [x2, #5, mul vl]
            ldnf1sb { z0.h }, p4/z, [x1, #5, mul vl]
            ldnf1h  { z1.s }, p3/z, [x2, #5, mul vl]
            ldnf1sb { z0.s }, p4/z, [x1, #5, mul vl]
            ldnf1h  { z1.d }, p3/z, [x2, #5, mul vl]
            ldnf1sb { z0.d }, p4/z, [x1, #5, mul vl]
            ldnf1b  { z2.b }, p5/z, [x3, #-4, mul vl]
            ldnf1b  { z2.h }, p5/z, [x3, #-2, mul vl]
            ldnf1b  { z2.s }, p5/z, [x3, #2, mul vl]
            ldnf1b  { z2.d }, p5/z, [x3, #1, mul vl]
            ldnt1b  { z0.b }, p1/z, [x2, #-5, mul vl]
            ldnt1d  { z3.d }, p4/z, [x5, #-1, mul vl]
            ldnt1h  { z6.h }, p7/z, [x8]
            ldnt1w  { z1.s }, p2/z, [x3, #-8, mul vl]
            ld1rob  { z0.b }, p1/z, [x2]
            ld1rod  { z4.d }, p5/z, [x6, #-0x20]
            ld1roh  { z8.h }, p3/z, [x1, #-0x100]
            ld1row  { z3.s }, p4/z, [x0, #0xE0]
            ld1rqb  { z6.b }, p7/z, [x8, #0x40]
            ld1rqd  { z9.d }, p0/z, [x1, #-0x80]
            ld1rqh  { z4.h }, p5/z, [x6, #0x70]
            ld1rqw  { z31.s }, p2/z, [x1, #-0x10]
            ld2q    { z0.q, z1.q }, p1/z, [x2, #-0x10, mul vl]
            ld2q    { z31.q, z0.q }, p1/z, [x2, #-0x10, mul vl]
            ld2q    { z0.q, z1.q }, p1/z, [x2, #0x0E, mul vl]
            ld3q    { z0.q - z2.q }, p4/z, [x5, #-0x18, mul vl]
            ld3q    { z0.q - z2.q }, p4/z, [x5, #0x15, mul vl]
            ld4q    { z0.q - z3.q }, p5/z, [x3, #-0x20, mul vl]
            ld4q    { z0.q - z3.q }, p5/z, [x3, #0x1C, mul vl]
            ld2q    { z12.q, z13.q }, p1/z, [x2, #-0x10, mul vl]
            ld2q    { z13.q, z14.q }, p1/z, [x2, #0x0E, mul vl]
            ld3q    { z14.q - z16.q }, p4/z, [x5, #-0x18, mul vl]
            ld3q    { z15.q - z17.q }, p4/z, [x5, #0x15, mul vl]
            ld4q    { z16.q - z19.q }, p5/z, [x3, #-0x20, mul vl]
            ld4q    { z27.q - z30.q }, p5/z, [x3, #0x1C, mul vl]
            ld4q    { z28.q - z31.q }, p5/z, [x3, #0x1C, mul vl]
            ld4q    { z29.q, z30.q, z31.q, z0.q }, p5/z, [x3, #0x1C, mul vl]
            ld4q    { z30.q, z31.q, z0.q, z1.q }, p5/z, [x3, #0x1C, mul vl]
            ld4q    { z31.q, z0.q, z1.q, z2.q }, p5/z, [x3, #0x1C, mul vl]
            ld2q    { z31.q, z0.q }, p1/z, [x2, #-0x10, mul vl]
            ld2b    { z0.b, z1.b }, p1/z, [x2, #-0x10, mul vl]
            ld2d    { z4.d, z5.d }, p5/z, [x7, #0x0E, mul vl]
            ld2h    { z6.h, z7.h }, p5/z, [x4, #0x08, mul vl]
            ld2w    { z0.s, z1.s }, p0/z, [x1, #0x02, mul vl]
            ld3b    { z0.b - z2.b }, p0/z, [x0, #0x15, mul vl]
            ld3d    { z0.d - z2.d }, p0/z, [x0, #-0x18, mul vl]
            ld3h    { z0.h - z2.h }, p0/z, [x0, #0x15, mul vl]
            ld3w    { z0.s - z2.s }, p0/z, [x0, #-0x18, mul vl]
            ld4b    { z31.b, z0.b, z1.b, z2.b }, p2/z, [x1, #-0x20, mul vl]
            ld4d    { z8.d - z11.d }, p0/z, [x0, #0x1C, mul vl]
            ld4h    { z5.h - z8.h }, p4/z, [x3, #-0x20, mul vl]
            ld4w    { z0.s - z3.s }, p1/z, [x2, #0x1C, mul vl]
            st2q    { z0.q, z1.q }, p3, [x0, #-0x10, mul vl]
            st3q    { z2.q - z4.q }, p3, [x4, #0x15, mul vl]
            st4q    { z7.q - z10.q }, p6, [x5, #0x1C, mul vl]
            stnt1b  { z1.b }, p2, [x3, #4, mul vl]
            stnt1d  { z8.d }, p7, [x6, #5, mul vl]
            stnt1h  { z9.h }, p1, [x0, #-5, mul vl]
            stnt1w  { z0.s }, p0, [x2, #-7, mul vl]
            st1d    { z1.d }, p2, [x3, #4, mul vl]
            st1w    { z3.q }, p4, [x5, #6, mul vl]
            st1d    { z2.q }, p1, [x0]
            st2b    { z0.b, z1.b }, p1, [x2, #-0x10, mul vl]
            st2d    { z5.d, z6.d }, p4, [x3, #-0x10, mul vl]
            st2h    { z6.h, z7.h }, p7, [x8, #-0x10, mul vl]
            st2w    { z8.s, z9.s }, p1, [x9, #-0x10, mul vl]
            st3b    { z7.b - z9.b }, p6, [x5, #-0x18, mul vl]
            st3d    { z2.d - z4.d }, p3, [x4, #-0x18, mul vl]
            st3h    { z1.h - z3.h }, p2, [x3, #-0x18, mul vl]
            st3w    { z1.s - z3.s }, p3, [x8, #-0x18, mul vl]
            st4b    { z0.b - z3.b }, p0, [x0, #-0x20, mul vl]
            st4d    { z2.d - z5.d }, p0, [x1, #-0x20, mul vl]
            st4h    { z3.h - z6.h }, p5, [x2, #-0x20, mul vl]
            st4w    { z0.s - z3.s }, p1, [x5, #0x1C, mul vl]
            st4w    { z31.s, z0.s, z1.s, z2.s }, p1, [x5, #0x1C, mul vl]
            st1b    { z4.b }, p1, [x2, x0]
            st1b    { z4.h }, p5, [x6, x2]
            st1b    { z6.s }, p5, [x7, x4]
						;; size=812 bbWeight=0.02 PerfScore 19.48
G_M65128_IG38:  ;; offset=0x1708
            st1b    { z4.d }, p0, [x1, x2]
            st1h    { z5.h }, p6, [x1, x2, lsl #1]
            st1h    { z1.s }, p2, [x3, x4, lsl #1]
            st1h    { z3.d }, p2, [x4, x0, lsl #1]
            st1w    { z0.s }, p1, [x2, x3, lsl #2]
            st1w    { z2.d }, p3, [x4, x5, lsl #2]
            st1d    { z0.d }, p1, [x2, z3.d, uxtw #3]
            st1d    { z0.d }, p1, [x2, z3.d, sxtw #3]
            st1h    { z3.s }, p1, [x5, z4.s, uxtw #1]
            st1h    { z3.s }, p1, [x5, z4.s, sxtw #1]
            st1w    { z0.s }, p3, [x1, z2.s, uxtw #2]
            st1w    { z0.s }, p3, [x1, z2.s, sxtw #2]
            st1d    { z3.d }, p1, [x2, z5.d, uxtw]
            st1d    { z3.d }, p1, [x2, z5.d, sxtw]
            st1h    { z2.d }, p3, [x1, z4.d, uxtw #1]
            st1h    { z2.d }, p3, [x1, z4.d, sxtw #1]
            st1w    { z1.d }, p4, [x2, z3.d, uxtw #2]
            st1w    { z1.d }, p4, [x2, z3.d, sxtw #2]
            st1h    { z1.d }, p5, [x1, z3.d, uxtw]
            st1h    { z1.d }, p5, [x1, z3.d, sxtw]
            st1w    { z0.d }, p2, [x3, z4.d, uxtw]
            st1w    { z0.d }, p2, [x3, z4.d, sxtw]
            st1h    { z7.s }, p5, [x4, z1.s, uxtw]
            st1h    { z7.s }, p5, [x4, z1.s, sxtw]
            st1w    { z1.s }, p2, [x3, z2.s, uxtw]
            st1w    { z1.s }, p2, [x3, z2.s, sxtw]
            st1b    { z4.d }, p2, [x0, z1.d, uxtw]
            st1b    { z4.d }, p2, [x0, z1.d, sxtw]
            st1b    { z1.s }, p4, [x3, z0.s, uxtw]
            st1b    { z1.s }, p4, [x3, z0.s, sxtw]
            st1b    { z3.b }, p2, [x1, #5, mul vl]
            st1b    { z3.h }, p2, [x1, #4, mul vl]
            st1h    { z0.h }, p3, [x4, #3, mul vl]
            st1b    { z3.s }, p2, [x1, #2, mul vl]
            st1h    { z0.s }, p3, [x4, #1, mul vl]
            st1b    { z3.d }, p2, [x1]
            st1h    { z0.d }, p3, [x4, #-2, mul vl]
            st1w    { z2.s }, p1, [x3, #5, mul vl]
            st1w    { z2.d }, p1, [x3, #1, mul vl]
            ld1b    { z1.d }, p2/z, [x3, z0.d, uxtw]
            ld1h    { z0.s }, p3/z, [x5, z4.s, uxtw #1]
            ld1sb   { z2.d }, p0/z, [x1, z5.d, uxtw]
            ld1sh   { z5.s }, p2/z, [x1, z3.s, uxtw #1]
            ld1w    { z4.s }, p5/z, [x6, z1.s, uxtw #2]
            ldff1b  { z3.d }, p2/z, [x1, z4.d, uxtw]
            ldff1h  { z2.s }, p1/z, [x3, z4.s, uxtw #1]
            ldff1sb { z1.d }, p4/z, [x3, z2.d, uxtw]
            ldff1sh { z3.s }, p2/z, [x4, z5.s, uxtw #1]
            ldff1w  { z4.s }, p1/z, [x2, z3.s, uxtw #2]
            ld1b    { z1.d }, p2/z, [x3, z0.d, sxtw]
            ld1h    { z0.s }, p3/z, [x5, z4.s, sxtw #1]
            ld1sb   { z2.d }, p0/z, [x1, z5.d, sxtw]
            ld1sh   { z5.s }, p2/z, [x1, z3.s, sxtw #1]
            ld1w    { z4.s }, p5/z, [x6, z1.s, sxtw #2]
            ldff1b  { z3.d }, p2/z, [x1, z4.d, sxtw]
            ldff1h  { z2.s }, p1/z, [x3, z4.s, sxtw #1]
            ldff1sb { z1.d }, p4/z, [x3, z2.d, sxtw]
            ldff1sh { z3.s }, p2/z, [x4, z5.s, sxtw #1]
            ldff1w  { z4.s }, p1/z, [x2, z3.s, sxtw #2]
            ld1b    { z5.s }, p4/z, [x3, z2.s, uxtw]
            ld1h    { z2.d }, p1/z, [x0, z1.d, uxtw #1]
            ld1sb   { z3.s }, p4/z, [x5, z1.s, uxtw]
            ld1sh   { z5.d }, p4/z, [x3, z2.d, uxtw #1]
            ld1w    { z4.d }, p3/z, [x2, z1.d, uxtw #2]
