dotproduct.c

// Unoptimized
// `0.81s to run 10000 tests (80750.60ns per test)`

// Method call 1x refactor
// `0.71s to run 10000 tests (70700.00ns per test)`
 
// Removing use of `get_vec_element` and fetching data upfront
// `0.22s to run 10000 tests (21686.00ns per test)`

// Loop unrolling using 5 accumulators
// `0.16s to run 10000 tests (16187.50ns per test)`

// Loop unrolling using 6 accumulators
// `0.16s to run 10000 tests (16145.60ns per test)`

// Variable renaming of temporary acc values
// `0.16s to run 10000 tests (16272.20ns per test)`
typedef long data_t;

data_t dotproduct(vec_ptr u, vec_ptr v) {
   data_t u_val1, v_val1, u_val2, v_val2, u_val3, v_val3, u_val4, v_val4, u_val5, v_val5;
   data_t sum1 = 0, sum2 = 0, sum3 = 0, sum4 = 0, sum5 = 0;
   long l = vec_length(u);

   data_t *u_data = u->data;
   data_t *v_data = v->data;
   long i = 0;

   for (; i < l - 4; i+=5) { // we can assume both vectors are same length
     u_val1 = u_data[i];
     v_val1 = v_data[i];
     sum1 += u_val1 * v_val1;

     u_val2 = u_data[i + 1];
     v_val2 = v_data[i + 1];
     sum2 += u_val2 * v_val2;

     u_val3 = u_data[i + 2];
     v_val3 = v_data[i + 2];
     sum3 += u_val3 * v_val3;

     u_val4 = u_data[i + 3];
     v_val4 = v_data[i + 3];
     sum4 += u_val4 * v_val4;

     u_val5 = u_data[i + 4];
     v_val5 = v_data[i + 4];
     sum5 += u_val5 * v_val5;
   }

   for (; i < l; i++) {
     u_val1 = u_data[i];
     v_val1 = v_data[i];
     sum1 += u_val1 * v_val1;
   }

   return sum1 + sum2 + sum3 + sum4 + sum5;
}