StarCore code optimizer ( sco ) is capable of handling different coding techniques that often used to implement many of DSP algorithms.
For example the classical DSP kernel for "NTAPS"-tap FIR filter processing "NPOINTS" points

may be coded with:

1. use of "pointers":

void fir( int *input,        /* pointer to input sample array of NTAPS points*/
             int *output,      /* pointer to output array of NPOINTS points */
             int *coefficient /* pointer to coefficient array of NTAPS values */
           )
 {
  int i, j;
  int *data, *coef, *initial;
  int sum;
  int tmp1, tmp2;

  initial = input;
  for ( i = 0; i < NPOINTS; i++ )
   {
    data = input++;
    coef = coefficient;
    sum = 0;
    for (j = 0; j < NTAPS; j++)
     {
      tmp1 = *data;
      if ( data == initial )
       {
        data = initial + ( NTAPS - 1 );
       }
      else
       {
        data--;
       }
      tmp2 = *coef++;
      sum += tmp1 * tmp2;
     }
    *output++ = sum;
   }
 }

2. use of "pointers with a swap":

3. use of "arrays":

4. use of "arrays with a swap":

void fir( int input[],        /* input sample array of NTAPS points */
             int output[],      /* output array of NPOINTS points */
             int coefficient[] /* array of NTAPS coefficients */
           )
 {
  int i;
  int j;
  int sum, tmp1, tmp2;

  for ( i = 0; i < NPOINTS; i++ )
   {
    sum = 0;
    tmp1 = input[i];
    tmp2 = coefficient[0];
    for (j = 1; j < NTAPS; j++)
     {
      sum += tmp1 * tmp2;
      if ( ( i - j ) < 0 )
       {
        tmp1 = input[i-j+NTAPS];
       }
      else
       {
        tmp1 = input[i-j];
       }
      tmp2 = coefficient[j];
     }
    sum += tmp1 * tmp2;
    output[i] = sum;
   }
}

The following table lists execution time improvements achieved by sco over different implementations of the
256-tap FIR filter processing 64 points

256-tap FIR filter processing 64 points

pointers 28% pointers with a swap 24% arrays 27% arrays with a swap 24%