            ldff1b  { z3.s }, p2/z, [x1, z0.s, uxtw]
            ldff1h  { z4.d }, p5/z, [x1, z2.d, uxtw #1]
            ldff1sb { z2.s }, p1/z, [x4, z3.s, uxtw]
            ldff1sh { z2.d }, p3/z, [x4, z5.d, uxtw #1]
            ldff1w  { z5.d }, p0/z, [x1, z2.d, uxtw #2]
            ld1b    { z5.s }, p4/z, [x3, z2.s, sxtw]
            ld1h    { z2.d }, p1/z, [x0, z1.d, sxtw #1]
            ld1sb   { z3.s }, p4/z, [x5, z1.s, sxtw]
            ld1sh   { z5.d }, p4/z, [x3, z2.d, sxtw #1]
            ld1w    { z4.d }, p3/z, [x2, z1.d, sxtw #2]
            ldff1b  { z3.s }, p2/z, [x1, z0.s, sxtw]
            ldff1h  { z4.d }, p5/z, [x1, z2.d, sxtw #1]
            ldff1sb { z2.s }, p1/z, [x4, z3.s, sxtw]
            ldff1sh { z2.d }, p3/z, [x4, z5.d, sxtw #1]
            ldff1w  { z5.d }, p0/z, [x1, z2.d, sxtw #2]
            ld1h    { z0.d }, p5/z, [x4, z3.d, uxtw]
            ld1sh   { z3.d }, p0/z, [x1, z2.d, uxtw]
            ld1w    { z4.d }, p5/z, [x3, z1.d, uxtw]
            ldff1h  { z1.d }, p3/z, [x4, z5.d, uxtw]
            ldff1sh { z2.d }, p1/z, [x3, z4.d, uxtw]
            ldff1w  { z1.d }, p4/z, [x2, z3.d, uxtw]
            ld1h    { z0.d }, p5/z, [x4, z3.d, sxtw]
            ld1sh   { z3.d }, p0/z, [x1, z2.d, sxtw]
            ld1w    { z4.d }, p5/z, [x3, z1.d, sxtw]
            ldff1h  { z1.d }, p3/z, [x4, z5.d, sxtw]
            ldff1sh { z2.d }, p1/z, [x3, z4.d, sxtw]
            ldff1w  { z1.d }, p4/z, [x2, z3.d, sxtw]
            ld1h    { z0.s }, p1/z, [x2, z3.s, uxtw]
            ld1sh   { z2.s }, p3/z, [x5, z4.s, uxtw]
            ld1w    { z1.s }, p2/z, [x3, z5.s, uxtw]
            ldff1h  { z4.s }, p2/z, [x1, z3.s, uxtw]
            ldff1sh { z5.s }, p4/z, [x1, z2.s, uxtw]
            ldff1w  { z3.s }, p5/z, [x2, z1.s, uxtw]
            ld1h    { z0.s }, p1/z, [x2, z3.s, sxtw]
            ld1sh   { z2.s }, p3/z, [x5, z4.s, sxtw]
            ld1w    { z1.s }, p2/z, [x3, z5.s, sxtw]
            ldff1h  { z4.s }, p2/z, [x1, z3.s, sxtw]
            ldff1sh { z5.s }, p4/z, [x1, z2.s, sxtw]
            ldff1w  { z3.s }, p5/z, [x2, z1.s, sxtw]
            ld1d    { z4.d }, p2/z, [x1, z3.d, uxtw #3]
            ld1sw   { z2.d }, p3/z, [x5, z4.d, uxtw #2]
            ldff1d  { z1.d }, p2/z, [x3, z5.d, uxtw #3]
            ldff1sw { z3.d }, p4/z, [x1, z2.d, uxtw #2]
            ld1d    { z4.d }, p2/z, [x1, z3.d, sxtw #3]
            ld1sw   { z2.d }, p3/z, [x5, z4.d, sxtw #2]
            ldff1d  { z1.d }, p2/z, [x3, z5.d, sxtw #3]
            ldff1sw { z3.d }, p4/z, [x1, z2.d, sxtw #2]
            ld1sw   { z2.d }, p3/z, [x4, z5.d, uxtw]
            ldff1d  { z0.d }, p2/z, [x3, z6.d, uxtw]
            ldff1sw { z1.d }, p4/z, [x2, z0.d, uxtw]
            ld1sw   { z2.d }, p3/z, [x4, z5.d, sxtw]
            ldff1d  { z0.d }, p2/z, [x3, z6.d, sxtw]
            ldff1sw { z1.d }, p4/z, [x2, z0.d, sxtw]
            ld1d    { z0.d }, p3/z, [x5, z2.d, uxtw]
            ld1d    { z0.d }, p3/z, [x5, z2.d, sxtw]
            ld1b    { z4.d }, p5/z, [x6, z1.d]
            ld1h    { z5.d }, p4/z, [x3, z2.d, lsl #1]
            ld1sb   { z6.d }, p3/z, [x1, z4.d]
            ld1sh   { z2.d }, p3/z, [x4, z5.d, lsl #1]
            ld1w    { z0.d }, p3/z, [x2, z1.d, lsl #2]
            ldff1b  { z2.d }, p5/z, [x4, z3.d]
            ldff1h  { z0.d }, p2/z, [x6, z1.d, lsl #1]
            ldff1sb { z3.d }, p4/z, [x5, z2.d]
            ldff1sh { z5.d }, p4/z, [x3, z2.d, lsl #1]
            ldff1w  { z3.d }, p2/z, [x0, z1.d, lsl #2]
            ld1h    { z4.d }, p2/z, [x1, z3.d]
            ld1sh   { z2.d }, p3/z, [x4, z5.d]
            ld1w    { z1.d }, p2/z, [x3, z4.d]
            ldff1h  { z2.d }, p3/z, [x1, z5.d]
            ldff1sh { z1.d }, p4/z, [x3, z2.d]
            ldff1w  { z4.d }, p3/z, [x2, z1.d]
            ldnt1b  { z3.s }, p2/z, [z1.s, x0]
            ldnt1h  { z0.s }, p1/z, [z2.s, x3]
            ldnt1sb { z2.s }, p3/z, [z5.s, x4]
            ldnt1sh { z3.s }, p4/z, [z1.s, x2]
            ldnt1w  { z1.s }, p3/z, [z4.s]
            ldnt1b  { z0.d }, p2/z, [z4.d, x3]
            ldnt1h  { z1.d }, p4/z, [z3.d, x2]
            ldnt1sb { z2.d }, p3/z, [z4.d, x5]
            ldnt1sh { z3.d }, p2/z, [z1.d, x0]
            ldnt1w  { z4.d }, p1/z, [z2.d]
            ldff1d  { z1.d }, p2/z, [x3, x4, lsl #3]
            ldff1sw { z2.d }, p3/z, [x4, x5, lsl #2]
            ldff1sw { z2.d }, p3/z, [x4]
            ldff1sb { z1.h }, p0/z, [x2, x4]
            ldff1sb { z1.s }, p0/z, [x2, x4]
            ldff1sb { z1.d }, p0/z, [x2, x4]
            ldff1b  { z3.b }, p2/z, [x0, x1]
            ldff1b  { z3.h }, p2/z, [x0, x1]
            ldff1b  { z3.s }, p2/z, [x0, x1]
            ldff1b  { z3.d }, p2/z, [x0, x1]
            ldff1sh { z4.s }, p3/z, [x1, x2, lsl #1]
            ldff1sh { z4.d }, p3/z, [x1, x2, lsl #1]
            ldff1w  { z1.s }, p0/z, [x2, x3, lsl #2]
            ldff1w  { z1.d }, p0/z, [x2, x3, lsl #2]
            ldff1w  { z1.d }, p0/z, [x2]
            ldff1h  { z3.h }, p1/z, [x4, x0, lsl #1]
            ldff1h  { z3.s }, p1/z, [x4, x0, lsl #1]
            ldff1h  { z3.d }, p1/z, [x4, x0, lsl #1]
            ldff1h  { z3.d }, p1/z, [x4]
            ld1d    { z0.d }, p2/z, [x1, x3, lsl #3]
            ld1d    { z1.q }, p0/z, [x3, x4, lsl #3]
            ld1w    { z5.s }, p3/z, [x4, x1, lsl #2]
            ld1w    { z5.d }, p3/z, [x4, x1, lsl #2]
            ld1w    { z5.q }, p3/z, [x4, x1, lsl #2]
            ld1sw   { z0.d }, p1/z, [x2, x3, lsl #2]
            ld1sb   { z1.h }, p2/z, [x3, x4]
            ld1sb   { z1.s }, p2/z, [x3, x4]
            ld1sb   { z1.d }, p2/z, [x3, x4]
            ld1sh   { z2.s }, p3/z, [x4, x5, lsl #1]
            ld1sh   { z2.d }, p3/z, [x4, x5, lsl #1]
            ld1b    { z3.b }, p4/z, [x5, x6]
            ld1b    { z3.h }, p4/z, [x5, x6]
            ld1b    { z3.s }, p4/z, [x5, x6]
            ld1b    { z3.d }, p4/z, [x5, x6]
            ld1h    { z4.h }, p2/z, [x3, x1, lsl #1]
            ld1h    { z4.s }, p2/z, [x3, x1, lsl #1]
            ld1h    { z4.d }, p2/z, [x3, x1, lsl #1]
            ldnt1b  { z4.b }, p2/z, [x1, x3]
            ldnt1d  { z0.d }, p1/z, [x2, x4, lsl #3]
            ldnt1h  { z0.h }, p3/z, [x4, x5, lsl #1]
            ldnt1w  { z2.s }, p0/z, [x3, x1, lsl #2]
            ld1rob  { z0.b }, p1/z, [x3, x2]
            ld1rod  { z0.d }, p2/z, [x1, x3, lsl #3]
            ld1roh  { z4.h }, p3/z, [x2, x1, lsl #1]
            ld1row  { z1.s }, p3/z, [x2, x4, lsl #2]
            ld1rqb  { z3.b }, p1/z, [x4, x2]
            ld1rqd  { z2.d }, p3/z, [x1, x4, lsl #3]
            ld1rqh  { z1.h }, p2/z, [x3, x4, lsl #1]
            ld1rqw  { z0.s }, p1/z, [x2, x3, lsl #2]
            ld2q    { z0.q, z1.q }, p3/z, [x2, x1, lsl #4]
            ld3q    { z3.q - z5.q }, p4/z, [x1, x2, lsl #4]
            ld4q    { z5.q - z8.q }, p1/z, [x4, x3, lsl #4]
            ld2b    { z0.b, z1.b }, p1/z, [x2, x3]
            ld2d    { z7.d, z8.d }, p6/z, [x5, x4, lsl #3]
						;; size=796 bbWeight=0.02 PerfScore 30.34
G_M65128_IG39:  ;; offset=0x1A24
            ld2h    { z8.h, z9.h }, p5/z, [x9, x10, lsl #1]
            ld2w    { z6.s, z7.s }, p5/z, [x4, x7, lsl #2]
            ld3b    { z1.b - z3.b }, p0/z, [x3, x2]
            ld3d    { z4.d - z6.d }, p3/z, [x8, x1, lsl #3]
            ld3h    { z30.h, z31.h, z0.h }, p2/z, [x9, x4, lsl #1]
            ld3w    { z1.s - z3.s }, p3/z, [x2, x4, lsl #2]
            ld4b    { z0.b - z3.b }, p1/z, [x2, x3]
            ld4d    { z0.d - z3.d }, p3/z, [x2, x1, lsl #3]
            ld4h    { z13.h - z16.h }, p6/z, [x5, x4, lsl #1]
            ld4w    { z10.s - z13.s }, p3/z, [x2, x5, lsl #2]
            ld1d    { z0.d }, p4/z, [x3, z2.d, lsl #3]
            ld1sw   { z4.d }, p3/z, [x2, z1.d, lsl #2]
            ldff1d  { z5.d }, p6/z, [x7, z8.d, lsl #3]
            ldff1sw { z3.d }, p0/z, [x10, z9.d, lsl #2]
            ld1sw   { z1.d }, p2/z, [x3, z0.d]
            ldff1d  { z2.d }, p6/z, [x5, z4.d]
