VMOVDQU AA2, (0*32)(oup); VMOVDQU BB2, (1*32)(oup); VMOVDQU CC2, (2*32)(oup); VMOVDQU DD2, (3*32)(oup)
	LEAQ    (4*32)(inp), inp
	LEAQ    (4*32)(oup), oup
	SUBQ    $4*32, inl

	JMP openAVX2TailLoop

// ----------------------------------------------------------------------------
// Special optimization for the last 384 bytes of ciphertext

	pc=5077843 func=35 file=36 line=32
Stack id=56 nframes=5
	pc=4753924 func=81 file=25 line=432
	pc=4744422 func=82 file=83 line=106
	pc=4823012 func=84 file=85 line=218
	pc=4824373 func=86 file=87 line=21
	pc=5079543 func=59 file=36 line=82
Stack id=60 nframes=5
	pc=4753924 func=81 file=25 line=432
	pc=4744422 func=82 file=83 line=106
	pc=4813961 func=145 file=29 line=129

	/* x^234560 mod p(x), x^234496 mod p(x) */
DATA ·IEEEConst+416(SB)/8,$0x00000001a0531540
DATA ·IEEEConst+424(SB)/8,$0x00000000c53dfb04

	/* x^233536 mod p(x), x^233472 mod p(x) */
DATA ·IEEEConst+432(SB)/8,$0x0000000132cd7ebc
DATA ·IEEEConst+440(SB)/8,$0x00000000e10c9ad6

	/* x^232512 mod p(x), x^232448 mod p(x) */
DATA ·IEEEConst+448(SB)/8,$0x0000000073ab7f36
DATA ·IEEEConst+456(SB)/8,$0x0000000025aa994a

      "integrity": "sha512-+5gy6OQfk+xx3q0d6jGZZC3f3KzAkXc/IanVxd1is/VIIziRqqt3ongQz0FiTUXqTk0c7aDB3OaFuKnuSoJicQ==",
      "dev": true,
      "dependencies": {
        "ajv": "^6.12.4",
        "debug": "^4.3.2",
        "espree": "^9.5.2",
        "globals": "^13.19.0",
        "ignore": "^5.2.0",
        "import-fresh": "^3.2.1",
        "js-yaml": "^4.1.0",
        "minimatch": "^3.1.2",

	SYS_OPENDIR                         = 0x1AD // 429
	SYS_PATHCONF                        = 0x1AE // 430
	SYS_PAUSE                           = 0x1AF // 431
	SYS_PIPE                            = 0x1B0 // 432
	SYS_W_GETPSENT                      = 0x1B1 // 433
	SYS_READ                            = 0x1B2 // 434
	SYS_READDIR                         = 0x1B3 // 435
	SYS_READLINK                        = 0x1B4 // 436

//
//   eval z = z[0] + (z[1] << 64) + (z[2] << 128) + (z[3] << 192) + (z[4] << 256) + (z[5] << 0x140) + (z[6] << 0x180) + (z[7] << 0x1c0) + (z[8] << 2^9)
//

  %3 = "tf.Const"() {value = dense<4096> : tensor<i32>} : () -> tensor<i32>
  %4 = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
  %5 = "tf.Range"(%4, %3, %2) : (tensor<i32>, tensor<i32>, tensor<i32>) -> tensor<4096xi32>
  %6 = "tf.Equal"(%1, %5) : (tensor<*xi32>, tensor<4096xi32>) -> tensor<*xi1>
  %7 = "tf.Cast"(%6) : (tensor<*xi1>) -> tensor<*xf32>

		case pack(0, ARNG_4H, ARNG_4S):
			immh, width = 2, 16
		case pack(1, ARNG_8H, ARNG_4S):
			immh, width = 2, 16
		case pack(0, ARNG_2S, ARNG_2D):
			immh, width = 4, 32
		case pack(1, ARNG_4S, ARNG_2D):
			immh, width = 4, 32
		default:
			c.ctxt.Diag("operand mismatch: %v\n", p)
		}
		if !(0 <= shift && shift <= int(width-1)) {
			c.ctxt.Diag("shift amount out of range: %v\n", p)
		}

func stack424()  { var buf [424]byte; use(buf[:]); C.callGoStackCheck() }
func stack428()  { var buf [428]byte; use(buf[:]); C.callGoStackCheck() }
func stack432()  { var buf [432]byte; use(buf[:]); C.callGoStackCheck() }
func stack436()  { var buf [436]byte; use(buf[:]); C.callGoStackCheck() }
func stack440()  { var buf [440]byte; use(buf[:]); C.callGoStackCheck() }

	428:  {region: 0x99, script: 0x5b, flags: 0x0},
	429:  {region: 0x166, script: 0x5b, flags: 0x0},
	430:  {region: 0xdc, script: 0x22, flags: 0x0},
	431:  {region: 0x166, script: 0x5b, flags: 0x0},
	432:  {region: 0x166, script: 0x5, flags: 0x0},
	433:  {region: 0x49, script: 0x5b, flags: 0x0},
	434:  {region: 0x166, script: 0x5, flags: 0x0},
	435:  {region: 0x166, script: 0x5b, flags: 0x0},