C / C ++脚本优化

Question

I have 3 matrices, gX , gY and gNorm , and one for loop like this:

for(y=1; y<n-1; y++){
      int base = y*this->_nxsIn;
      for(x=1; x<n-1; x++){ 
          i1 = base + x;
          i2 = i1 + 1;
          i3 = i1 + n;
          den = ( (gX[i1]+gY[i1])*gNorm[i1] - gX[i2]*gNorm[i2] - gY[i3]*gNorm[i3];
}

If I extract the products gX * gNorm and gY * gNorm in this way:

for(y=0; y<n; y++){
      int base = y*n;
      for(x=0; x<n; x++){
           i = base + x;
          GxGnorm[i] = gX[i]*gNorm[i];
       }
   }

//Gy * GNorm

for(y=0; y<n; y++){
     int base = y*n;
     for(x=0; x<n; x++){
          i = base + x;
         GyGnorm[i] = gY[i]*gNorm[i];
      }
}


for(y=1; y<this->_nysIn-1; y++){
      int base = y*this->_nxsIn;
      for(x=1; x<this->_nxsIn-1; x++){  
      i1 = base + x;
      i2 = i1 + 1;
      i3 = i1 + n;

      i1Sum = GxGnorm[i1] + GyGnorm[i1];         //(gX[i1]+gY[i1])*gNorm[i1];
      float sum1 = GxGnorm[i2] + GyGnorm[i3];    //gX[i2]*gNorm[i2];

     }
}

Can I get improvements in computation time?

If no, why?

Answer 1

Well, both scripts have the same asymptotic time, O(n^2), which is bad if you have a substantial data set.

As you have repetitions of the nested loops in your second script, you are actually increasing the constant factor. Add to that the costs of creating the extra arrays.

Other than these both scripts are quite similar. So, the second script might actually be slower.