VPCLMULQDQ $127, X22, X28, X15                     // 62331d0044fe7f or 62339d0044fe7f
	VPCLMULQDQ $127, X7, X28, X15                      // 62731d0044ff7f or 62739d0044ff7f
	VPCLMULQDQ $127, X19, X28, X15                     // 62331d0044fb7f or 62339d0044fb7f
	VPCLMULQDQ $127, -17(BP)(SI*8), X28, X15           // 62731d0044bcf5efffffff7f or 62739d0044bcf5efffffff7f

	VPCONFLICTD (BP), K6, X17                          // 62e27d0ec44d00
	VPCONFLICTD X6, K6, X28                            // 62627d0ec4e6
	VPCONFLICTD X1, K6, X28                            // 62627d0ec4e1
	VPCONFLICTD X8, K6, X28                            // 62427d0ec4e0
	VPCONFLICTD 15(R8), K6, X28                        // 62427d0ec4a00f000000
	VPCONFLICTD (BP), K6, X28                          // 62627d0ec46500

	VRSQRT28SD X20, X20, K2, X28                       // 6222dd02cde4
	VRSQRT28SD X16, X20, K2, X28                       // 6222dd02cde0
	VRSQRT28SD X12, X20, K2, X28                       // 6242dd02cde4
	VRSQRT28SD X20, X24, K2, X28                       // 6222bd02cde4
	VRSQRT28SD X16, X24, K2, X28                       // 6222bd02cde0
	VRSQRT28SD X12, X24, K2, X28                       // 6242bd02cde4

package main

func main() {
	var x00, x01, x02, x03, x04, x05, x06, x07, x08, x09 int
	var x10, x11, x12, x13, x14, x15, x16, x17, x18, x19 int
	var x20, x21, x22, x23, x24, x25, x26, x27, x28, x29 int
	var x30, x31, x32 int

	_ = x00
	_ = x01
	_ = x02
	_ = x03
	_ = x04
	_ = x05
	_ = x06
	_ = x07
	_ = x08
	_ = x09

	_ = x10
	_ = x11
	_ = x12
	_ = x13
	_ = x14

	VPDPBUSD -15(BX), X16, K2, X6                      // 62f27d0250b3f1ffffff
	VPDPBUSD X15, X28, K2, X6                          // 62d21d0250f7
	VPDPBUSD X11, X28, K2, X6                          // 62d21d0250f3
	VPDPBUSD X1, X28, K2, X6                           // 62f21d0250f1
	VPDPBUSD -15(R14)(R15*1), X28, K2, X6              // 62921d0250b43ef1ffffff
	VPDPBUSD -15(BX), X28, K2, X6                      // 62f21d0250b3f1ffffff

	VPINSRD $82, -7(CX)(DX*1), X7, X28                 // 6263450822a411f9ffffff52
	VPINSRD $82, -15(R14)(R15*4), X7, X28              // 6203450822a4bef1ffffff52
	VPINSRD $82, R9, X19, X28                          // 6243650022e152
	VPINSRD $82, CX, X19, X28                          // 6263650022e152
	VPINSRD $82, -7(CX)(DX*1), X19, X28                // 6263650022a411f9ffffff52
	VPINSRD $82, -15(R14)(R15*4), X19, X28             // 6203650022a4bef1ffffff52

	VGF2P8AFFINEQB $82, X22, X28, K1, X15              // 62339d01cefe52
	VGF2P8AFFINEQB $82, X7, X28, K1, X15               // 62739d01ceff52
	VGF2P8AFFINEQB $82, X19, X28, K1, X15              // 62339d01cefb52
	VGF2P8AFFINEQB $82, 17(SP)(BP*2), X28, K1, X15     // 62739d01cebc6c1100000052
	VGF2P8AFFINEQB $82, -7(DI)(R8*4), X28, K1, X15     // 62339d01cebc87f9ffffff52

	x26, x27 = bits.Add64(x20, uint64(0x0), uint64(p224Uint1(x25)))
	var x28 uint64
	_, x28 = bits.Mul64(x22, 0xffffffffffffffff)
	var x30 uint64
	var x31 uint64
	x31, x30 = bits.Mul64(x28, 0xffffffff)
	var x32 uint64
	var x33 uint64
	x33, x32 = bits.Mul64(x28, 0xffffffffffffffff)
	var x34 uint64
	var x35 uint64
	x35, x34 = bits.Mul64(x28, 0xffffffff00000000)
	var x36 uint64
	var x37 uint64

	VADDPD X12, X29, X28        // 6241950058e4
	VADDPD X30, X29, X28        // 6201950058e6
	VADDPD (AX), X29, X0        // 62f195005800
	VADDPD (AX), X1, X28        // 6261f5085820
	VADDPD (AX), X29, X28       // 626195005820
	VADDPD (AX), X29, X10       // 627195005810
	VADDPD (AX), X10, X28       // 6261ad085820
	VADDPD (CX)(AX*1), X29, X0  // 62f19500580401
	VADDPD (CX)(AX*1), X1, X28  // 6261f508582401

	MOVBU	1(X12), X16
	MOVBU	2(X12), X18
	MOVBU	3(X12), X20
	MOVBU	4(X12), X22
	MOVBU	5(X12), X24
	MOVBU	6(X12), X28
	MOVBU	7(X12), X30
	BNE	X8, X9, cmp1a
	BNE	X15, X16, cmp1b
	BNE	X17, X18, cmp1c
	BNE	X19, X20, cmp1d
	BNE	X21, X22, cmp1e
	BNE	X23, X24, cmp1f
	BNE	X25, X28, cmp1g
	BNE	X29, X30, cmp1h
	ADD	$8, X10
	ADD	$8, X12
	SUB	$8, X5
	BGE	X5, X6, compare8_unaligned
	BEQZ	X5, cmp_len