Codebook Inner-Loop Code Listing: Carmel vs. Conventional



ITU-T fixed-point "C"reference

for (i3=1; i3<L_SUBFR; i3+=STEP)
      {
    ps2 = add(ps1, dn[i3]);
    alp2 = L_mac(alp1, *p3++, _1_8);
    alp2 = L_mac(alp2, *p4++, _1_2);
    sq2 = mult(ps2, ps2);
    alp_16 = round(alp2);
    s = L_msu(L_mult(alp,sq2),sq,alp_16);
    if (s > 0)
{
        sq = sq2;
        alp = alp_16;
        ix = i2;
        iy = i3;
}
      }



Carmel DSP implementation -- 3 cycles

repeat(8) block
  {
    cliw aclp1(r0 += rn0, r2++, r4++)
      {
 a5 = a4 + *ma2 * *ma3		/* L_mac(alp1, *p3++, _1_8) */
       || a0h = ff1 + *ma1 || nop		/* ps2 = add(ps1, dn[i3]) */
       || a2 = a2 - a5; /* s=L_msu(L_mult(alp_16,sq),sq2,alp); */
      }
    cliw aclp2(r1++, r2--, r6++)
      {
          a0 = a0h * a0h	  		 /* sq2 = mult(ps2, ps2) */
       || ifexe btr1 = *ma3  		 /* save index information */
       || a1h = round(a5 + *ma1 * *ma2)
      			/* round(alp2 = L_mac(alp2, *p4++, _1_2)) */
       || ifexe a3 = a1;       		/* alp = alp_16, sq = sq2 */
      }
    cliw aclp3()
      {
          a2 = a3l * a1h			/* L_mult(sq, alp_16) */
       || a1l = a0h
       || a5 = a3h * a0h			/* L_mult(sq2, alp) */
}
  }