Speed up code on GPU by removing nested for loops

Hello! I'm trying to speed up the following element-by-element array operation using the parallel computing toolbox. Is this a good candidate for arrayfun? What might that look like? Any advice would be greatly appreciated.

for ii = 1:Nx
  for jj = 1:Ny
       f1(ii,jj) = exp(i*pi*(ii + jj));
  end
end

Answer 1

[Bruno Luong]

f1 = exp((1i*pi) * (ii(:) + jj(:).'));

or on older MATLAB version (without auto expansion).

f1 = exp(bsxfun(@plus, (1i*pi) * ii(:), jj(:).'))

Avoid using ARRAYFUN with you like clear and fast code.

Answer 2

[Noah Walcutt]

Aw, heck yes!! Thanks!

Answer 3

[Noah Walcutt]

Another array expansion problem, but slightly different:

for ii = 1:Nx 
  for jj = 1:Ny
      u = delta0*(ii - Nx/2 -1);
      v = delta0*(jj - Ny/2 -1);
      p(ii,jj) = exp(1i*pi*lambda*z0*(u^2 + v^2));

    end
  end

Using bsxfun, I've got:

pp = exp(bsxfun(@plus, (1i*pi*lambda*z0*(delta0*(iii - Nx/2 -1)).^2), (jjj - Ny/2 -1).^2) );

What am I doing wrong here?

Answer 4

[Noah Walcutt]

Got it!

pp = exp(bsxfun(@plus, (1i*pi*lambda*z0*(delta0*(iii - Nx/2 -1)).^2), (1i*pi*lambda*z0*(delta0*(jjj - Ny/2 -1)).^2)));