MATLAB: Codes runs too long

acoustic

Hi, i wonder if there is somebody who can help. I have this program where it calculates pressure inside a cavity. For normal case, the program just took 30 minutes to give result. I did modification to the program where the concerned codes a shown below, could this codes are the culprit where it took 15 hrs to run and give result. a, b, c, d are bessel functions that i dont bother to write here. Note: this is just partial of the program, i doubt other part is the reason because i only change numeric values for them.
HHvp=1e-10;
HHvp1=1e-10;
HHvp2=1e-10;
HHvp3=1e-10;
HHvp4=1e-10;
HHvp5=1e-10;
     for m1=1:5
         for n1=1:5
             for l1=1:5
              omega_n(m1,n1,l1)=m1*n1*l1 % a lot more to this, to long to write)
             end
         end
     end
     for n1=1:5
       for l1=1:5
          HHvp1= HHvp1 + a*n1*l1
       end
     end
     for n1=1:5
       for l1=1:5
          HHvp2= HHvp2 + b*n1*l1
       end
     end
     for n1=1:5
       for l1=1:5
          HHvp3= HHvp3 + c*n1*l1
       end
     end
     for n1=1:5
       for l1=1:5
          HHvp4= HHvp4 + d*n1*l1
       end
     end        
     HHvp=HHvp1+HHvp2+HHvp3+HHvp4;
     Hvp(mmm2,nnn2,lll2,num2)= HHvp;
[EDITED, Jan, code formatted]

Best Answer

In reply to your posted code:
When you use the profile to find the most time consuming lines, a standard procedure gets obvious: Avoid repeated calculations inside a loop. Here the Bessel function waste a lot of time and using temporary variables accelerate the code by a factor > 100:
r=1e-10+(lll2-1)*dr;  % After this line insert this:
tmp1 = 7.7*besselj(0,48.3*r)-0.32*bessely(0,48.3*r);
tmp2 = -22*besselj(0,96.7*r)-18.2*bessely(0,96.7*r);
tmp3 = -3*besselj(0,145*r)-17.5*bessely(0,145*r);
tmp4 = -0.8*besselj(0,193*r)-1.6*bessely(0,193*r);
tmp5 = -1.8*besselj(0,242*r)-0.37*bessely(0,242*r);
% And now replace the correspondig calculations by "tmpX" using the
% Replace dialog of the editor.
Btw., the code is extremely ugly. A substantial cleanup would at first increase the speed and at second allow a successful debugging. Most of all consider the MLint warnings (see the orange bars in the Editor). Then add comments and split the very large lines into readable pieces. Then it will be much easier to identify the parts, which are calculated repeatedly. Finally I guess, that a speedup of factor 1000 is possible without any magic or sophisticated tricks.
  • Is sqrt(m1*pi/(R2-R1))^2 really wanted?
  • In the calculation of Hvp_0_1_0 you assume that n1 has the last value from the last FOR loop, which uses n1 as counter. Better define n1 and l1 explicitly outside the loops.
  • Is "3.142" a crude approximation of pi? If so, don't do it. Never loose accuracy, when there is no benefit.
Related Question