Building an array using parfor

Hi,

I am building an array using parfor in the following fashion :

parfor ifreq = 1:80
  for iSig   =  1:140
    [SigVal,~] = somefunction(NecessaryInputs)
    AllSig(ifreq,iSig,:) = SigVal;
  end
end

However this is throwing an error that the variable AllSig cannot be classified. Any ideas how to get this to work ?

Answer 1

[Edric Ellis]

What version of MATLAB are you using? I tried this in R2015b, and it worked as expected:

parfor ifreq = 1:80
  for iSig   =  1:140
    SigVal = rand(3,1);
    AllSig(ifreq,iSig,:) = SigVal;
  end
end

[Moiz]

I am using R2014b

Answer 2

[Walter Roberson]

parfor ifreq = 1:80
  SigVal = zeros(140, 3);  %for example
  for iSig   =  1:140
    [SigVal(iSig, :),~] = somefunction(NecessaryInputs)
  end
  AllSig(ifreq,:,:) = SigVal;
end

[Moiz]

Hi Walter,

This works except that I found parfor to take a huge amount of time (more than the regular for loop). So I am not sure what was going on..

[Walter Roberson]

parfor is not necessarily faster, due to the various overheads involved. Also because the workers are restricted to one CPU each whereas non-parallel array computations that are "big enough" get sent to BLAS or LAPACK which operates on all CPUs on the host. Therefore any one computation can take several times longer in parallel.