            ldff1sw { z3.d }, p4/z, [x6, z5.d]
            ld1d    { z0.d }, p1/z, [x3, z4.d]
            ld1q    { z0.q }, p1/z, [z2.d, x3]
            ld1q    { z0.q }, p1/z, [z2.d]
            ldnt1d  { z4.d }, p2/z, [z1.d, x3]
            ldnt1sw { z7.d }, p1/z, [z0.d, x1]
            ldnt1sw { z7.d }, p1/z, [z0.d]
            st1q    { z1.q }, p2, [z3.d, x4]
            st1q    { z1.q }, p2, [z3.d]
            stnt1b  { z0.s }, p2, [z3.s, x4]
            stnt1h  { z2.s }, p7, [z6.s, x5]
            stnt1w  { z1.s }, p3, [z2.s, x0]
            stnt1w  { z1.s }, p3, [z2.s]
            stnt1b  { z0.d }, p4, [z6.d, x8]
            stnt1h  { z5.d }, p3, [z1.d, x2]
            stnt1w  { z3.d }, p1, [z2.d, x0]
            stnt1b  { z0.d }, p4, [z6.d]
            stnt1d  { z1.d }, p3, [z4.d, x5]
            stnt1d  { z0.d }, p4, [z5.d]
            stnt1b  { z6.b }, p5, [x4, x3]
            stnt1d  { z7.d }, p6, [x5, x4, lsl #3]
            stnt1h  { z0.h }, p1, [x2, x3, lsl #1]
            stnt1w  { z0.s }, p5, [x6, x7, lsl #2]
            st2b    { z0.b, z1.b }, p1, [x2, x4]
            st2d    { z1.d, z2.d }, p7, [x6, x5, lsl #3]
            st2h    { z2.h, z3.h }, p3, [x5, x6, lsl #1]
            st2w    { z0.s, z1.s }, p2, [x8, x7, lsl #2]
            st3b    { z0.b - z2.b }, p1, [x3, x4]
            st3d    { z2.d - z4.d }, p3, [x4, x6, lsl #3]
            st3h    { z1.h - z3.h }, p0, [x3, x8, lsl #1]
            st3w    { z0.s - z2.s }, p1, [x2, x3, lsl #2]
            st4b    { z0.b - z3.b }, p6, [x5, x4]
            st4d    { z5.d - z8.d }, p2, [x1, x0, lsl #3]
            st4h    { z1.h - z4.h }, p0, [x9, x8, lsl #1]
            st4w    { z0.s - z3.s }, p1, [x4, x5, lsl #2]
            st1d    { z1.d }, p4, [x5, x6, lsl #3]
            st1w    { z2.q }, p1, [x8, x7, lsl #2]
            st1d    { z3.q }, p5, [x6, x1, lsl #3]
            st2q    { z0.q, z1.q }, p2, [x3, x5, lsl #4]
            st3q    { z1.q - z3.q }, p4, [x2, x8, lsl #4]
            st4q    { z4.q - z7.q }, p1, [x8, x2, lsl #4]
            st1d    { z0.d }, p3, [x2, z1.d, lsl #3]
            st1h    { z1.d }, p2, [x3, z4.d, lsl #1]
            st1w    { z2.d }, p3, [x4, z5.d, lsl #2]
            st1d    { z3.d }, p4, [x5, z6.d]
            st1h    { z1.d }, p4, [x3, z2.d]
            st1w    { z3.d }, p5, [x1, z0.d]
            st1b    { z6.d }, p3, [x0, z4.d]
            fcadd   z0.h, p1/m, z0.h, z2.h, #90
            fcadd   z0.h, p1/m, z0.h, z2.h, #270
            fcadd   z0.s, p1/m, z0.s, z2.s, #270
            fcadd   z0.d, p1/m, z0.d, z2.d, #270
            fcmla   z2.h, p1/m, z3.h, z4.h, #0
            fcmla   z0.s, p2/m, z1.s, z5.s, #90
            fcmla   z2.d, p3/m, z0.d, z6.d, #180
            fcmla   z2.d, p3/m, z0.d, z6.d, #270
            histcnt z0.s, p0/z, z1.s, z2.s
            histcnt z3.d, p7/z, z4.d, z5.d
            rax1    z0.d, z1.d, z2.d
            sm4ekey z3.s, z4.s, z5.s
            fcmeq   p2.h, p3/z, z4.h, #0.0
            fcmge   p1.s, p2/z, z3.s, #0.0
            fcmgt   p11.d, p5/z, z2.d, #0.0
            fcmle   p2.h, p7/z, z1.h, #0.0
            fcmlt   p3.s, p1/z, z0.s, #0.0
            fcmne   p1.d, p0/z, z5.d, #0.0
            fadd    z0.h, p0/m, z0.h, #0.5000
            fadd    z0.h, p1/m, z0.h, #1.0000
            fmax    z1.s, p0/m, z1.s, #0.0000
            fmax    z1.s, p0/m, z1.s, #1.0000
            fmaxnm  z3.d, p4/m, z3.d, #0.0000
            fmaxnm  z3.d, p4/m, z3.d, #1.0000
            fmin    z6.h, p5/m, z6.h, #0.0000
            fmin    z6.h, p5/m, z6.h, #1.0000
            fminnm  z2.s, p4/m, z2.s, #0.0000
            fminnm  z2.s, p4/m, z2.s, #1.0000
            fmul    z5.d, p1/m, z5.d, #0.5000
            fmul    z5.d, p1/m, z5.d, #2.0000
            fsub    z7.h, p2/m, z7.h, #0.5000
            fsub    z7.h, p2/m, z7.h, #1.0000
            fsubr   z6.s, p3/m, z6.s, #0.5000
            fsubr   z6.s, p3/m, z6.s, #1.0000
            ftmad   z0.h, z0.h, z2.h, #0
            ftmad   z3.s, z3.s, z5.s, #1
            ftmad   z4.d, z4.d, z2.d, #7
            flogb   z31.h, p7/m, z31.h
            flogb   z31.s, p7/m, z31.s
            flogb   z31.d, p7/m, z31.d
            bfmla   z1.h, p5/m, z4.h, z3.h
            bfmls   z0.h, p6/m, z1.h, z2.h
            fmad    z3.h, p4/m, z5.h, z6.h
            fmsb    z1.s, p2/m, z3.s, z7.s
            fnmad   z0.d, p5/m, z4.d, z3.d
            fnmsb   z5.s, p3/m, z1.s, z2.s
            ldr     p1, [x5]
            ldr     p1, [x5, #76, mul vl]
            ldr     p1, [x5, #-25, mul vl]
            ldr     p1, [x5, #-256, mul vl]
            ldr     p1, [x5, #255, mul vl]
            str     p3, [x1]
            str     p3, [x1, #221, mul vl]
            str     p3, [x1, #-117, mul vl]
            str     p3, [x1, #-256, mul vl]
            str     p3, [x1, #255, mul vl]
            ldr     z3, [x4]
            ldr     z3, [x4, #33, mul vl]
            ldr     z3, [x4, #-173, mul vl]
            ldr     z3, [x4, #-256, mul vl]
            ldr     z3, [x4, #255, mul vl]
            str     z2, [x3]
            str     z2, [x3, #71, mul vl]
            str     z2, [x3, #-165, mul vl]
            str     z2, [x3, #-256, mul vl]
            str     z2, [x3, #255, mul vl]
            prfb    pldl1keep, p1, [x2, z3.s, uxtw]
            prfd    pldl1keep, p6, [x5, z4.s, uxtw #3]
            prfh    pldl1keep, p7, [x8, z9.s, sxtw #1]
            prfw    pldl1keep, p0, [x2, z1.s, uxtw #2]
            prfb    pldl1strm, p1, [x2, z3.s, uxtw]
            prfb    pldl2keep, p1, [x2, z3.s, uxtw]
            prfb    pldl2strm, p1, [x2, z3.s, uxtw]
            prfb    pldl3keep, p1, [x2, z3.s, uxtw]
            prfb    pldl3strm, p1, [x2, z3.s, uxtw]
            prfb    pstl1keep, p1, [x2, z3.s, uxtw]
            prfb    pstl1strm, p1, [x2, z3.s, uxtw]
            prfb    pstl2keep, p1, [x2, z3.s, uxtw]
            prfb    pstl2strm, p1, [x2, z3.s, uxtw]
            prfb    pstl3keep, p1, [x2, z3.s, uxtw]
            prfb    pstl3strm, p1, [x2, z3.s, uxtw]
            prfb    #6, p1, [x2, z3.s, sxtw]
            prfb    #7, p1, [x2, z3.s, sxtw]
            prfb    #0xE, p1, [x2, z3.s, uxtw]
            prfb    #0xF, p1, [x2, z3.s, uxtw]
            prfb    pldl1keep, p1, [x2, z3.d, uxtw]
            prfd    pldl1keep, p6, [x5, z4.d, uxtw #3]
            prfh    pldl1keep, p7, [x8, z9.d, sxtw #1]
            prfw    pldl1keep, p0, [x2, z1.d, uxtw #2]
            prfb    pldl1keep, p0, [x1, z2.d]
            prfd    pldl1keep, p7, [x4, z3.d, lsl #3]
            prfh    pldl1keep, p6, [x5, z4.d, lsl #1]
            prfw    pldl1keep, p3, [x2, z1.d, lsl #2]
            prfb    pldl1keep, p1, [z2.s]
            prfd    pldl1keep, p4, [z3.s, #0xF8]
            prfh    pldl1keep, p0, [z4.s, #0x3E]
            prfw    pldl1keep, p7, [z5.s, #0x7C]
            prfb    pldl1keep, p1, [z2.d, #0x1F]
            prfd    pldl1keep, p4, [z3.d, #0xF8]
            prfh    pldl1keep, p0, [z4.d, #0x3E]
            prfw    pldl1keep, p7, [z5.d, #0x7C]
            prfb    pldl1keep, p2, [x3, #-32, mul vl]
            prfd    pldl1keep, p7, [x4, #31, mul vl]
            prfh    pldl1keep, p1, [x2]
            prfw    pldl1keep, p0, [x5, #-32, mul vl]
            prfb    pldl1keep, p2, [x3, #17, mul vl]
            prfb    pldl1keep, p0, [x1, x2]
            prfd    pldl1keep, p5, [x4, x3, lsl #3]
            prfh    pldl1keep, p6, [x7, x8, lsl #1]
            prfw    pldl1keep, p7, [x1, x9, lsl #2]
            ld1b    { z0.s }, p0/z, [z1.s]
            ld1sb   { z2.s }, p7/z, [z3.s, #5]
            ldff1b  { z4.s }, p3/z, [z1.s, #5]
            ldff1sb { z2.s }, p6/z, [z0.s, #31]
            ld1b    { z0.d }, p0/z, [z1.d]
            ld1sb   { z2.d }, p7/z, [z3.d, #5]
            ldff1b  { z4.d }, p3/z, [z1.d, #5]
            ldff1sb { z2.d }, p6/z, [z0.d, #31]
            ld1h    { z1.s }, p0/z, [z2.s]
            ld1sh   { z2.s }, p4/z, [z3.s, #2]
            ld1w    { z1.s }, p2/z, [z9.s, #0x7C]
