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 lpA, lpB;

lpA = lpB = 0;

int32_t coefA = __USAT((1 << 27) - param_attenA - inlet_attenA, 27) >>
                (27 - 14);
coefA = attr_invA ? coefA : -coefA;

int32_t coefB = __USAT((1 << 27) - param_attenB - inlet_attenB, 27) >>
                (27 - 14);
coefB = attr_invB ? coefB : -coefB;

int32_t c, s;
{
  uint32_t theta = ((param_theta + inlet_theta) & 0x07FFFFFF) << 5;
  sinCos_q27(theta, c, s);
}
int32_t cLPA;
MTOFEXTENDED(param_LPA + inlet_LPA, cLPA);
cLPA = __USAT(cLPA, 30);
int32_t cLPB;
MTOFEXTENDED(param_LPB + inlet_LPB, cLPB);
cLPB = __USAT(cLPB, 30);

// 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 differ
// 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++) {
  //                              q15              q14    q29   q29     q27
  lpA = ___SMMLA((attr_delayNameA.array[a_rw] * coefA - lpA), cLPA, lpA >> 2)
        << 2;
  lpB = ___SMMLA((attr_delayNameB.array[b_rw] * coefB - lpB), cLPB, lpB >> 2)
        << 2;
  //                                      q24   q29 q27 24-15
  attr_delayNameA.array[a_rw] =
      __SSAT(___SMMLS(lpB, s, ___SMMUL(lpA, c)) >> (24 - 15), 16);
  a_rw = (a_rw + 1) & attr_delayNameA.LENGTHMASK;
  attr_delayNameB.array[b_rw] =
      __SSAT(___SMMLA(lpB, c, ___SMMUL(lpA, s)) >> (24 - 15), 16);
  b_rw = (b_rw + 1) & attr_delayNameB.LENGTHMASK;
}