The variable in parfor cant classified

parpool(3)
parfor n = 1:N

      a=n/60;
      for u_p=1:length(t)
          x_a(n,u_p)=sqrt(1-1i*cot(a*pi/2)) *integral(@(u) ...
              rectpuls(u).*exp(1i*pi* (  cot(a*pi/2)*t(u_p)^2   -2*csc(a*pi/2)*u*t(u_p)   +cot(a*pi/2)*u.^2  )) ,-               HalfDur,HalfDur );
      end

end

Pretty sure I do not access the same location, but I might be wrong. What am I doing wrong?

on a side note, how can I write that inner loop better? i couldnt vector it

Answer 1

[Walter Roberson]

parfor requires that it be obvious that the same location cannot be written to multiple times.

You should write your output to a vector indexed just by n. You can then reshape the vector afterwards.

[htrh5]

so im going to have to copy an array around each loop. thats annoying.

i have another question: forget about parfor, how would i get rid of the inner loop?

[Walter Roberson]

The change to your code would be minor:

parpool(3)
parfor n = 1:N
  a=n/60;
  x_au = zeros(length(t),1);
  for u_p=1:length(t)
    x_au(u_p) = sqrt(1-1i*cot(a*pi/2)) *integral(@(u) ...
      rectpuls(u).*exp(1i*pi* (  cot(a*pi/2)*t(u_p)^2   -2*csc(a*pi/2)*u*t(u_p)   +cot(a*pi/2)*u.^2  )), -HalfDur, HalfDur );
  end
  x_a(n,:) = x_au;
end

With regards to getting rid of the inner loop: Is HalfDur certain to be < 1/2, or is it certain to be > 1/2 ? Either way there is a closed form for the integral, but the forms differ in how messy they are to write out. As there is a closed form, the u_p loop can be vectorized.

[htrh5]

rectpuls inside the integral could have been anything, so I am guessing that integral() cannot be evaluated with a vector parameter to produce a vector output?

[Walter Roberson]

Consider using the ArrayValued option of integral()

[htrh5]

Thanks, that's it.