void sinCos_q27(uint32_t theta, int32_t &c, int32_t &s) {
  int32_t a = (theta & ((1 << 20) - 1)) << 8; // q28
  uint32_t i = theta >> 20;
  s = ___SMMLA(a, sine2t[i + 1] - sine2t[i], sine2t[i] >> 4);
  i = (i + 1024) & 4095;
  c = ___SMMLA(a, sine2t[i + 1] - sine2t[i], sine2t[i] >> 4);
}
int32_t coefA = attr_invA ? 1 << 14 : -1 << 14;
int32_t coefB = attr_invB ? 1 << 14 : -1 << 14;

int32_t c, s;
{
  uint32_t theta = ((param_theta + inlet_theta) & 0x07FFFFFF) << 5;
  sinCos_q27(theta, c, s);
}

// note: read/write indexes are set BUFSIZE samples back as we will process a
// batch of BUFSIZE samples so beware of overlappings when param_delays defer by
// less than 16 samples
int32_t a_rw = (attr_delayNameA.writepos - (BUFSIZE - 1) - param_posA) &
               attr_delayNameA.LENGTHMASK;
int32_t b_rw = (attr_delayNameB.writepos - (BUFSIZE - 1) - param_posB) &
               attr_delayNameB.LENGTHMASK;

// _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
for (int i = 0; i < BUFSIZE; i++) {
  int32_t xA = attr_delayNameA.array[a_rw] << 14;
  int32_t xB = attr_delayNameB.array[b_rw] << 14;
  attr_delayNameA.array[a_rw] =
      __SSAT(___SMMLS(xB, s, ___SMMUL(xA, c)) >> (24 - 15), 16);
  a_rw = (a_rw + 1) & attr_delayNameA.LENGTHMASK;
  attr_delayNameB.array[b_rw] =
      __SSAT(___SMMLA(xB, c, ___SMMUL(xA, s)) >> (24 - 15), 16);
  b_rw = (b_rw + 1) & attr_delayNameB.LENGTHMASK;
}