layer3.c

motorolfa ezx平台的mp3播放器

  if (header->mode_extension & MS_STEREO) {
    register mad_fixed_t invsqrt2;

    header->flags |= MAD_FLAG_MS_STEREO;

    invsqrt2 = root_table[3 + -2];

    for (sfbi = l = 0; l < 576; ++sfbi, l += n) {
      n = sfbwidth[sfbi];

      if (modes[sfbi] != MS_STEREO)
	continue;

      for (i = 0; i < n; ++i) {
	register mad_fixed_t m, s;

	m = xr[0][l + i];
	s = xr[1][l + i];

	xr[0][l + i] = mad_f_mul(m + s, invsqrt2);  /* l = (m + s) / sqrt(2) */
	xr[1][l + i] = mad_f_mul(m - s, invsqrt2);  /* r = (m - s) / sqrt(2) */
      }
    }
  }

  return MAD_ERROR_NONE;
}

/*
 * NAME:	III_aliasreduce()
 * DESCRIPTION:	perform frequency line alias reduction
 */
static
void III_aliasreduce(mad_fixed_t xr[576], int lines)
{
  mad_fixed_t const *bound;
  int i;

  bound = &xr[lines];
  for (xr += 18; xr < bound; xr += 18) {
    for (i = 0; i < 8; ++i) {
      register mad_fixed_t a, b;
      register mad_fixed64hi_t hi;
      register mad_fixed64lo_t lo;

      a = xr[-1 - i];
      b = xr[     i];

# if defined(ASO_ZEROCHECK)
      if (a | b) {
# endif
	MAD_F_ML0(hi, lo,  a, cs[i]);
	MAD_F_MLA(hi, lo, -b, ca[i]);

	xr[-1 - i] = MAD_F_MLZ(hi, lo);

	MAD_F_ML0(hi, lo,  b, cs[i]);
	MAD_F_MLA(hi, lo,  a, ca[i]);

	xr[     i] = MAD_F_MLZ(hi, lo);
# if defined(ASO_ZEROCHECK)
      }
# endif
    }
  }
}

# if defined(ASO_IMDCT)
void III_imdct_l2(mad_fixed_t const [18], mad_fixed_t [36], unsigned int);
# else
#  if 1
static
void fastsdct(mad_fixed_t const x[9], mad_fixed_t y[18])
{
  mad_fixed_t a0,  a1,  a2,  a3,  a4,  a5,  a6,  a7,  a8,  a9,  a10, a11, a12;
  mad_fixed_t a13, a14, a15, a16, a17, a18, a19, a20, a21, a22, a23, a24, a25;
  mad_fixed_t m0,  m1,  m2,  m3,  m4,  m5,  m6,  m7;

  enum {
    c0 =  MAD_F(0x1f838b8d),  /* 2 * cos( 1 * PI / 18) */
    c1 =  MAD_F(0x1bb67ae8),  /* 2 * cos( 3 * PI / 18) */
    c2 =  MAD_F(0x18836fa3),  /* 2 * cos( 4 * PI / 18) */
    c3 =  MAD_F(0x1491b752),  /* 2 * cos( 5 * PI / 18) */
    c4 =  MAD_F(0x0af1d43a),  /* 2 * cos( 7 * PI / 18) */
    c5 =  MAD_F(0x058e86a0),  /* 2 * cos( 8 * PI / 18) */
    c6 = -MAD_F(0x1e11f642)   /* 2 * cos(16 * PI / 18) */
  };

  a0 = x[3] + x[5];
  a1 = x[3] - x[5];
  a2 = x[6] + x[2];
  a3 = x[6] - x[2];
  a4 = x[1] + x[7];
  a5 = x[1] - x[7];
  a6 = x[8] + x[0];
  a7 = x[8] - x[0];

