VGF2P8AFFINEINVQB $64, X0, X16, K3, X26            // 6263fd03cfd040
	VGF2P8AFFINEINVQB $64, -17(BP), X16, K3, X26       // 6263fd03cf95efffffff40
	VGF2P8AFFINEINVQB $64, -15(R14)(R15*8), X16, K3, X26 // 6203fd03cf94fef1ffffff40
	VGF2P8AFFINEINVQB $64, X8, X7, K3, X26             // 6243c50bcfd040
	VGF2P8AFFINEINVQB $64, X1, X7, K3, X26             // 6263c50bcfd140

	VAESENCLAST X11, X16, X26                          // 62427d00ddd3 or 6242fd00ddd3
	VAESENCLAST X31, X16, X26                          // 62027d00ddd7 or 6202fd00ddd7
	VAESENCLAST (AX), X16, X26                         // 62627d00dd10 or 6262fd00dd10
	VAESENCLAST 7(SI), X16, X26                        // 62627d00dd9607000000 or 6262fd00dd9607000000

	VRCP28SD X2, X2, K2, X18                           // 62e2ed0acbd2 or 62e2ed2acbd2 or 62e2ed4acbd2
	VRCP28SD X27, X2, K2, X18                          // 6282ed0acbd3 or 6282ed2acbd3 or 6282ed4acbd3
	VRCP28SD X26, X2, K2, X18                          // 6282ed0acbd2 or 6282ed2acbd2 or 6282ed4acbd2
	VRCP28SD 17(SP)(BP*8), X2, K2, X18                 // 62e2ed0acb94ec11000000 or 62e2ed2acb94ec11000000 or 62e2ed4acb94ec11000000

	VPLZCNTD X26, K3, X17                              // 62827d0b44ca
	VPLZCNTD X23, K3, X17                              // 62a27d0b44cf
	VPLZCNTD 15(DX)(BX*1), K3, X17                     // 62e27d0b448c1a0f000000
	VPLZCNTD -7(CX)(DX*2), K3, X17                     // 62e27d0b448c51f9ffffff
	VPLZCNTD X3, K3, X15                               // 62727d0b44fb
	VPLZCNTD X26, K3, X15                              // 62127d0b44fa

	var x24 uint64
	var x25 uint64
	x25, x24 = bits.Mul64(x11, 0xffffffffffffffff)
	var x26 uint64
	var x27 uint64
	x26, x27 = bits.Add64(x25, x22, uint64(0x0))
	x28 := (uint64(p256Uint1(x27)) + x23)
	var x30 uint64
	_, x30 = bits.Add64(x11, x24, uint64(0x0))
	var x31 uint64
	var x32 uint64
	x31, x32 = bits.Add64(x13, x26, uint64(p256Uint1(x30)))
	var x33 uint64
	var x34 uint64

	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

	x22, x21 = bits.Mul64(x19, 0xffffffff)
	var x23 uint64
	var x24 uint64
	x24, x23 = bits.Mul64(x19, 0xffffffffffffffff)
	var x25 uint64
	var x26 uint64
	x26, x25 = bits.Mul64(x19, 0xffffffff00000000)
	var x27 uint64
	var x28 uint64
	x27, x28 = bits.Add64(x26, x23, uint64(0x0))
	var x29 uint64
	var x30 uint64
	x29, x30 = bits.Add64(x24, x21, uint64(p224Uint1(x28)))
	var x32 uint64

TEXT asmtest_avx512_vbmi(SB), NOSPLIT, $0
	VPERMB X26, X20, K1, X23                           // 62825d018dfa
	VPERMB X19, X20, K1, X23                           // 62a25d018dfb
	VPERMB X0, X20, K1, X23                            // 62e25d018df8
	VPERMB 7(SI)(DI*4), X20, K1, X23                   // 62e25d018dbcbe07000000
	VPERMB -7(DI)(R8*2), X20, K1, X23                  // 62a25d018dbc47f9ffffff
	VPERMB X26, X2, K1, X23                            // 62826d098dfa

	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
	MAX	X26, X28				// 336eae0b
	MAXU	X28, X29, X30				// 33ffce0b
	MAXU	X28, X29				// b3fece0b
	MIN	X29, X30, X5				// b342df0b
	MIN	X29, X30				// 334fdf0b
	MINU	X30, X5, X6				// 33d3e20b

	VPDPWSSD X26, X2, K3, X18                          // 62826d0b52d2
	VPDPWSSD (SI), X2, K3, X18                         // 62e26d0b5216
	VPDPWSSD 7(SI)(DI*2), X2, K3, X18                  // 62e26d0b52947e07000000
	VPDPWSSD X2, X24, K3, X18                          // 62e23d0352d2
	VPDPWSSD X27, X24, K3, X18                         // 62823d0352d3
	VPDPWSSD X26, X24, K3, X18                         // 62823d0352d2