            ldff1h  { z4.s }, p7/z, [z3.s, #6]
            ldff1sh { z3.s }, p5/z, [z4.s, #0x3E]
            ldff1w  { z2.s }, p1/z, [z3.s, #0x7C]
            ld1h    { z1.d }, p0/z, [z2.d]
            ld1sh   { z2.d }, p4/z, [z3.d, #2]
            ld1w    { z1.d }, p2/z, [z9.d, #0x7C]
            ldff1h  { z4.d }, p7/z, [z3.d, #6]
            ldff1sh { z3.d }, p5/z, [z4.d, #0x3E]
            ldff1w  { z2.d }, p1/z, [z3.d, #0x7C]
            ld1d    { z1.d }, p2/z, [z3.d]
            ld1sw   { z6.d }, p5/z, [z4.d]
            ldff1d  { z7.d }, p3/z, [z1.d, #0xF8]
            ldff1sw { z2.d }, p0/z, [z4.d, #0x7C]
            st1b    { z1.s }, p2, [z3.s]
            st1b    { z1.s }, p2, [z3.s, #31]
						;; size=800 bbWeight=0.02 PerfScore 16.53
G_M65128_IG40:  ;; offset=0x1D44
            st1h    { z5.s }, p3, [z2.s]
            st1h    { z5.s }, p3, [z2.s, #0x3E]
            st1w    { z5.s }, p4, [z1.s]
            st1w    { z5.s }, p4, [z1.s, #0x7C]
            st1b    { z1.d }, p2, [z3.d]
            st1b    { z1.d }, p2, [z3.d, #31]
            st1h    { z5.d }, p3, [z2.d]
            st1h    { z5.d }, p3, [z2.d, #0x3E]
            st1w    { z5.d }, p4, [z1.d]
            st1w    { z5.d }, p4, [z1.d, #0x7C]
            st1d    { z3.d }, p7, [z4.d]
            st1d    { z3.d }, p7, [z4.d, #0xF8]
            ld1rd   { z1.d }, p2/z, [x3, #0x1F8]
            ld1rsw  { z4.d }, p5/z, [x6, #0xFC]
            ld1rsh  { z0.s }, p1/z, [x2]
            ld1rw   { z5.s }, p4/z, [x3]
            ld1rsh  { z0.d }, p1/z, [x2, #0x7E]
            ld1rw   { z5.d }, p4/z, [x3, #0xFC]
            ld1rh   { z0.h }, p2/z, [x3]
            ld1rsb  { z6.h }, p5/z, [x4]
            ld1rh   { z5.s }, p4/z, [x3, #0x7E]
            ld1rsb  { z2.s }, p1/z, [x0, #0x3F]
            ld1rh   { z3.d }, p2/z, [x1, #0x7E]
            ld1rsb  { z4.d }, p5/z, [x6, #0x3F]
            ld1rb   { z1.b }, p2/z, [x3]
            ld1rb   { z5.h }, p4/z, [x3, #0x3F]
            ld1rb   { z6.s }, p7/z, [x8]
            ld1rb   { z1.b }, p0/z, [x9, #0x3F]
            frecpe  z0.h, z2.h
            frsqrte z5.s, z3.s
            frsqrte z9.d, z5.d
            sunpkhi z2.s, z4.h
            sunpklo z1.d, z5.s
            uunpkhi z5.h, z1.b
            uunpklo z8.d, z6.s
            rev     z2.b, z3.b
            rev     z2.h, z4.h
            rev     z7.s, z1.s
            rev     z2.d, z5.d
            mov     z0.b, w1
            mov     z2.h, w3
            mov     z1.s, w5
            mov     z4.d, sp
            mov     z4.b, w2
            mov     z4.h, w2
            mov     z1.s, w3
            mov     z5.d, sp
            mov     z2.d, x9
            fexpa   z0.h, z1.h
            fexpa   z3.s, z0.s
            fexpa   z1.d, z0.d
            movprfx z3, z5
            asr     z0.b, z0.b, #1
            asr     z31.b, z31.b, #7
            asr     z31.b, z31.b, #8
            asr     z0.b, z31.b, #5
            asr     z0.b, z31.b, #4
            asr     z0.h, z0.h, #1
            asr     z31.h, z31.h, #16
            asr     z0.h, z31.h, #9
            asr     z0.h, z31.h, #8
            asr     z0.s, z0.s, #1
            asr     z31.s, z31.s, #32
            asr     z0.s, z31.s, #15
            asr     z0.s, z31.s, #16
            asr     z0.d, z0.d, #1
            asr     z31.d, z31.d, #64
            asr     z0.d, z31.d, #33
            asr     z0.d, z31.d, #32
            lsl     z0.b, z0.b, #0
            lsl     z31.b, z31.b, #7
            lsl     z0.b, z31.b, #5
            lsl     z0.b, z31.b, #4
            lsl     z0.h, z0.h, #0
            lsl     z31.h, z31.h, #15
            lsl     z0.h, z31.h, #9
            lsl     z0.h, z31.h, #8
            lsl     z0.s, z0.s, #0
            lsl     z31.s, z31.s, #31
            lsl     z0.s, z31.s, #15
            lsl     z0.s, z31.s, #16
            lsl     z0.d, z0.d, #0
            lsl     z31.d, z31.d, #63
            lsl     z0.d, z31.d, #33
            lsl     z0.d, z31.d, #32
            lsr     z0.b, z0.b, #1
            lsr     z31.b, z31.b, #8
            lsr     z0.b, z31.b, #5
            lsr     z0.b, z31.b, #4
            lsr     z0.h, z0.h, #1
            lsr     z31.h, z31.h, #16
            lsr     z0.h, z31.h, #9
            lsr     z0.h, z31.h, #8
            lsr     z0.s, z0.s, #1
            lsr     z31.s, z31.s, #32
            lsr     z0.s, z31.s, #15
            lsr     z0.s, z31.s, #16
            lsr     z0.d, z0.d, #1
            lsr     z31.d, z31.d, #64
            lsr     z0.d, z31.d, #33
            lsr     z0.d, z31.d, #32
            sli     z0.b, z0.b, #0
            sli     z31.b, z31.b, #7
            sli     z0.b, z31.b, #3
            sli     z0.b, z31.b, #4
            sli     z0.h, z0.h, #0
            sli     z31.h, z31.h, #15
            sli     z0.h, z31.h, #7
            sli     z0.h, z31.h, #8
            sli     z0.s, z0.s, #0
            sli     z31.s, z31.s, #31
            sli     z0.s, z31.s, #17
            sli     z0.s, z31.s, #16
            sli     z0.d, z0.d, #0
            sli     z31.d, z31.d, #63
            sli     z0.d, z31.d, #31
            sli     z0.d, z31.d, #32
            sri     z0.b, z0.b, #1
            sri     z31.b, z31.b, #8
            sri     z0.b, z31.b, #3
            sri     z0.b, z31.b, #4
            sri     z0.h, z0.h, #1
            sri     z31.h, z31.h, #16
            sri     z0.h, z31.h, #7
            sri     z0.h, z31.h, #8
            sri     z0.s, z0.s, #1
            sri     z31.s, z31.s, #32
            sri     z0.s, z31.s, #17
            sri     z0.s, z31.s, #16
            sri     z0.d, z0.d, #1
            sri     z31.d, z31.d, #64
            sri     z0.d, z31.d, #31
            sri     z0.d, z31.d, #32
            srsra   z0.b, z0.b, #1
            srsra   z31.b, z31.b, #8
            srsra   z0.b, z31.b, #3
            srsra   z0.b, z31.b, #4
            srsra   z0.h, z0.h, #1
            srsra   z31.h, z31.h, #16
            srsra   z0.h, z31.h, #7
            srsra   z0.h, z31.h, #8
            srsra   z0.s, z0.s, #1
            srsra   z31.s, z31.s, #32
            srsra   z0.s, z31.s, #17
            srsra   z0.s, z31.s, #16
            srsra   z0.d, z0.d, #1
            srsra   z31.d, z31.d, #64
            srsra   z0.d, z31.d, #31
            srsra   z0.d, z31.d, #32
            ssra    z0.b, z0.b, #1
            ssra    z31.b, z31.b, #8
            ssra    z0.b, z31.b, #3
            ssra    z0.b, z31.b, #4
            ssra    z0.h, z0.h, #1
            ssra    z31.h, z31.h, #16
            ssra    z0.h, z31.h, #7
            ssra    z0.h, z31.h, #8
            ssra    z0.s, z0.s, #1
            ssra    z31.s, z31.s, #32
            ssra    z0.s, z31.s, #17
            ssra    z0.s, z31.s, #16
            ssra    z0.d, z0.d, #1
            ssra    z31.d, z31.d, #64
            ssra    z0.d, z31.d, #31
            ssra    z0.d, z31.d, #32
            ursra   z0.b, z0.b, #1
            ursra   z31.b, z31.b, #8
            ursra   z0.b, z31.b, #3
            ursra   z0.b, z31.b, #4
            ursra   z0.h, z0.h, #1
            ursra   z31.h, z31.h, #16
            ursra   z0.h, z31.h, #7
            ursra   z0.h, z31.h, #8
            ursra   z0.s, z0.s, #1
            ursra   z31.s, z31.s, #32
            ursra   z0.s, z31.s, #17
            ursra   z0.s, z31.s, #16
            ursra   z0.d, z0.d, #1
            ursra   z31.d, z31.d, #64
            ursra   z0.d, z31.d, #31
            ursra   z0.d, z31.d, #32
            usra    z0.b, z0.b, #1
            usra    z31.b, z31.b, #8
            usra    z0.b, z31.b, #3
            usra    z0.b, z31.b, #4
            usra    z0.h, z0.h, #1
            usra    z31.h, z31.h, #16
            usra    z0.h, z31.h, #7
            usra    z0.h, z31.h, #8
            usra    z0.s, z0.s, #1
            usra    z31.s, z31.s, #32
            usra    z0.s, z31.s, #17
            usra    z0.s, z31.s, #16
            usra    z0.d, z0.d, #1
