T.AddMethod(m)
		}

		// check method order
		if i == 0 {
			// first round: collect methods in given order
			methods = make([]string, T.NumMethods())
			for j := range methods {
				methods[j] = T.Method(j).Name()
			}
		} else {
			// successive rounds: methods must appear in the same order
			if got := T.NumMethods(); got != len(methods) {

  /**
   * Returns x rounded to either the greatest double less than or equal to the precise value of x,
   * or the least double greater than or equal to the precise value of x.
   */
  abstract double roundToDoubleArbitrarily(X x);

  /** Returns the sign of x: either -1, 0, or 1. */
  abstract int sign(X x);

  /** Returns d's value as an X, rounded with the specified mode. */

  /**
   * Returns x rounded to either the greatest double less than or equal to the precise value of x,
   * or the least double greater than or equal to the precise value of x.
   */
  abstract double roundToDoubleArbitrarily(X x);

  /** Returns the sign of x: either -1, 0, or 1. */
  abstract int sign(X x);

  /** Returns d's value as an X, rounded with the specified mode. */

				// Row is determined by the 1st and 6th bit.
				// Column is the middle four bits.
				row := uint8(((i & 2) << 4) | i&1)
				col := uint8(j << 1)
				t := row | col

				// The rotation was performed in the feistel rounds, being factored out and now mixed into the feistelBox.
				f = (f << 1) | (f >> 31)

				feistelBox[s][t] = uint32(f)
			}
		}
	}
}

// permuteInitialBlock is equivalent to the permutation defined

  int pw;  // Padding on "width" dimension (both sides).

  int sh;  // Stride on "height" dimension.
  int sw;  // Stride on "width" dimension.

  bool ceil_mode;  // Rounding strategy (ceil or floor).
};

// Rounds the dimension based on the ceil mode.
int RoundDim(float dim, bool ceil_mode) {
  if (ceil_mode) {
    return std::ceil(dim);
  }
  return std::floor(dim);
}

// the value of [x0, x1, x2] is x[0] + x[1] * 2⁶⁴ + x[2] * 2¹²⁸.
type macState struct {
	// h is the main accumulator. It is to be interpreted modulo 2¹³⁰ - 5, but
	// can grow larger during and after rounds. It must, however, remain below
	// 2 * (2¹³⁰ - 5).
	h [3]uint64
	// r and s are the private key components.
	r [2]uint64
	s [2]uint64
}

type macGeneric struct {
	macState

	buffer [TagSize]byte

	ROUND1(Rb, Rc, Rd, Ra,  3, 22, Rc3)

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND1(Ra, Rb, Rc, Rd,  4,	7, Rc0)
	ROUND1(Rd, Ra, Rb, Rc,  5, 12, Rc1)
	ROUND1(Rc, Rd, Ra, Rb,  6, 17, Rc2)
	ROUND1(Rb, Rc, Rd, Ra,  7, 22, Rc3)

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND1(Ra, Rb, Rc, Rd,  8,	7, Rc0)
	ROUND1(Rd, Ra, Rb, Rc,  9, 12, Rc1)
	ROUND1(Rc, Rd, Ra, Rb, 10, 17, Rc2)
	ROUND1(Rb, Rc, Rd, Ra, 11, 22, Rc3)

	ROUND2(R4,R5,R6,R7,10,0xd62f105d, 5);
	ROUND2(R7,R4,R5,R6,15, 0x2441453, 9);
	ROUND2(R6,R7,R4,R5, 4,0xd8a1e681,14);
	ROUND2(R5,R6,R7,R4, 9,0xe7d3fbc8,20);
	ROUND2(R4,R5,R6,R7,14,0x21e1cde6, 5);
	ROUND2(R7,R4,R5,R6, 3,0xc33707d6, 9);
	ROUND2(R6,R7,R4,R5, 8,0xf4d50d87,14);
	ROUND2(R5,R6,R7,R4,13,0x455a14ed,20);
	ROUND2(R4,R5,R6,R7, 2,0xa9e3e905, 5);
	ROUND2(R7,R4,R5,R6, 7,0xfcefa3f8, 9);
	ROUND2(R6,R7,R4,R5,12,0x676f02d9,14);

	ROUND2(R4,R5,R22,R3,M15,0xd8a1e681,14);
	ROUND2(R3,R4,R5,R22,M04,0xe7d3fbc8,20);
	ROUND2(R22,R3,R4,R5,M09,0x21e1cde6, 5);
	ROUND2(R5,R22,R3,R4,M14,0xc33707d6, 9);
	ROUND2(R4,R5,R22,R3,M03,0xf4d50d87,14);
	ROUND2(R3,R4,R5,R22,M08,0x455a14ed,20);
	ROUND2(R22,R3,R4,R5,M13,0xa9e3e905, 5);
	ROUND2(R5,R22,R3,R4,M02,0xfcefa3f8, 9);
	ROUND2(R4,R5,R22,R3,M07,0x676f02d9,14);
	ROUND2(R3,R4,R5,R22,M12,0x8d2a4c8a,20);

	ROUND4(R4,R5,R2,R3, 5,0xab9423a7,15);
	ROUND4(R3,R4,R5,R2,12,0xfc93a039,21);
	ROUND4(R2,R3,R4,R5, 3,0x655b59c3, 6);
	ROUND4(R5,R2,R3,R4,10,0x8f0ccc92,10);
	ROUND4(R4,R5,R2,R3, 1,0xffeff47d,15);
	ROUND4(R3,R4,R5,R2, 8,0x85845dd1,21);
	ROUND4(R2,R3,R4,R5,15,0x6fa87e4f, 6);
	ROUND4(R5,R2,R3,R4, 6,0xfe2ce6e0,10);
	ROUND4(R4,R5,R2,R3,13,0xa3014314,15);
	ROUND4(R3,R4,R5,R2, 4,0x4e0811a1,21);
	ROUND4(R2,R3,R4,R5,11,0xf7537e82, 6);