  a8  = a0  + a2;
  a9  = a0  - a2;
  a10 = a0  - a6;
  a11 = a2  - a6;
  a12 = a8  + a6;
  a13 = a1  - a3;
  a14 = a13 + a7;
  a15 = a3  + a7;
  a16 = a1  - a7;
  a17 = a1  + a3;

  m0 = mad_f_mul(a17, -c3);
  m1 = mad_f_mul(a16, -c0);
  m2 = mad_f_mul(a15, -c4);
  m3 = mad_f_mul(a14, -c1);
  m4 = mad_f_mul(a5,  -c1);
  m5 = mad_f_mul(a11, -c6);
  m6 = mad_f_mul(a10, -c5);
  m7 = mad_f_mul(a9,  -c2);

  a18 =     x[4] + a4;
  a19 = 2 * x[4] - a4;
  a20 = a19 + m5;
  a21 = a19 - m5;
  a22 = a19 + m6;
  a23 = m4  + m2;
  a24 = m4  - m2;
  a25 = m4  + m1;

  /* output to every other slot for convenience */

  y[ 0] = a18 + a12;
  y[ 2] = m0  - a25;
  y[ 4] = m7  - a20;
  y[ 6] = m3;
  y[ 8] = a21 - m6;
  y[10] = a24 - m1;
  y[12] = a12 - 2 * a18;
  y[14] = a23 + m0;
  y[16] = a22 + m7;
}

static 
#ifndef __WINS__
inline
#endif
void sdctII(mad_fixed_t const x[18], mad_fixed_t X[18])
{
  mad_fixed_t tmp[9];
  int i;

  /* scale[i] = 2 * cos(PI * (2 * i + 1) / (2 * 18)) */
  static mad_fixed_t const scale[9] = {
    MAD_F(0x1fe0d3b4), MAD_F(0x1ee8dd47), MAD_F(0x1d007930),
    MAD_F(0x1a367e59), MAD_F(0x16a09e66), MAD_F(0x125abcf8),
    MAD_F(0x0d8616bc), MAD_F(0x08483ee1), MAD_F(0x02c9fad7)
  };

  /* divide the 18-point SDCT-II into two 9-point SDCT-IIs */

  /* even input butterfly */

  for (i = 0; i < 9; i += 3) {
    tmp[i + 0] = x[i + 0] + x[18 - (i + 0) - 1];
    tmp[i + 1] = x[i + 1] + x[18 - (i + 1) - 1];
    tmp[i + 2] = x[i + 2] + x[18 - (i + 2) - 1];
  }

  fastsdct(tmp, &X[0]);

  /* odd input butterfly and scaling */

  for (i = 0; i < 9; i += 3) {
    tmp[i + 0] = mad_f_mul(x[i + 0] - x[18 - (i + 0) - 1], scale[i + 0]);
    tmp[i + 1] = mad_f_mul(x[i + 1] - x[18 - (i + 1) - 1], scale[i + 1]);
    tmp[i + 2] = mad_f_mul(x[i + 2] - x[18 - (i + 2) - 1], scale[i + 2]);
  }

  fastsdct(tmp, &X[1]);

  /* output accumulation */

  for (i = 3; i < 18; i += 8) {
    X[i + 0] -= X[(i + 0) - 2];
    X[i + 2] -= X[(i + 2) - 2];
    X[i + 4] -= X[(i + 4) - 2];
    X[i + 6] -= X[(i + 6) - 2];
  }
}

static 
#ifndef __WINS__
inline
#endif
void dctIV(mad_fixed_t const y[18], mad_fixed_t X[18])
{
  mad_fixed_t tmp[18];
  int i;