            usra    z31.d, z31.d, #64
            usra    z0.d, z31.d, #31
            usra    z0.d, z31.d, #32
            dupq    z0.b, z0.b[0]
            dupq    z21.b, z10.b[10]
            dupq    z31.b, z31.b[15]
            dupq    z0.h, z0.h[0]
            dupq    z21.h, z10.h[5]
            dupq    z31.h, z31.h[7]
            dupq    z0.s, z0.s[0]
            dupq    z21.s, z10.s[2]
            dupq    z31.s, z31.s[3]
            dupq    z0.d, z0.d[0]
            dupq    z31.d, z31.d[1]
            extq    z0.b, z0.b, z0.b, #0
            extq    z31.b, z31.b, z31.b, #15
						;; size=840 bbWeight=0.02 PerfScore 10.16
G_M65128_IG41:  ;; offset=0x208C
            nop     
            nop     
            nop     
            nop     
            b       G_M65128_IG23
						;; size=20 bbWeight=0.02 PerfScore 0.07

; Total bytes of code 8352, prolog size 44, PerfScore 280.61, instruction count 2088, allocated bytes for code 8352 (MethodHash=5a2e0197) for method System.Text.Unicode.Utf16Utility:GetPointerToFirstInvalidChar(ulong,int,byref,byref):ulong (FullOpts)
; ============================================================

************************** System.Text.Unicode.Utf16Utility:GetPointerToFirstInvalidChar size 0x20A0 **************************

Base address : 0000020A3FFE0920h (RW: 0000020A3FFE0920h)
0: fd 7b b9 a9	stp	x29, x30, [sp, #-112]!
4: e8 a7 02 6d	stp	d8, d9, [sp, #40]
8: f3 d3 03 a9	stp	x19, x20, [sp, #56]
c: f5 db 04 a9	stp	x21, x22, [sp, #72]
10: f7 e3 05 a9	stp	x23, x24, [sp, #88]
14: f9 37 00 f9	str	x25, [sp, #104]
18: fd 03 00 91	mov	x29, sp
1c: f3 03 00 aa	mov	x19, x0
20: f4 03 01 2a	mov	w20, w1
24: f5 03 02 aa	mov	x21, x2
28: f6 03 03 aa	mov	x22, x3
2c: 94 01 f8 36	tbz	w20, #31, #48
30: 00 05 81 d2	mov	x0, #2088
34: c0 6e a9 f2	movk	x0, #19318, lsl #16
38: c0 3b c0 f2	movk	x0, #478, lsl #32
3c: 01 01 80 d2	mov	x1, #8
40: c1 6e a9 f2	movk	x1, #19318, lsl #16
44: c1 3b c0 f2	movk	x1, #478, lsl #32
48: 02 d8 9e d2	mov	x2, #63168
4c: a2 ea a1 f2	movk	x2, #3925, lsl #16
50: 62 ff cf f2	movk	x2, #32763, lsl #32
54: 42 00 40 f9	ldr	x2, [x2]
58: 40 00 3f d6	blr	x2
5c: 13 02 00 b5	cbnz	x19, #64
60: 9f 02 00 71	cmp	w20, #0
64: e0 17 9f 9a	cset	x0, eq
68: 1f 1c 00 72	tst	w0, #0xff
6c: 81 01 00 54	b.ne	#48
70: 00 11 81 d2	mov	x0, #2184
74: c0 6e a9 f2	movk	x0, #19318, lsl #16
78: c0 3b c0 f2	movk	x0, #478, lsl #32
7c: 01 01 80 d2	mov	x1, #8
80: c1 6e a9 f2	movk	x1, #19318, lsl #16
84: c1 3b c0 f2	movk	x1, #478, lsl #32
88: 02 d8 9e d2	mov	x2, #63168
8c: a2 ea a1 f2	movk	x2, #3925, lsl #16
90: 62 ff cf f2	movk	x2, #32763, lsl #32
94: 42 00 40 f9	ldr	x2, [x2]
98: 40 00 3f d6	blr	x2
9c: e1 03 14 2a	mov	w1, w20
a0: e0 03 13 aa	mov	x0, x19
a4: 02 cc 9b d2	mov	x2, #56928
a8: 42 f0 a1 f2	movk	x2, #3970, lsl #16
ac: 62 ff cf f2	movk	x2, #32763, lsl #32
b0: 42 00 40 f9	ldr	x2, [x2]
b4: 40 00 3f d6	blr	x2
b8: f7 03 00 2a	mov	w23, w0
bc: b7 00 f8 37	tbnz	w23, #31, #20
c0: ff 02 14 6b	cmp	w23, w20
c4: e0 c7 9f 9a	cset	x0, le
c8: 1f 1c 00 72	tst	w0, #0xff
cc: 81 01 00 54	b.ne	#48
d0: 00 01 80 d2	mov	x0, #8
d4: c0 6e a9 f2	movk	x0, #19318, lsl #16
d8: c0 3b c0 f2	movk	x0, #478, lsl #32
dc: 01 01 80 d2	mov	x1, #8
e0: c1 6e a9 f2	movk	x1, #19318, lsl #16
e4: c1 3b c0 f2	movk	x1, #478, lsl #32
e8: 02 d8 9e d2	mov	x2, #63168
ec: a2 ea a1 f2	movk	x2, #3925, lsl #16
f0: 62 ff cf f2	movk	x2, #32763, lsl #32
f4: 42 00 40 f9	ldr	x2, [x2]
f8: 40 00 3f d6	blr	x2
fc: e0 7e 7f d3	ubfiz	x0, x23, #1, #32
100: 73 02 00 8b	add	x19, x19, x0
104: 94 02 17 4b	sub	w20, w20, w23
108: 74 01 00 35	cbnz	w20, #44
10c: bf 02 00 f9	str	xzr, [x21]
110: df 02 00 b9	str	wzr, [x22]
114: e0 03 13 aa	mov	x0, x19
118: f9 37 40 f9	ldr	x25, [sp, #104]
11c: f7 e3 45 a9	ldp	x23, x24, [sp, #88]
120: f5 db 44 a9	ldp	x21, x22, [sp, #72]
124: f3 d3 43 a9	ldp	x19, x20, [sp, #56]
128: e8 a7 42 6d	ldp	d8, d9, [sp, #40]
12c: fd 7b c7 a8	ldp	x29, x30, [sp], #112
130: c0 03 5f d6	ret
134: f7 03 1f aa	mov	x23, xzr
138: f8 03 1f 2a	mov	w24, wzr
13c: 80 7e 7f d3	ubfiz	x0, x20, #1, #32
140: 79 02 00 8b	add	x25, x19, x0
144: 9f 22 00 71	cmp	w20, #8
148: ab 09 00 54	b.lt	#308
14c: 08 a5 00 4f	movi	v8.8h, #8, lsl #8
150: 34 43 00 d1	sub	x20, x25, #16
154: 9f 02 13 eb	cmp	x20, x19
158: c2 01 00 54	b.hs	#56
15c: 00 01 80 d2	mov	x0, #8
160: c0 6e a9 f2	movk	x0, #19318, lsl #16
164: c0 3b c0 f2	movk	x0, #478, lsl #32
168: 01 01 80 d2	mov	x1, #8
16c: c1 6e a9 f2	movk	x1, #19318, lsl #16
170: c1 3b c0 f2	movk	x1, #478, lsl #32
174: 02 d8 9e d2	mov	x2, #63168
178: a2 ea a1 f2	movk	x2, #3925, lsl #16
17c: 62 ff cf f2	movk	x2, #32763, lsl #32
180: 42 00 40 f9	ldr	x2, [x2]
184: 09 45 08 6e	mov	v9.d[0], v8.d[1]
188: 40 00 3f d6	blr	x2
18c: 28 05 18 6e	mov	v8.d[1], v9.d[0]
190: 70 02 c0 3d	ldr	q16, [x19]
194: 11 3e 68 6e	cmhs	v17.8h, v16.8h, v8.8h
198: 12 84 04 4f	movi	v18.8h, #128
19c: 12 3e 72 6e	cmhs	v18.8h, v16.8h, v18.8h
1a0: 13 04 00 4f	movi	v19.4s, #0
1a4: 72 86 72 6e	sub	v18.8h, v19.8h, v18.8h
1a8: 51 86 71 6e	sub	v17.8h, v18.8h, v17.8h
1ac: e0 03 1f 2a	mov	w0, wzr
1b0: 20 3e 08 4e	mov	x0, v17.d[0]
1b4: 21 00 80 52	mov	w1, #1
1b8: 21 3e 18 4e	mov	x1, v17.d[1]
1bc: 20 00 00 8b	add	x0, x1, x0
1c0: 01 fc 60 d3	lsr	x1, x0, #32
1c4: 00 00 01 0b	add	w0, w0, w1
1c8: 01 3c 00 53	uxth	w1, w0
1cc: 20 40 40 0b	add	w0, w1, w0, lsr #16
1d0: 11 a7 06 4f	movi	v17.8h, #216, lsl #8
1d4: 10 86 71 6e	sub	v16.8h, v16.8h, v17.8h
1d8: 11 35 70 6e	cmhi	v17.8h, v8.8h, v16.8h
1dc: 32 a6 b1 6e	umaxp	v18.4s, v17.4s, v17.4s
1e0: 41 3e 08 4e	mov	x1, v18.d[0]
1e4: 3f 00 00 f1	cmp	x1, #0
1e8: e0 03 00 54	b.eq	#124
1ec: 92 a4 00 4f	movi	v18.8h, #4, lsl #8
1f0: 50 36 70 6e	cmhi	v16.8h, v18.8h, v16.8h
1f4: 31 1e 70 4e	bic	v17.16b, v17.16b, v16.16b
1f8: 21 3e 02 0e	umov	w1, v17.h[0]
1fc: 41 04 00 35	cbnz	w1, #136
200: e1 03 1f 2a	mov	w1, wzr
204: e2 03 1f 2a	mov	w2, wzr
208: a3 43 00 91	add	x3, x29, #16
20c: 70 00 80 3d	str	q16, [x3]
210: 63 78 62 78	ldrh	w3, [x3, x2, lsl #1]
214: 21 00 03 4b	sub	w1, w1, w3
218: 21 3c 00 53	uxth	w1, w1
21c: 42 04 00 11	add	w2, w2, #1
220: 44 7c 00 13	asr	w4, w2, #0
224: a5 43 00 91	add	x5, x29, #16
228: b1 00 80 3d	str	q17, [x5]
22c: a5 78 64 78	ldrh	w5, [x5, x4, lsl #1]
230: bf 00 03 6b	cmp	w5, w3
234: 41 02 00 54	b.ne	#72
238: 5f 1c 00 71	cmp	w2, #7
23c: 6b fe ff 54	b.lt	#-52
240: e2 00 80 52	mov	w2, #7
244: 02 3e 1e 0e	umov	w2, v16.h[7]
248: 62 00 00 34	cbz	w2, #12
24c: 73 0a 00 d1	sub	x19, x19, #2
250: 00 08 00 51	sub	w0, w0, #2
254: e1 03 01 2a	mov	w1, w1
258: 18 03 01 4b	sub	w24, w24, w1
25c: e2 02 01 cb	sub	x2, x23, x1
260: 57 00 01 cb	sub	x23, x2, x1
264: f7 42 20 8b	add	x23, x23, w0, uxtw
268: 73 42 00 91	add	x19, x19, #16
26c: 7f 02 14 eb	cmp	x19, x20
270: 09 f9 ff 54	b.ls	#-224
274: 02 00 00 14	b	#8
278: 73 0a 00 91	add	x19, x19, #2
27c: 7f 02 19 eb	cmp	x19, x25
280: 63 01 00 54	b.lo	#44
284: b7 02 00 f9	str	x23, [x21]
288: d8 02 00 b9	str	w24, [x22]
28c: e0 03 13 aa	mov	x0, x19
290: f9 37 40 f9	ldr	x25, [sp, #104]
294: f7 e3 45 a9	ldp	x23, x24, [sp, #88]
298: f5 db 44 a9	ldp	x21, x22, [sp, #72]
29c: f3 d3 43 a9	ldp	x19, x20, [sp, #56]
2a0: e8 a7 42 6d	ldp	d8, d9, [sp, #40]
2a4: fd 7b c7 a8	ldp	x29, x30, [sp], #112
2a8: c0 03 5f d6	ret
2ac: 60 02 40 79	ldrh	w0, [x19]
2b0: 1f fc 01 71	cmp	w0, #127
2b4: 29 fe ff 54	b.ls	#-60
2b8: e1 17 15 32	mov	w1, #129024
2bc: 01 00 01 0b	add	w1, w0, w1
2c0: 21 7c 10 53	lsr	w1, w1, #16
2c4: f7 42 21 8b	add	x23, x23, w1, uxtw
2c8: e1 ff 9a 12	mov	w1, #-55296
2cc: 00 00 01 0b	add	w0, w0, w1
2d0: 1f fc 1f 71	cmp	w0, #2047
2d4: 28 fd ff 54	b.hi	#-92
2d8: f7 0a 00 d1	sub	x23, x23, #2
2dc: 20 03 13 cb	sub	x0, x25, x19
2e0: 1f 10 00 f1	cmp	x0, #4
2e4: 03 fd ff 54	b.lo	#-96
2e8: 60 02 40 b9	ldr	w0, [x19]
2ec: 01 00 85 52	mov	w1, #10240
2f0: e1 7f a4 72	movk	w1, #9215, lsl #16
2f4: 00 00 01 0b	add	w0, w0, w1
2f8: 1f 94 06 72	tst	w0, #0xfc00fc00
2fc: 41 fc ff 54	b.ne	#-120
300: 18 07 00 51	sub	w24, w24, #1
304: f7 0a 00 91	add	x23, x23, #2
308: 73 0a 00 91	add	x19, x19, #2
30c: 60 07 00 14	b	#7552
