Parfor to speed up loops

Question

Jw 2012-2-10

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/28678-parfor-to-speed-up-loops

Hi, i have a dual core processor, i wish to make use of parfor to speed up my for loop processing how do i do that? Basically i understand the i have to initialise matlabpool(2) and parfor i=1:NPARTICLES and also matlabpool close.

How do i implement it in my coding? Thanks

      if CDiff < ODiff
          dt = CDiff;
          % Predict step
            for i=1:NPARTICLES
              particles(i)= predict (particles(i), V,G,Qe, WHEELBASE,dt, SWITCH_PREDICT_NOISE);
            end
          cIndex = cIndex + 1;
      else
          dt = ODiff;
          z1 = Ob(:,OIndex);
          z = cell2mat (z1);
          z(1) = z(1)/100;
          % Predict step
          for i=1:NPARTICLES
              particles(i)= predict (particles(i), V,G,Qe, WHEELBASE,dt, SWITCH_PREDICT_NOISE);
          end
          % Perform update
          for i=1:NPARTICLES
              [zf,idf,zn]= data_associate(particles(i),z,R);
              if ~isempty(zf) % observe map features
                  w= compute_weight(particles(i), zf,idf, R); % w = p(z_k | x_k)
                     if w <= 1e-5
                       w = 1e-5;
                     end
                  particles(i).w= particles(i).w * w;
                  particles(i)= feature_update(particles(i), zf, idf, R);
              end
              if ~isempty(zn) % observe new features, augment map
                  particles(i)= add_feature(particles(i), zn,R);
              end
          end
          particles= resample_particles(particles, NEFFECTIVE, SWITCH_RESAMPLE);
          dt2 = timestamp(cIndex+1)-Obser_time(OIndex);
          for i=1:NPARTICLES
             particles(i)= predict (particles(i), V,G,Qe, WHEELBASE,dt2, SWITCH_PREDICT_NOISE);
          end

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Ken Atwell 2012-2-10

2
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/28678-parfor-to-speed-up-loops#answer_37013

You code looks good to go.

Replace the "for" with "parfor"
"matlabpool open" before starting
"matlabpool close" when you're all done

But, before you do any of that, in deference to Amdahl's Law, I would be tempted to run this code through the profiler first to be certain that the bits to be parallelized make up a meaningful percentage of the program.