VPSHLDVW X1, X15, K7, X24                          // 6262850f70c1
	VPSHLDVW -17(BP)(SI*8), X15, K7, X24               // 6262850f7084f5efffffff
	VPSHLDVW (R15), X15, K7, X24                       // 6242850f7007
	VPSHLDVW X18, X26, K7, X24                         // 6222ad0770c2
	VPSHLDVW X21, X26, K7, X24                         // 6222ad0770c5

TEXT asmtest_aes_avx512f(SB), NOSPLIT, $0
	VAESDEC X24, X7, X11                               // 62124508ded8 or 6212c508ded8
	VAESDEC X20, X7, X11                               // 62324508dedc or 6232c508dedc
	VAESDEC X24, X0, X11                               // 62127d08ded8 or 6212fd08ded8
	VAESDEC X20, X0, X11                               // 62327d08dedc or 6232fd08dedc
	VAESDEC X24, X7, X31                               // 62024508def8 or 6202c508def8

	VPMAXUB 7(AX)(CX*8), Z13, K3, Z21                  // 62e1154bdeacc807000000 or 62e1954bdeacc807000000
	VPMAXUW X9, X24, K7, X28                           // 62423d073ee1 or 6242bd073ee1
	VPMAXUW -17(BP)(SI*8), X24, K7, X28                // 62623d073ea4f5efffffff or 6262bd073ea4f5efffffff
	VPMAXUW (R15), X24, K7, X28                        // 62423d073e27 or 6242bd073e27

    X7,
    X8,
    X9,
    X10,
    X11,
    X12,
    X13,
    X14,
    X15,
    X16,
    X17,
    X18,
    X19,
    X20,
    X21,
    X22,
    X23,
    X24,
    X25,
    X26,
    X27,
    X28,
    X29,
    X30,
    X31,
    X32,
    X33,
    X34,
    X35,
    X36,
    X37,
    X38,
    X39,
    X40,
    X41,

	ANDN	X19, X20, X21				// b37a3a41 or 93caf9ffb37a5a01
	ANDN	X19, X20				// 337a3a41 or 93cff9ff337afa01
	CLZ	X20, X21				// 931a0a60
	CLZW	X21, X22				// 1b9b0a60
	CPOP	X22, X23				// 931b2b60
	CPOPW	X23, X24				// 1b9c2b60
	CTZ	X24, X25				// 931c1c60
	CTZW	X25, X26				// 1b9d1c60
	MAX	X26, X28, X29				// b36eae0b or b32fae01b30ff041b34eae01b3fedf01b34ede01
	MAX	X26, X28				// 336eae0b or b32fcd01b30ff041334ecd0133fecf01334ecd01

	// EVEX gather (also tests Z as VSIB index).
	VPGATHERDD 360(AX)(X2*4), K1, X1    // 62f27d09904c905a
	VPGATHERDD 640(BP)(X15*8), K3, X14  // 62327d0b90b4fd80020000
	VPGATHERDD 960(R10)(X25*2), K7, X24 // 62027d0790844ac0030000
	VPGATHERDD 1280(R10)(X1*4), K4, X0  // 62d27d0c90848a00050000
	VPGATHERDD 360(AX)(Y2*4), K1, Y1    // 62f27d29904c905a
	VPGATHERDD 640(BP)(Y15*8), K3, Y14  // 62327d2b90b4fd80020000

x6) var x20 uint64 _, x20 = bits.Mul64(x11, 0xd2b51da312547e1b) var x22 uint64 var x23 uint64 x23, x22 = bits.Mul64(x20, 0x1000000000000000) var x24 uint64 var x25 uint64 x25, x24 = bits.Mul64(x20, 0x14def9dea2f79cd6) var x26 uint64 var x27 uint64 x27, x26 = bits.Mul64(x20, 0x5812631a5cf5d3ed) var x28 uint64 var x29 uint64 x28, x29 = bits.Add64(x27, x24, uint64(0x0)) x30 := (uint64(fiatScalarUint1(x29)) + x25) var x32 uint64 _, x32 = bits.Add64(x11, x26, uint64(0x0)) var x33 uint64 var x34 uint64...