310: 1f 20 03 d5	nop
314: 40 04 1a 04	and	z0.b, p1/m, z0.b, z2.b
318: a3 10 5b 04	bic	z3.h, p4/m, z3.h, z5.h
31c: 0e 16 99 04	eor	z14.s, p5/m, z14.s, z16.s
320: fd 1f d8 04	orr	z29.d, p7/m, z29.d, z31.d
324: e5 18 00 04	add	z5.b, p6/m, z5.b, z7.b
328: af 1f 41 04	sub	z15.h, p7/m, z15.h, z29.h
32c: a2 01 83 04	subr	z2.s, p0/m, z2.s, z13.s
330: 58 00 08 04	smax	z24.b, p0/m, z24.b, z2.b
334: 69 07 4a 04	smin	z9.h, p1/m, z9.h, z27.h
338: c5 08 0c 04	sabd	z5.b, p2/m, z5.b, z6.b
33c: 37 0d 8d 04	uabd	z23.s, p3/m, z23.s, z9.s
340: 4f 10 89 04	umax	z15.s, p4/m, z15.s, z2.s
344: 0c 1c cb 04	umin	z12.d, p7/m, z12.d, z0.d
348: 65 04 d0 04	mul	z5.d, p1/m, z5.d, z3.d
34c: b1 14 92 04	smulh	z17.s, p5/m, z17.s, z5.s
350: 0c 0b 13 04	umulh	z12.b, p2/m, z12.b, z24.b
354: a5 82 90 04	asr	z5.s, p0/m, z5.s, z21.s
358: 81 9e 14 04	asrr	z1.b, p7/m, z1.b, z20.b
35c: 00 88 53 04	lsl	z0.h, p2/m, z0.h, z0.h
360: fb 9b d7 04	lslr	z27.d, p6/m, z27.d, z31.d
364: c5 94 11 04	lsr	z5.b, p5/m, z5.b, z6.b
368: 2f 92 95 04	lsrr	z15.s, p4/m, z15.s, z17.s
36c: 4f 81 10 44	shadd	z15.b, p0/m, z15.b, z10.b
370: 70 85 52 44	shsub	z16.h, p1/m, z16.h, z11.h
374: 91 89 96 44	shsubr	z17.s, p2/m, z17.s, z12.s
378: b2 8d d4 44	srhadd	z18.d, p3/m, z18.d, z13.d
37c: d3 91 11 44	uhadd	z19.b, p4/m, z19.b, z14.b
380: f4 95 53 44	uhsub	z20.h, p5/m, z20.h, z15.h
384: 15 9a 97 44	uhsubr	z21.s, p6/m, z21.s, z16.s
388: 36 9e d5 44	urhadd	z22.d, p7/m, z22.d, z17.d
38c: 57 ba 11 44	addp	z23.b, p6/m, z23.b, z18.b
390: 78 b6 54 44	smaxp	z24.h, p5/m, z24.h, z19.h
394: 99 b2 96 44	sminp	z25.s, p4/m, z25.s, z20.s
398: ba ae d5 44	umaxp	z26.d, p3/m, z26.d, z21.d
39c: db aa 17 44	uminp	z27.b, p2/m, z27.b, z22.b
3a0: fc 86 18 44	sqadd	z28.b, p1/m, z28.b, z23.b
3a4: 1d 83 5a 44	sqsub	z29.h, p0/m, z29.h, z24.h
3a8: 3e 87 5e 44	sqsubr	z30.h, p1/m, z30.h, z25.h
3ac: 5f 8b 1c 44	suqadd	z31.b, p2/m, z31.b, z26.b
3b0: 60 8f 99 44	uqadd	z0.s, p3/m, z0.s, z27.s
3b4: 81 93 db 44	uqsub	z1.d, p4/m, z1.d, z28.d
3b8: a2 97 1f 44	uqsubr	z2.b, p5/m, z2.b, z29.b
3bc: c3 9b 1d 44	usqadd	z3.b, p6/m, z3.b, z30.b
3c0: e4 9f 0a 44	sqrshl	z4.b, p7/m, z4.b, z31.b
3c4: c5 83 4e 44	sqrshlr	z5.h, p0/m, z5.h, z30.h
3c8: a6 87 88 44	sqshl	z6.s, p1/m, z6.s, z29.s
3cc: 87 8b cc 44	sqshlr	z7.d, p2/m, z7.d, z28.d
3d0: 68 8f 02 44	srshl	z8.b, p3/m, z8.b, z27.b
3d4: 49 93 46 44	srshlr	z9.h, p4/m, z9.h, z26.h
3d8: 2a 97 8b 44	uqrshl	z10.s, p5/m, z10.s, z25.s
3dc: 0b 9b cf 44	uqrshlr	z11.d, p6/m, z11.d, z24.d
3e0: ec 9e 09 44	uqshl	z12.b, p7/m, z12.b, z23.b
3e4: cd 82 4d 44	uqshlr	z13.h, p0/m, z13.h, z22.h
3e8: ae 86 83 44	urshl	z14.s, p1/m, z14.s, z21.s
3ec: 8f 8a c7 44	urshlr	z15.d, p2/m, z15.d, z20.d
3f0: 23 09 94 04	sdiv	z3.s, p2/m, z3.s, z9.s
3f4: bf 0f d6 04	sdivr	z31.d, p3/m, z31.d, z29.d
3f8: 01 00 95 04	udiv	z1.s, p0/m, z1.s, z0.s
3fc: ed 1d d7 04	udivr	z13.d, p7/m, z13.d, z15.d
400: ed 23 10 04	movprfx	z13.b, p0/z, z31.b
404: 0e 24 50 04	movprfx	z14.h, p1/z, z0.h
408: 2d 28 90 04	movprfx	z13.s, p2/z, z1.s
40c: 4f 2c d0 04	movprfx	z15.d, p3/z, z2.d
410: 70 30 11 04	movprfx	z16.b, p4/m, z3.b
414: 91 35 51 04	movprfx	z17.h, p5/m, z12.h
418: a0 39 91 04	movprfx	z0.s, p6/m, z13.s
41c: df 3e d1 04	movprfx	z31.d, p7/m, z22.d
420: e0 81 00 04	asr	z0.b, p0/m, z0.b, #1
424: c1 85 00 04	asr	z1.b, p1/m, z1.b, #2
428: a2 89 00 04	asr	z2.b, p2/m, z2.b, #3
42c: 23 8d 00 04	asr	z3.b, p3/m, z3.b, #7
430: 04 9d 00 04	asr	z4.b, p7/m, z4.b, #8
434: e5 83 04 04	asrd	z5.h, p0/m, z5.h, #1
438: 06 93 04 04	asrd	z6.h, p4/m, z6.h, #8
43c: 27 9e 04 04	asrd	z7.h, p7/m, z7.h, #15
440: 1f 9a 04 04	asrd	z31.h, p6/m, z31.h, #16
444: e8 97 41 04	lsr	z8.s, p5/m, z8.s, #1
448: 69 9b 41 04	lsr	z9.s, p6/m, z9.s, #5
44c: ff 9d 41 04	lsr	z31.s, p7/m, z31.s, #17
450: 20 80 41 04	lsr	z0.s, p0/m, z0.s, #31
454: 05 84 41 04	lsr	z5.s, p1/m, z5.s, #32
458: e6 8f cc 04	srshr	z6.d, p3/m, z6.d, #1
45c: a7 93 cc 04	srshr	z7.d, p4/m, z7.d, #3
460: 75 95 cc 04	srshr	z21.d, p5/m, z21.d, #21
464: 36 98 8c 04	srshr	z22.d, p6/m, z22.d, #63
468: 1f 9c 8d 04	urshr	z31.d, p7/m, z31.d, #64
46c: 01 81 03 04	lsl	z1.b, p0/m, z1.b, #0
470: 23 85 03 04	lsl	z3.b, p1/m, z3.b, #1
474: e5 89 03 04	lsl	z5.b, p2/m, z5.b, #7
478: 07 8e 06 04	sqshl	z7.h, p3/m, z7.h, #0
47c: e9 92 06 04	sqshl	z9.h, p4/m, z9.h, #7
480: 2b 83 06 04	sqshl	z11.h, p0/m, z11.h, #9
484: ed 9f 06 04	sqshl	z13.h, p7/m, z13.h, #15
488: 0e 98 4f 04	sqshlu	z14.s, p6/m, z14.s, #0
48c: 8f 81 4f 04	sqshlu	z15.s, p0/m, z15.s, #12
490: f0 85 4f 04	sqshlu	z16.s, p1/m, z16.s, #15
494: e0 8b 4f 04	sqshlu	z0.s, p2/m, z0.s, #31
498: 11 9c 87 04	uqshl	z17.d, p7/m, z17.d, #0
49c: 52 82 87 04	uqshl	z18.d, p0/m, z18.d, #18
4a0: 13 8c c7 04	uqshl	z19.d, p3/m, z19.d, #32
4a4: e0 8b c7 04	uqshl	z0.d, p2/m, z0.d, #63
4a8: 04 8f 18 04	asr	z4.b, p3/m, z4.b, z24.d
4ac: 73 9c 5b 04	lsl	z19.h, p7/m, z19.h, z3.d
4b0: 00 80 99 04	lsr	z0.s, p0/m, z0.s, z0.d
Decode Failure @ offset      4b4
4b4: 82 39 2a 05
Decode Failure @ offset      4b8
4b8: 47 38 2c 05
Decode Failure @ offset      4bc
4bc: ef 38 ee 05
Decode Failure @ offset      4c0
4c0: 07 3a a8 05
Decode Failure @ offset      4c4
4c4: e0 3b 2e 05
Decode Failure @ offset      4c8
4c8: 21 38 2c 05
Decode Failure @ offset      4cc
4cc: 23 39 6e 05
Decode Failure @ offset      4d0
4d0: 8a 38 68 05
Decode Failure @ offset      4d4
4d4: 8b 39 2b 05
Decode Failure @ offset      4d8
4d8: e2 38 69 05
Decode Failure @ offset      4dc
4dc: 06 39 ef 05
Decode Failure @ offset      4e0
4e0: e9 38 a9 05
Decode Failure @ offset      4e4
4e4: 88 38 2f 05
Decode Failure @ offset      4e8
4e8: 25 39 2d 05
Decode Failure @ offset      4ec
4ec: 4e 38 6f 05
Decode Failure @ offset      4f0
4f0: e3 39 69 05
4f4: 41 40 34 05	rev	p1.b, p2.b
4f8: a4 40 74 05	rev	p4.h, p5.h
4fc: e3 40 b4 05	rev	p3.s, p7.s
500: c0 40 f4 05	rev	p0.d, p6.d
504: e1 40 31 05	punpkhi	p1.h, p7.b
508: 65 40 30 05	punpklo	p5.h, p3.b
50c: ff 9f 28 05	clasta	z31.b, p7, z31.b, z31.b
510: de 9b e9 05	clastb	z30.d, p6, z30.d, z30.d
514: ec 85 6a 05	clasta	h12, p1, h12, z15.h
518: 0d 8a ab 05	clastb	s13, p2, s13, z16.s
51c: 2e 82 eb 05	clastb	d14, p0, d14, z17.d
520: 00 a0 30 05	clasta	w0, p0, w0, z0.b
524: 61 a8 70 05	clasta	w1, p2, w1, z3.h
528: 97 b5 b1 05	clastb	w23, p5, w23, z12.s
52c: 23 b9 f1 05	clastb	x3, p6, x3, z9.d
Decode Failure @ offset      530
530: c1 80 2e 05
Decode Failure @ offset      534
534: 85 9d 2e 05
538: 0f a0 0a 24	cmpeq	p15.b, p0/z, z0.b, z10.b
53c: 4e 85 57 24	cmpge	p14.h, p1/z, z10.h, z23.h
540: 9d 8a 98 24	cmpgt	p13.s, p2/z, z20.s, z24.s
544: dc 0f d9 24	cmphi	p12.d, p3/z, z30.d, z25.d
548: 2b 10 1a 24	cmphs	p11.b, p4/z, z1.b, z26.b
54c: 6a 97 4b 24	cmpge	p10.h, p5/z, z27.h, z11.h
550: 9b 1b 8c 24	cmphi	p11.s, p6/z, z28.s, z12.s
554: a0 1f cd 24	cmphs	p0.d, p7/z, z29.d, z13.d
558: d2 83 0e 24	cmpgt	p2.b, p0/z, z30.b, z14.b
55c: f3 a5 54 24	cmpne	p3.h, p1/z, z15.h, z20.h
560: ef 3f 03 24	cmpeq	p15.b, p7/z, z31.b, z3.d
564: ae 5a 4d 24	cmpge	p14.h, p6/z, z21.h, z13.d
568: 7d 55 97 24	cmpgt	p13.s, p5/z, z11.s, z23.d
56c: 3c d0 1f 24	cmphi	p12.b, p4/z, z1.b, z31.d
570: 0b cc 5e 24	cmphs	p11.h, p3/z, z0.h, z30.d
574: 04 48 8a 24	cmpge	p4.s, p2/z, z0.s, z10.d
578: 33 c4 14 24	cmphi	p3.b, p1/z, z1.b, z20.d
57c: 42 c0 5e 24	cmphs	p2.h, p0/z, z2.h, z30.d
580: 11 5d 98 24	cmpgt	p1.s, p7/z, z8.s, z24.d
584: d0 21 1c 24	cmpne	p0.b, p0/z, z14.b, z28.d
588: ef 83 08 25	cmpeq	p15.b, p0/z, z31.b, #8
58c: ab 1e 41 25	cmpge	p11.h, p7/z, z21.h, #1
590: 5a 06 84 25	cmpgt	p10.s, p1/z, z18.s, #4
594: 78 39 cf 25	cmple	p8.d, p6/z, z11.d, #15
598: 07 29 10 25	cmplt	p7.b, p2/z, z8.b, #-16
59c: 10 94 52 25	cmpne	p0.h, p5/z, z0.h, #-14
5a0: 7f 1e 20 24	cmphi	p15.b, p7/z, z19.b, #0
5a4: 0b 04 69 24	cmphs	p11.h, p1/z, z0.h, #36
5a8: a8 36 b0 24	cmplo	p8.s, p5/z, z21.s, #64
5ac: 30 ed ff 24	cmpls	p0.d, p3/z, z9.d, #127
5b0: c0 4d 0f 25	and	p0.b, p3/z, p14.b, p15.b
5b4: 02 59 4d 25	ands	p2.b, p6/z, p8.b, p13.b
5b8: 94 65 0b 25	bic	p4.b, p9/z, p12.b, p11.b
5bc: 16 70 49 25	bics	p6.b, p12/z, p0.b, p9.b
