How to optimize matlab code for gpu

Question

I have a function like,

function ObjVal=fun(Chrom,a)

  [Nind,Nvar] = size(Chrom);
  [m n]=size(a);

  for i=1:Nind
    c=Chrom(i,:);
    Cmat=repmat(c',1,n);
    ax=abs(sum(Cmat.*a)).^2;
    ObjVal(i)= 10*log10(max(ax)./(mean(ax)));
  end;

Chrom=16*16 and a=16*1024 array. I'm trying to find the fastest way on gpu. Only gpuarray is slower. When I use arrayfun or bsxfun I get some errors.

function valmult=mult(Cmat,a)

  valmult=abs(sum(Cmat.*a)).^2;

ax=arrayfun(@mult,Cmat,a); I get Function passed as first input argument contains unsupported or unknown function 'sum'.)

I am new in MATLAB with GPU. Any suggestions which way is the best and how can I maximize code performance with gpuarray?

kangshiyin · Accepted Answer · 2016-06-20 13:02:31Z

0

Things will be better if you can eliminate the for-loop. It can be done by the following code.

ax_mat = (Chrom * a).^2;
ObjVal = 10 * log10(max(ax_mat, [], 2) ./ mean(ax_mat, 2));

On the other hand, the heaviest operation in the code, the matrix multiplication of Chrom * a, is too small (16x16x1024) to fully utilize the GPU. You may find that even with the improved code, GPU is still slower.

edited Jun 20, 2016 at 13:02

answered Jun 20, 2016 at 12:57

kangshiyin

9,8291 gold badge19 silver badges30 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

How to optimize matlab code for gpu

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related