  /* scale[i] = 2 * cos(PI * (2 * i + 1) / (4 * 18)) */
  static mad_fixed_t const scale[18] = {
    MAD_F(0x1ff833fa), MAD_F(0x1fb9ea93), MAD_F(0x1f3dd120),
    MAD_F(0x1e84d969), MAD_F(0x1d906bcf), MAD_F(0x1c62648b),
    MAD_F(0x1afd100f), MAD_F(0x1963268b), MAD_F(0x1797c6a4),
    MAD_F(0x159e6f5b), MAD_F(0x137af940), MAD_F(0x11318ef3),
    MAD_F(0x0ec6a507), MAD_F(0x0c3ef153), MAD_F(0x099f61c5),
    MAD_F(0x06ed12c5), MAD_F(0x042d4544), MAD_F(0x0165547c)
  };

  /* scaling */

  for (i = 0; i < 18; i += 3) {
    tmp[i + 0] = mad_f_mul(y[i + 0], scale[i + 0]);
    tmp[i + 1] = mad_f_mul(y[i + 1], scale[i + 1]);
    tmp[i + 2] = mad_f_mul(y[i + 2], scale[i + 2]);
  }

  /* SDCT-II */

  sdctII(tmp, X);

  /* scale reduction and output accumulation */

  X[0] /= 2;
  for (i = 1; i < 17; i += 4) {
    X[i + 0] = X[i + 0] / 2 - X[(i + 0) - 1];
    X[i + 1] = X[i + 1] / 2 - X[(i + 1) - 1];
    X[i + 2] = X[i + 2] / 2 - X[(i + 2) - 1];
    X[i + 3] = X[i + 3] / 2 - X[(i + 3) - 1];
  }
  X[17] = X[17] / 2 - X[16];
}

/*
 * NAME:	imdct36
 * DESCRIPTION:	perform X[18]->x[36] IMDCT using Szu-Wei Lee's fast algorithm
 */
static 
#ifndef __WINS__
inline
#endif
void imdct36(mad_fixed_t const x[18], mad_fixed_t y[36])
{
  mad_fixed_t tmp[18];
  int i;

  /* DCT-IV */

  dctIV(x, tmp);

  /* convert 18-point DCT-IV to 36-point IMDCT */

  for (i =  0; i <  9; i += 3) {
    y[i + 0] =  tmp[9 + (i + 0)];
    y[i + 1] =  tmp[9 + (i + 1)];
    y[i + 2] =  tmp[9 + (i + 2)];
  }
  for (i =  9; i < 27; i += 3) {
    y[i + 0] = -tmp[36 - (9 + (i + 0)) - 1];
    y[i + 1] = -tmp[36 - (9 + (i + 1)) - 1];
    y[i + 2] = -tmp[36 - (9 + (i + 2)) - 1];
  }
  for (i = 27; i < 36; i += 3) {
    y[i + 0] = -tmp[(i + 0) - 27];
    y[i + 1] = -tmp[(i + 1) - 27];
    y[i + 2] = -tmp[(i + 2) - 27];
  }
}
#  else
/*
 * NAME:	imdct36
 * DESCRIPTION:	perform X[18]->x[36] IMDCT
 */
static 
#ifndef __WINS__
inline
#endif
void imdct36(mad_fixed_t const X[18], mad_fixed_t x[36])
{
  mad_fixed_t t0, t1, t2,  t3,  t4,  t5,  t6,  t7;
  mad_fixed_t t8, t9, t10, t11, t12, t13, t14, t15;
  register mad_fixed64hi_t hi;
  register mad_fixed64lo_t lo;

  MAD_F_ML0(hi, lo, X[4],  MAD_F(0x0ec835e8));
  MAD_F_MLA(hi, lo, X[13], MAD_F(0x061f78aa));

  t6 = MAD_F_MLZ(hi, lo);

  MAD_F_MLA(hi, lo, (t14 = X[1] - X[10]), -MAD_F(0x061f78aa));
  MAD_F_MLA(hi, lo, (t15 = X[7] + X[16]), -MAD_F(0x0ec835e8));

  t0 = MAD_F_MLZ(hi, lo);

