t = r ^ uint32(k0>>32)
	l ^= feistelBox[7][t&0x3f] ^
		feistelBox[5][(t>>8)&0x3f] ^
		feistelBox[3][(t>>16)&0x3f] ^
		feistelBox[1][(t>>24)&0x3f]

	t = ((r << 28) | (r >> 4)) ^ uint32(k0)
	l ^= feistelBox[6][(t)&0x3f] ^
		feistelBox[4][(t>>8)&0x3f] ^
		feistelBox[2][(t>>16)&0x3f] ^
		feistelBox[0][(t>>24)&0x3f]

	t = l ^ uint32(k1>>32)
	r ^= feistelBox[7][t&0x3f] ^
		feistelBox[5][(t>>8)&0x3f] ^

        final int i = ((inData[inIndex] & 0xff) << 16) + ((inData[inIndex + 1] & 0xff) << 8) + (inData[inIndex + 2] & 0xff);
        outData[outIndex] = ENCODE_TABLE[i >> 18];
        outData[outIndex + 1] = ENCODE_TABLE[i >> 12 & 0x3f];
        outData[outIndex + 2] = ENCODE_TABLE[i >> 6 & 0x3f];
        outData[outIndex + 3] = ENCODE_TABLE[i & 0x3f];
    }

	CBZ	R5, not_equal
	BNE	chunk64_loop
	AND	$0x3f, R2, R2
	CBZ	R2, equal
chunk16:
	// work with 16-byte chunks
	BIC	$0xf, R2, R3
	CBZ	R3, tail
	ADD	R3, R0, R6	// end of chunks
chunk16_loop:
	LDP.P	16(R0), (R4, R5)
	LDP.P	16(R1), (R7, R9)
	EOR	R4, R7
	CBNZ	R7, not_equal
	EOR	R5, R9
	CBNZ	R9, not_equal
	CMP	R0, R6
	BNE	chunk16_loop
	AND	$0xf, R2, R2
	CBZ	R2, equal
tail:

	0x28, 0xd7, 0xd9, 0x79, 0x95, 0xd3, 0xdd, 0x60, 0x34, 0x91, 0x22, 0xf7, 0x7f, 0x0f, 0x62, 0x1f,
	0x35, 0xfe, 0xaa, 0x3f, 0xe7, 0x29, 0x98, 0x65, 0x4b, 0xb3, 0x4f, 0xdc, 0xe6, 0x96, 0xe4, 0x16,
	0xe1, 0xff, 0x33, 0xd9, 0xec, 0x20, 0x63, 0xd7, 0x14, 0x1b, 0xa8, 0x14, 0x43, 0x5e, 0x24, 0x52,
	0xf1, 0x97, 0xc7, 0x65, 0x0e, 0x8a, 0xcd, 0x17, 0x50, 0xe4, 0x50, 0xf3, 0x57, 0xc5, 0xad, 0xde,

	BYTE $0x30; BYTE $0x31; BYTE $0x1a; BYTE $0x33
	BYTE $0x34; BYTE $0x35; BYTE $0x36; BYTE $0x08
	BYTE $0x38; BYTE $0x39; BYTE $0x3a; BYTE $0x3b
	BYTE $0x04; BYTE $0x14; BYTE $0x3e; BYTE $0xff
	BYTE $0x41; BYTE $0xaa; BYTE $0x4a; BYTE $0xb1
	BYTE $0x9f; BYTE $0xb2; BYTE $0x6a; BYTE $0xb5
	BYTE $0xbb; BYTE $0xb4; BYTE $0x9a; BYTE $0x8a
	BYTE $0xb0; BYTE $0xca; BYTE $0xaf; BYTE $0xbc

// trailing CCC shifted left 2 bits. The two least-significant bits of tccc
// are the number of trailing non-starters.

const (
	qcInfoMask      = 0x3F // to clear all but the relevant bits in a qcInfo
	headerLenMask   = 0x3F // extract the length value from the header byte
	headerFlagsMask = 0xC0 // extract the qcInfo bits from the header byte
)

	// to how other linkers handle this (for example this code in lld:
	// https://github.com/llvm/llvm-project/blob/2a3a79ce4c2149d7787d56f9841b66cacc9061d0/lld/MachO/Writer.cpp#L524).
	rv[6] &= 0xcf
	rv[6] |= 0x30
	rv[8] &= 0x3f
	rv[8] |= 0xc0

	return rv
}

// machoRewriteUuid copies over the contents of the Macho executable
// exef into the output file outexe, and in the process updates the

   * also demonstrates how sparse the lookup table is.
   */
  @Test
  fun allFormattedFlagsWithValidBits() {
    val formattedFlags = mutableListOf<String>() // Highest valid flag is 0x20.
    for (i in 0..0x3f) formattedFlags.add(formatFlags(TYPE_HEADERS, i))
    assertThat(formattedFlags).containsExactly(
      "",
      "END_STREAM",
      "00000010",
      "00000011",
      "END_HEADERS",

			}
			outbuf[out4] = byte(t4 >> 8)
			out4++
			rbr4.cnt -= uint32(t4 & 0xff)
		}

		outbuf[out1] = byte(t1 >> 8)
		out1++
		rbr1.cnt -= uint32(t1 & 0xff)

		outbuf[out2] = byte(t2 >> 8)
		out2++
		rbr2.cnt -= uint32(t2 & 0xff)

		outbuf[out3] = byte(t3 >> 8)
		out3++
		rbr3.cnt -= uint32(t3 & 0xff)
	}

	return outbuf, nil

}

func TestScalarNonAdjacentForm(t *testing.T) {
	s, _ := (&Scalar{}).SetCanonicalBytes([]byte{
		0x1a, 0x0e, 0x97, 0x8a, 0x90, 0xf6, 0x62, 0x2d,
		0x37, 0x47, 0x02, 0x3f, 0x8a, 0xd8, 0x26, 0x4d,
		0xa7, 0x58, 0xaa, 0x1b, 0x88, 0xe0, 0x40, 0xd1,
		0x58, 0x9e, 0x7b, 0x7f, 0x23, 0x76, 0xef, 0x09,
	})

	expectedNaf := [256]int8{
		0, 13, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, -9, 0, 0, 0, 0, -11, 0, 0, 0, 0, 3, 0, 0, 0, 0, 1,