5c0: 68 7f 07 25	eor	p8.b, p15/z, p11.b, p7.b
5c4: aa 42 43 25	eors	p10.b, p0/z, p5.b, p3.b
5c8: 4c 45 0a 25	mov	p12.b, p1/z, p10.b
5cc: 0e 50 40 25	movs	p14.b, p4/z, p0.b
5d0: 51 5e 8f 25	nand	p1.b, p7/z, p2.b, p15.b
5d4: d3 42 c9 25	nands	p3.b, p0/z, p6.b, p9.b
5d8: 45 77 87 25	nor	p5.b, p13/z, p10.b, p7.b
5dc: c7 4b c5 25	nors	p7.b, p2/z, p14.b, p5.b
5e0: e9 57 05 25	not	p9.b, p5/z, p15.b
5e4: 0b 62 48 25	nots	p11.b, p8/z, p0.b
5e8: 3d 4c 8b 25	orn	p13.b, p3/z, p1.b, p11.b
5ec: 3f 50 c8 25	orns	p15.b, p4/z, p1.b, p8.b
5f0: e0 41 8c 25	orr	p0.b, p0/z, p15.b, p12.b
5f4: 22 4d ce 25	orrs	p2.b, p3/z, p9.b, p14.b
5f8: b4 5b 0a 25	sel	p4.b, p6, p13.b, p10.b
5fc: e0 7d cf 25	movs	p0.b, p15.b
600: f0 63 00 25	mov	p0.b, p8/m, p15.b
604: e0 7d 8f 25	mov	p0.b, p15.b
608: 40 c5 0f 25	brkpa	p0.b, p1/z, p10.b, p15.b
60c: e3 d1 40 25	brkpas	p3.b, p4/z, p15.b, p0.b
610: 77 e1 0d 25	brkpb	p7.b, p8/z, p11.b, p13.b
614: 3b e4 43 25	brkpbs	p11.b, p9/z, p1.b, p3.b
618: e0 61 10 25	brka	p0.b, p8/z, p15.b
61c: c2 65 90 25	brkb	p2.b, p9/z, p14.b
620: f0 61 10 25	brka	p0.b, p8/m, p15.b
624: d2 65 90 25	brkb	p2.b, p9/m, p14.b
628: 60 79 50 25	brkas	p0.b, p14/z, p11.b
62c: af 45 d0 25	brkbs	p15.b, p1/z, p13.b
630: e0 61 18 25	brkn	p0.b, p8/z, p15.b, p0.b
634: a1 5d 58 25	brkns	p1.b, p7/z, p13.b, p1.b
638: e0 c1 58 25	pfirst	p0.b, p15, p0.b
63c: 00 e0 18 25	ptrue	p0.b, pow2
640: c7 e3 58 25	ptrue	p7.h, mul3
644: e8 e3 99 25	ptrues	p8.s
648: ef e3 d9 25	ptrues	p15.d
64c: e0 c5 19 25	pnext	p0.b, p15, p0.b
650: ea f1 18 25	rdffr	p10.b, p15/z
654: c7 f1 58 25	rdffrs	p7.b, p14/z
658: 08 f0 19 25	rdffr	p8.b
65c: 0d e4 18 25	pfalse	p13.b
660: c0 c9 50 25	ptest	p2, p14.b
664: fd 81 e0 25	cntp	x29, p0, p15.d
668: af 82 20 45	match	p15.b, p0/z, z21.b, z0.b
66c: 70 9d 7f 45	nmatch	p0.h, p7/z, z11.h, z31.h
670: 83 a0 8a 64	bfcvtnt	z3.h, p0/m, z4.s
674: 20 bc cb 64	fcvtlt	z0.d, p7/m, z1.s
678: 8e be 89 64	fcvtlt	z14.s, p7/m, z20.h
67c: 32 ad 88 64	fcvtnt	z18.h, p3/m, z9.s
680: ac ac ca 64	fcvtnt	z12.s, p3/m, z5.d
684: 61 a8 0a 64	fcvtxnt	z1.s, p2/m, z3.d
688: 70 8e 50 64	faddp	z16.h, p3/m, z16.h, z19.h
68c: 51 92 94 64	fmaxnmp	z17.s, p4/m, z17.s, z18.s
690: 32 96 d6 64	fmaxp	z18.d, p5/m, z18.d, z17.d
694: 13 9a 95 64	fminnmp	z19.s, p6/m, z19.s, z16.s
698: f4 9d 57 64	fminp	z20.h, p7/m, z20.h, z15.h
69c: d5 39 58 65	fadda	h21, p6, h21, z14.h
6a0: b6 35 98 65	fadda	s22, p5, s22, z13.s
6a4: 97 31 d8 65	fadda	d23, p4, d23, z12.d
6a8: 78 8d 48 65	fabd	z24.h, p3/m, z24.h, z11.h
6ac: 59 89 80 65	fadd	z25.s, p2/m, z25.s, z10.s
6b0: fc 80 8d 65	fdiv	z28.s, p0/m, z28.s, z7.s
6b4: dd 84 cc 65	fdivr	z29.d, p1/m, z29.d, z6.d
6b8: be 88 46 65	fmax	z30.h, p2/m, z30.h, z5.h
6bc: 9f 8c 84 65	fmaxnm	z31.s, p3/m, z31.s, z4.s
6c0: 60 90 c7 65	fmin	z0.d, p4/m, z0.d, z3.d
6c4: 41 94 45 65	fminnm	z1.h, p5/m, z1.h, z2.h
6c8: 22 98 82 65	fmul	z2.s, p6/m, z2.s, z1.s
6cc: 03 9c ca 65	fmulx	z3.d, p7/m, z3.d, z0.d
6d0: e4 9b 49 65	fscale	z4.h, p6/m, z4.h, z31.h
6d4: c5 97 81 65	fsub	z5.s, p5/m, z5.s, z30.s
6d8: a6 93 c3 65	fsubr	z6.d, p4/m, z6.d, z29.d
Decode Failure @ offset      6dc
6dc: 20 80 00 65
Decode Failure @ offset      6e0
6e0: 62 84 06 65
Decode Failure @ offset      6e4
6e4: a4 88 04 65
Decode Failure @ offset      6e8
6e8: e6 8c 07 65
Decode Failure @ offset      6ec
6ec: 28 91 05 65
Decode Failure @ offset      6f0
6f0: 6a 95 02 65
Decode Failure @ offset      6f4
6f4: ac 99 01 65
6f8: 23 a9 8a 65	bfcvt	z3.h, p2/m, z9.s
6fc: 27 bc cb 65	fcvt	z7.d, p7/m, z1.s
700: 9d ad ca 65	fcvt	z29.s, p3/m, z12.d
704: a0 b1 c8 65	fcvt	z0.h, p4/m, z13.d
708: c1 b5 c9 65	fcvt	z1.d, p5/m, z14.h
70c: e2 b9 88 65	fcvt	z2.h, p6/m, z15.s
710: 03 be 89 65	fcvt	z3.s, p7/m, z16.h
714: c2 a0 0a 65	fcvtx	z2.s, p0/m, z6.d
718: 69 a4 9c 65	fcvtzs	z9.s, p1/m, z3.s
71c: 05 a3 dc 65	fcvtzs	z5.d, p0/m, z24.s
720: cc ac d8 65	fcvtzs	z12.s, p3/m, z6.d
724: 22 a6 de 65	fcvtzs	z2.d, p1/m, z17.d
728: 43 aa 5a 65	fcvtzs	z3.h, p2/m, z18.h
72c: 64 ae 5c 65	fcvtzs	z4.s, p3/m, z19.h
730: 85 b2 5e 65	fcvtzs	z5.d, p4/m, z20.h
734: 43 a9 9d 65	fcvtzu	z3.s, p2/m, z10.s
738: 2a bc dd 65	fcvtzu	z10.d, p7/m, z1.s
73c: a4 ad d9 65	fcvtzu	z4.s, p3/m, z13.d
740: 96 b8 df 65	fcvtzu	z22.d, p6/m, z4.d
744: b7 bc 5b 65	fcvtzu	z23.h, p7/m, z5.h
748: d8 a0 5d 65	fcvtzu	z24.s, p0/m, z6.h
74c: f9 a4 5f 65	fcvtzu	z25.d, p1/m, z7.h
750: 13 a9 94 65	scvtf	z19.s, p2/m, z8.s
754: 61 b6 d0 65	scvtf	z1.d, p5/m, z19.s
758: c4 a1 d4 65	scvtf	z4.s, p0/m, z14.d
75c: 00 a0 d6 65	scvtf	z0.d, p0/m, z0.d
760: cc b5 52 65	scvtf	z12.h, p5/m, z14.h
764: 0e be 54 65	scvtf	z14.h, p7/m, z16.s
768: 50 a6 56 65	scvtf	z16.h, p1/m, z18.d
76c: 71 b9 95 65	ucvtf	z17.s, p6/m, z11.s
770: 83 ae d1 65	ucvtf	z3.d, p3/m, z20.s
774: e8 a4 d5 65	ucvtf	z8.s, p1/m, z7.d
778: 28 b1 d7 65	ucvtf	z8.d, p4/m, z9.d
77c: ed b9 53 65	ucvtf	z13.h, p6/m, z15.h
780: 2f a2 55 65	ucvtf	z15.h, p0/m, z17.s
784: 71 aa 57 65	ucvtf	z17.h, p2/m, z19.d
788: 50 c1 5f 65	facge	p0.h, p0/z, z10.h, z31.h
78c: 9f e6 95 65	facgt	p15.s, p1/z, z20.s, z21.s
790: 71 c9 c0 65	facge	p1.d, p2/z, z11.d, z0.d
794: 3e ec 5e 65	facgt	p14.h, p3/z, z1.h, z30.h
798: 82 73 88 65	fcmeq	p2.s, p4/z, z28.s, z8.s
79c: 0d 55 d2 65	fcmge	p13.d, p5/z, z8.d, z18.d
7a0: 53 5a 5c 65	fcmgt	p3.h, p6/z, z18.h, z28.h
7a4: cc 5f 81 65	fcmge	p12.s, p7/z, z30.s, z1.s
7a8: 14 40 cb 65	fcmgt	p4.d, p0/z, z0.d, z11.d
7ac: bb 66 4a 65	fcmne	p11.h, p1/z, z21.h, z10.h
7b0: e5 cb 94 65	fcmuo	p5.s, p2/z, z31.s, z20.s
7b4: 20 00 62 65	fmla	z0.h, p0/m, z1.h, z2.h
7b8: 83 28 a5 65	fmls	z3.s, p2/m, z4.s, z5.s
7bc: e6 50 e8 65	fnmla	z6.d, p4/m, z7.d, z8.d
7c0: 49 79 6b 65	fnmls	z9.h, p6/m, z10.h, z11.h
7c4: 00 20 1a 04	andv	b0, p0, z0.b
7c8: 21 24 59 04	eorv	h1, p1, z1.h
7cc: 42 28 98 04	orv	s2, p2, z2.s
7d0: 63 2c d8 04	orv	d3, p3, z3.d
7d4: 41 30 00 04	saddv	d1, p4, z2.b
7d8: 62 34 40 04	saddv	d2, p5, z3.h
7dc: 83 38 81 04	uaddv	d3, p6, z4.s
7e0: 8f 3c c8 04	smaxv	d15, p7, z4.d
7e4: d0 39 8a 04	sminv	s16, p6, z14.s
7e8: 11 37 49 04	umaxv	h17, p5, z24.h
7ec: f2 33 0b 04	uminv	b18, p4, z31.b
7f0: 1f a0 18 04	cls	z31.b, p0/m, z0.b
7f4: 3e a4 59 04	clz	z30.h, p1/m, z1.h
7f8: 5d a8 9b 04	cnot	z29.s, p2/m, z2.s
7fc: 7c ac da 04	cnt	z28.d, p3/m, z3.d
800: 9b b0 5c 04	fabs	z27.h, p4/m, z4.h
804: ba b4 9d 04	fneg	z26.s, p5/m, z5.s
808: d9 b8 1e 04	not	z25.b, p6/m, z6.b
80c: f8 bc 16 04	abs	z24.b, p7/m, z7.b
810: 17 a1 97 04	neg	z23.s, p0/m, z8.s
814: 36 a5 50 04	sxtb	z22.h, p1/m, z9.h
818: 36 a5 90 04	sxtb	z22.s, p1/m, z9.s
81c: 36 a5 d0 04	sxtb	z22.d, p1/m, z9.d
820: 55 a9 92 04	sxth	z21.s, p2/m, z10.s
824: 55 a9 d2 04	sxth	z21.d, p2/m, z10.d
828: 74 ad d4 04	sxtw	z20.d, p3/m, z11.d
82c: 93 b1 51 04	uxtb	z19.h, p4/m, z12.h
830: 93 b1 91 04	uxtb	z19.s, p4/m, z12.s
834: 93 b1 d1 04	uxtb	z19.d, p4/m, z12.d
838: b2 b5 93 04	uxth	z18.s, p5/m, z13.s
83c: b2 b5 d3 04	uxth	z18.d, p5/m, z13.d
840: d1 b9 d5 04	uxtw	z17.d, p6/m, z14.d
844: 40 40 13 04	mla	z0.b, p0/m, z2.b, z19.b
848: e2 67 5f 04	mls	z2.h, p1/m, z31.h, z31.h
84c: 84 c9 97 04	mad	z4.s, p2/m, z23.s, z12.s
850: c8 fe cc 04	msb	z8.d, p7/m, z12.d, z22.d
854: 00 00 20 04	add	z0.b, z0.b, z0.b
858: e3 13 6c 04	sqadd	z3.h, z31.h, z12.h
85c: 07 18 bf 04	sqsub	z7.s, z0.s, z31.s
860: f3 04 ed 04	sub	z19.d, z7.d, z13.d
864: 97 17 3d 04	uqadd	z23.b, z28.b, z29.b
868: ff 1f 7f 04	uqsub	z31.h, z31.h, z31.h
86c: 05 60 3f 04	mul	z5.b, z0.b, z31.b
870: e0 6b 65 04	smulh	z0.h, z31.h, z5.h
874: bf 6c e0 04	umulh	z31.d, z5.d, z0.d
878: 87 73 20 04	sqdmulh	z7.b, z28.b, z0.b
87c: 77 74 7f 04	sqrdmulh	z23.h, z3.h, z31.h
880: 11 b2 ef 04	ftssel	z17.d, z16.d, z15.d
884: 20 70 22 05	trn1	z0.b, z1.b, z2.b
888: 83 70 65 05	trn1	z3.h, z4.h, z5.h
88c: e6 74 a8 05	trn2	z6.s, z7.s, z8.s
890: 49 75 eb 05	trn2	z9.d, z10.d, z11.d
894: ac 69 2e 05	uzp1	z12.b, z13.b, z14.b
898: 0f 6a 71 05	uzp1	z15.h, z16.h, z17.h
89c: 72 6e b4 05	uzp2	z18.s, z19.s, z20.s
8a0: d5 6e f7 05	uzp2	z21.d, z22.d, z23.d
8a4: 38 63 3a 05	zip1	z24.b, z25.b, z26.b
8a8: 9b 63 7d 05	zip1	z27.h, z28.h, z29.h
8ac: fe 67 a0 05	zip2	z30.s, z31.s, z0.s