  MAD_F_MLA(hi, lo, (t8  = X[0] - X[11] - X[12]),  MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, (t9  = X[2] - X[9]  - X[14]),  MAD_F(0x09bd7ca0));
  MAD_F_MLA(hi, lo, (t10 = X[3] - X[8]  - X[15]), -MAD_F(0x0cb19346));
  MAD_F_MLA(hi, lo, (t11 = X[5] - X[6]  - X[17]), -MAD_F(0x0fdcf549));

  x[7]  = MAD_F_MLZ(hi, lo);
  x[10] = -x[7];

  MAD_F_ML0(hi, lo, t8,  -MAD_F(0x0cb19346));
  MAD_F_MLA(hi, lo, t9,   MAD_F(0x0fdcf549));
  MAD_F_MLA(hi, lo, t10,  MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, t11, -MAD_F(0x09bd7ca0));

  x[19] = x[34] = MAD_F_MLZ(hi, lo) - t0;

  t12 = X[0] - X[3] + X[8] - X[11] - X[12] + X[15];
  t13 = X[2] + X[5] - X[6] - X[9]  - X[14] - X[17];

  MAD_F_ML0(hi, lo, t12, -MAD_F(0x0ec835e8));
  MAD_F_MLA(hi, lo, t13,  MAD_F(0x061f78aa));

  x[22] = x[31] = MAD_F_MLZ(hi, lo) + t0;

  MAD_F_ML0(hi, lo, X[1],  -MAD_F(0x09bd7ca0));
  MAD_F_MLA(hi, lo, X[7],   MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, X[10], -MAD_F(0x0fdcf549));
  MAD_F_MLA(hi, lo, X[16],  MAD_F(0x0cb19346));

  t1 = MAD_F_MLZ(hi, lo) + t6;

  MAD_F_ML0(hi, lo, X[0],   MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[3],  -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[5],  -MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[6],   MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[8],  -MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[9],   MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[11],  MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[15], -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x0f9ee890));

  x[6]  = MAD_F_MLZ(hi, lo) + t1;
  x[11] = -x[6];

  MAD_F_ML0(hi, lo, X[0],  -MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[2],  -MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[3],   MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[5],   MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[6],   MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[8],  -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[9],  -MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[15],  MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[17],  MAD_F(0x04cfb0e2));

  x[23] = x[30] = MAD_F_MLZ(hi, lo) + t1;

  MAD_F_ML0(hi, lo, X[0],  -MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[3],  -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[5],   MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[6],   MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[8],  -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[9],  -MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[11],  MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[15],  MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x0acf37ad));

  x[18] = x[35] = MAD_F_MLZ(hi, lo) - t1;

  MAD_F_ML0(hi, lo, X[4],   MAD_F(0x061f78aa));
  MAD_F_MLA(hi, lo, X[13], -MAD_F(0x0ec835e8));

  t7 = MAD_F_MLZ(hi, lo);

  MAD_F_MLA(hi, lo, X[1],  -MAD_F(0x0cb19346));
  MAD_F_MLA(hi, lo, X[7],   MAD_F(0x0fdcf549));
  MAD_F_MLA(hi, lo, X[10],  MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, X[16], -MAD_F(0x09bd7ca0));

  t2 = MAD_F_MLZ(hi, lo);

  MAD_F_MLA(hi, lo, X[0],   MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[3],  -MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[5],   MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[6],  -MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[8],  -MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[9],   MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[12],  MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[15],  MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[17],  MAD_F(0x0f426cb5));

  x[5]  = MAD_F_MLZ(hi, lo);
  x[12] = -x[5];

  MAD_F_ML0(hi, lo, X[0],   MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[2],  -MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[3],   MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[5],  -MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[6],  -MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[8],   MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[9],  -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[11],  MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[15],  MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x0bcbe352));

  x[0]  = MAD_F_MLZ(hi, lo) + t2;
  x[17] = -x[0];

  MAD_F_ML0(hi, lo, X[0],  -MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[2],  -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[3],  -MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[5],   MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[6],   MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[8],   MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[9],   MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[14], -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[15], -MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x03768962));

  x[24] = x[29] = MAD_F_MLZ(hi, lo) + t2;

  MAD_F_ML0(hi, lo, X[1],  -MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, X[7],  -MAD_F(0x09bd7ca0));
  MAD_F_MLA(hi, lo, X[10],  MAD_F(0x0cb19346));
  MAD_F_MLA(hi, lo, X[16],  MAD_F(0x0fdcf549));

  t3 = MAD_F_MLZ(hi, lo) + t7;

  MAD_F_ML0(hi, lo, X[0],   MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[3],  -MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[5],  -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[6],   MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[8],   MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[9],  -MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[12],  MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[15], -MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x0ffc19fd));

  x[8] = MAD_F_MLZ(hi, lo) + t3;
  x[9] = -x[8];

  MAD_F_ML0(hi, lo, X[0],  -MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[3],   MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[5],  -MAD_F(0x0898c779));
  MAD_F_MLA(hi, lo, X[6],  -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[8],   MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[9],   MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[12], -MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[14], -MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[15],  MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[17],  MAD_F(0x07635284));

  x[21] = x[32] = MAD_F_MLZ(hi, lo) + t3;

  MAD_F_ML0(hi, lo, X[0],  -MAD_F(0x0d7e8807));
  MAD_F_MLA(hi, lo, X[2],   MAD_F(0x0f426cb5));
  MAD_F_MLA(hi, lo, X[3],   MAD_F(0x0acf37ad));
  MAD_F_MLA(hi, lo, X[5],  -MAD_F(0x0ffc19fd));
  MAD_F_MLA(hi, lo, X[6],  -MAD_F(0x07635284));
  MAD_F_MLA(hi, lo, X[8],   MAD_F(0x0f9ee890));
  MAD_F_MLA(hi, lo, X[9],   MAD_F(0x03768962));
  MAD_F_MLA(hi, lo, X[11], -MAD_F(0x0e313245));
  MAD_F_MLA(hi, lo, X[12],  MAD_F(0x00b2aa3e));
  MAD_F_MLA(hi, lo, X[14],  MAD_F(0x0bcbe352));
  MAD_F_MLA(hi, lo, X[15], -MAD_F(0x04cfb0e2));
  MAD_F_MLA(hi, lo, X[17], -MAD_F(0x0898c779));

  x[20] = x[33] = MAD_F_MLZ(hi, lo) - t3;

  MAD_F_ML0(hi, lo, t14, -MAD_F(0x0ec835e8));
  MAD_F_MLA(hi, lo, t15,  MAD_F(0x061f78aa));

  t4 = MAD_F_MLZ(hi, lo) - t7;

  MAD_F_ML0(hi, lo, t12, MAD_F(0x061f78aa));
  MAD_F_MLA(hi, lo, t13, MAD_F(0x0ec835e8));

  x[4]  = MAD_F_MLZ(hi, lo) + t4;
  x[13] = -x[4];

  MAD_F_ML0(hi, lo, t8,   MAD_F(0x09bd7ca0));
  MAD_F_MLA(hi, lo, t9,  -MAD_F(0x0216a2a2));
  MAD_F_MLA(hi, lo, t10,  MAD_F(0x0fdcf549));
  MAD_F_MLA(hi, lo, t11, -MAD_F(0x0cb19346));

  x[1]  = MAD_F_MLZ(hi, lo) + t4;
  x[16] = -x[1];

  MAD_F_ML0(hi, lo, t8,  -MAD_F(0x0fdcf549));
  MAD_F_MLA(hi, lo, t9,  -MAD_F(0x0cb19346));
  MAD_F_MLA(hi, lo, t10, -MAD_F(0x09bd7ca0));
  MAD_F_MLA(hi, lo, t11, -MAD_F(0x0216a2a2));