8b0: 41 64 e3 05	zip2	z1.d, z2.d, z3.d
Decode Failure @ offset      8b4
8b4: 20 34 22 05
Decode Failure @ offset      8b8
8b8: 83 34 65 05
Decode Failure @ offset      8bc
8bc: e6 34 a8 05
Decode Failure @ offset      8c0
8c0: 49 35 eb 05
Decode Failure @ offset      8c4
8c4: 20 c0 02 44
Decode Failure @ offset      8c8
8c8: 83 c0 45 44
Decode Failure @ offset      8cc
8cc: e6 c4 88 44
Decode Failure @ offset      8d0
8d0: 49 c5 cb 44
Decode Failure @ offset      8d4
8d4: 20 24 62 64
Decode Failure @ offset      8d8
8d8: 83 24 a5 64
Decode Failure @ offset      8dc
8dc: e6 24 e8 64
8e0: 20 90 02 45	eorbt	z0.b, z1.b, z2.b
8e4: 83 90 45 45	eorbt	z3.h, z4.h, z5.h
8e8: e6 94 88 45	eortb	z6.s, z7.s, z8.s
8ec: 49 95 cb 45	eortb	z9.d, z10.d, z11.d
8f0: 20 b4 02 45	bdep	z0.b, z1.b, z2.b
8f4: 83 b0 45 45	bext	z3.h, z4.h, z5.h
8f8: e6 b8 88 45	bgrp	z6.s, z7.s, z8.s
8fc: 49 b9 cb 45	bgrp	z9.d, z10.d, z11.d
900: 20 00 42 65	fadd	z0.h, z1.h, z2.h
904: 83 08 85 65	fmul	z3.s, z4.s, z5.s
908: e6 18 c8 65	frecps	z6.d, z7.d, z8.d
90c: 49 1d 4b 65	frsqrts	z9.h, z10.h, z11.h
910: ac 05 8e 65	fsub	z12.s, z13.s, z14.s
914: 0f 0e d1 65	ftsmul	z15.d, z16.d, z17.d
918: f8 4f 29 04	index	z24.b, wzr, w9
91c: ec 4d e0 04	index	z12.d, x15, x0
920: 20 64 22 04	pmul	z0.b, z1.b, z2.b
924: e9 83 22 04	asr	z9.b, z31.b, z2.d
928: 13 8c 6c 04	lsl	z19.h, z0.h, z12.d
92c: 5d 85 b6 04	lsr	z29.s, z10.s, z22.d
930: 44 a0 e0 04	adr	z4.d, [z2.d, z0.d]
934: 3d a4 aa 04	adr	z29.s, [z1.s, z10.s, lsl #1]
938: e9 a0 29 04	adr	z9.d, [z7.d, z9.d, sxtw]
93c: 6c a8 25 04	adr	z12.d, [z3.d, z5.d, sxtw #2]
940: 49 a1 6e 04	adr	z9.d, [z10.d, z14.d, uxtw]
944: e3 ad 6b 04	adr	z3.d, [z15.d, z11.d, uxtw #3]
Decode Failure @ offset      948
948: 20 18 a2 05
Decode Failure @ offset      94c
94c: 83 1c a5 05
Decode Failure @ offset      950
950: e6 08 a8 05
Decode Failure @ offset      954
954: 49 0d ab 05
Decode Failure @ offset      958
958: ac 01 ae 05
Decode Failure @ offset      95c
95c: 0f 06 b1 05
960: 80 0f 80 05	and	z0.b, z0.b, #0xaa
964: 81 0f 80 05	and	z1.b, z1.b, #0xaa
968: e2 44 80 05	and	z2.h, z2.h, #0xff00
96c: e3 44 80 05	and	z3.h, z3.h, #0xff00
970: 64 91 40 05	eor	z4.s, z4.s, #0x3ffc000
974: 65 91 40 05	eor	z5.s, z5.s, #0x3ffc000
978: 66 e2 02 05	orr	z6.d, z6.d, #0xfffff000000000
97c: 67 e2 02 05	orr	z7.d, z7.d, #0xfffff000000000
980: 40 26 c0 05	dupm	z0.b, #0x70
984: 61 75 c0 05	mov	z1.h, #16380
988: 42 a0 c0 05	dupm	z2.s, #0x7000
98c: e3 85 c3 05	mov	z3.d, #0xffffffffffff0000
990: a4 06 c0 05	dupm	z4.b, #0x3f
994: 45 44 c0 05	dupm	z5.h, #0x700
998: 66 e2 c0 05	mov	z6.s, #0xfffff0
99c: 67 a5 c2 05	mov	z7.d, #0xfffff00000ffffff
9a0: 0f 00 15 05	mov	z15.b, p5/z, #0
9a4: 65 03 1f 05	mov	z5.b, p15/z, #27
9a8: 1f 10 10 05	mov	z31.b, p0/z, #-128
9ac: e0 0f 15 05	mov	z0.b, p5/z, #127
9b0: 0f 00 55 05	mov	z15.h, p5/z, #0
9b4: 57 01 9c 05	mov	z23.s, p12/z, #10
9b8: 04 10 d0 05	mov	z4.d, p0/z, #-128
9bc: f3 0f 5f 05	mov	z19.h, p15/z, #127
9c0: 21 20 90 05	mov	z1.s, p0/z, #256
9c4: 81 21 d0 05	mov	z1.d, p0/z, #3072
9c8: 81 3e 50 05	mov	z1.h, p0/z, #-3072
9cc: 01 30 90 05	mov	z1.s, p0/z, #-32768
9d0: e0 2f d0 05	mov	z0.d, p0/z, #32512
9d4: a1 40 1c 05	mov	z1.b, p12/m, #5
9d8: db 62 5d 05	mov	z27.h, p13/m, #5632
9dc: 5b 7d 5d 05	mov	z27.h, p13/m, #-5632
9e0: 00 40 51 05	mov	z0.h, p1/m, #0
9e4: 02 40 93 05	mov	z2.s, p3/m, #0
9e8: 04 40 d5 05	mov	z4.d, p5/m, #0
9ec: 84 21 ff 05	mov	z4.b, z12.b[63]
9f0: 28 21 fe 05	mov	z8.h, z9.h[31]
9f4: 8b 23 fc 05	mov	z11.s, z28.s[15]
9f8: 95 21 f8 05	mov	z21.d, z12.d[7]
9fc: ee 20 f0 05	mov	z14.q, z7.q[3]
a00: 0d 21 21 05	mov	z13.b, b8
a04: 02 20 22 05	mov	z2.h, h0
a08: ef 23 24 05	mov	z15.s, s31
a0c: 77 23 28 05	mov	z23.d, d27
a10: 64 20 30 05	mov	z4.q, q3
a14: 01 22 ff 05	mov	z1.b, z16.b[63]
a18: 51 22 fe 05	mov	z17.h, z18.h[31]
a1c: 69 21 fc 05	mov	z9.s, z11.s[15]
a20: 62 20 f8 05	mov	z2.d, z3.d[7]
a24: 03 21 f0 05	mov	z3.q, z8.q[3]
a28: 2d 21 21 05	mov	z13.b, b9
a2c: cc 20 22 05	mov	z12.h, h6
a30: e2 20 24 05	mov	z2.s, s7
a34: 00 20 28 05	mov	z0.d, d0
a38: 8a 22 30 05	mov	z10.q, q20
a3c: 20 30 22 05	tbl	z0.b, { z1.b }, z2.b
a40: 83 30 65 05	tbl	z3.h, { z4.h }, z5.h
a44: e6 2c a8 05	tbx	z6.s, z7.s, z8.s
a48: 49 2d eb 05	tbx	z9.d, z10.d, z11.d
a4c: 20 28 22 05	tbl	z0.b, { z1.b, z2.b }, z2.b
a50: 83 28 65 05	tbl	z3.h, { z4.h, z5.h }, z5.h
a54: e6 28 a8 05	tbl	z6.s, { z7.s, z8.s }, z8.s
a58: 49 29 eb 05	tbl	z9.d, { z10.d, z11.d }, z11.d
a5c: 20 00 82 44	sdot	z0.s, z1.b, z2.b
a60: 83 00 c5 44	sdot	z3.d, z4.h, z5.h
a64: e6 04 88 44	udot	z6.s, z7.b, z8.b
a68: 49 05 cb 44	udot	z9.d, z10.h, z11.h
a6c: 20 40 42 44	smlalb	z0.h, z1.b, z2.b
a70: 83 44 85 44	smlalt	z3.s, z4.h, z5.h
a74: e6 50 c8 44	smlslb	z6.d, z7.s, z8.s
a78: 49 55 4b 44	smlslt	z9.h, z10.b, z11.b
a7c: ac 49 8e 44	umlalb	z12.s, z13.h, z14.h
a80: 0f 4e d1 44	umlalt	z15.d, z16.s, z17.s
a84: 72 5a 54 44	umlslb	z18.h, z19.b, z20.b
a88: d5 5e 97 44	umlslt	z21.s, z22.h, z23.h
a8c: 20 08 42 44	sqdmlalbt	z0.h, z1.b, z2.b
a90: 83 0c 85 44	sqdmlslbt	z3.s, z4.h, z5.h
a94: e6 0c c8 44	sqdmlslbt	z6.d, z7.s, z8.s
a98: 20 60 42 44	sqdmlalb	z0.h, z1.b, z2.b
a9c: 83 64 85 44	sqdmlalt	z3.s, z4.h, z5.h
aa0: e6 68 c8 44	sqdmlslb	z6.d, z7.s, z8.s
aa4: 49 6d 4b 44	sqdmlslt	z9.h, z10.b, z11.b
aa8: 20 c0 42 45	sabalb	z0.h, z1.b, z2.b
aac: 83 c4 85 45	sabalt	z3.s, z4.h, z5.h
ab0: e6 c8 c8 45	uabalb	z6.d, z7.s, z8.s
ab4: 49 cd 4b 45	uabalt	z9.h, z10.b, z11.b
ab8: 20 70 02 44	sqrdmlah	z0.b, z1.b, z2.b
abc: 83 70 45 44	sqrdmlah	z3.h, z4.h, z5.h
ac0: e6 74 88 44	sqrdmlsh	z6.s, z7.s, z8.s
ac4: 49 75 cb 44	sqrdmlsh	z9.d, z10.d, z11.d
Decode Failure @ offset      ac8
ac8: 20 f8 02 44
Decode Failure @ offset      acc
acc: 83 e8 45 44
Decode Failure @ offset      ad0
ad0: e6 ec 88 44
Decode Failure @ offset      ad4
ad4: 49 e1 cb 44
Decode Failure @ offset      ad8
ad8: ac e5 0e 44
adc: 20 30 42 45	sabdlb	z0.h, z1.b, z2.b
ae0: 83 34 85 45	sabdlt	z3.s, z4.h, z5.h
ae4: e6 00 c8 45	saddlb	z6.d, z7.s, z8.s
ae8: 49 05 4b 45	saddlt	z9.h, z10.b, z11.b
aec: ac 11 8e 45	ssublb	z12.s, z13.h, z14.h
af0: 0f 16 d1 45	ssublt	z15.d, z16.s, z17.s
af4: 72 3a 54 45	uabdlb	z18.h, z19.b, z20.b
af8: d5 3e 98 45	uabdlt	z21.s, z22.h, z24.h
afc: 38 0b da 45	uaddlb	z24.d, z25.s, z26.s
b00: 9b 0f 5d 45	uaddlt	z27.h, z28.b, z29.b
b04: fe 1b 80 45	usublb	z30.s, z31.h, z0.h
b08: 41 1c c3 45	usublt	z1.d, z2.s, z3.s
b0c: 20 68 42 45	pmullb	z0.h, z1.b, z2.b
b10: 83 6c c5 45	pmullt	z3.d, z4.s, z5.s
b14: e6 70 48 45	smullb	z6.h, z7.b, z8.b
b18: 49 75 cb 45	smullt	z9.d, z10.s, z11.s
b1c: ac 61 4e 45	sqdmullb	z12.h, z13.b, z14.b
b20: 0f 66 d1 45	sqdmullt	z15.d, z16.s, z17.s
b24: 72 7a 54 45	umullb	z18.h, z19.b, z20.b
b28: d5 7e d7 45	umullt	z21.d, z22.s, z23.s
b2c: 20 80 42 45	saddlbt	z0.h, z1.b, z2.b
b30: 83 88 85 45	ssublbt	z3.s, z4.h, z5.h
b34: e6 8c c8 45	ssubltb	z6.d, z7.s, z8.s
b38: 20 40 42 45	saddwb	z0.h, z1.h, z2.b
b3c: 83 44 85 45	saddwt	z3.s, z4.s, z5.h
b40: e6 50 c8 45	ssubwb	z6.d, z7.d, z8.s
b44: 49 55 4b 45	ssubwt	z9.h, z10.h, z11.b
b48: ac 49 8e 45	uaddwb	z12.s, z13.s, z14.h
b4c: 0f 4e d1 45	uaddwt	z15.d, z16.d, z17.s
b50: 72 5a 54 45	usubwb	z18.h, z19.h, z20.b
b54: d5 5e 97 45	usubwt	z21.s, z22.s, z23.h
Decode Failure @ offset      b58
b58: 20 68 02 45
Decode Failure @ offset      b5c
b5c: 83 6c 05 45
b60: 20 98 02 45	smmla	z0.s, z1.b, z2.b
b64: 83 98 c5 45	ummla	z3.s, z4.b, z5.b
b68: e6 98 88 45	usmmla	z6.s, z7.b, z8.b
b6c: 20 f8 02 45	saba	z0.b, z1.b, z2.b
b70: 83 f8 45 45	saba	z3.h, z4.h, z5.h
b74: e6 fc 88 45	uaba	z6.s, z7.s, z8.s
b78: 49 fd cb 45	uaba	z9.d, z10.d, z11.d
b7c: 20 a0 22 45	histseg	z0.b, z1.b, z2.b
Decode Failure @ offset      b80
b80: 20 24 22 64
Too many failures

Using jit(D:\git\runtime2\artifacts\obj\coreclr\windows.x64.Checked\ide\jit\Debug\clrjit_universal_arm64_x64.dll) with input (E:\PRs\bugfix#2\all_tests.mch)
 indexCount=1 (1)
Jit startup took 26.210100ms
Loaded 1  Jitted 1  FailedCompile 0 Excluded 0 Missing 0
Total time: 195.635700ms
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment