Parfor gives NaN when for does not

I'm having issues parallelizing some code I use for a "homemade" spectrum analyzer.

It works fine in a for loop, however the

    data=smooth(data,window);
    Data=data(1+floor(window/2):length(data)-ceil(window/2));

part below (BoydFilter) causes issues (if I change it to Data=data; I get actual numbers) and my Data in ReadDAQTraces becomes a matrix of NaNs.

Can anyone tell me what I'm doing wrong?

*I and Q are arrays of 34.4 million doubles, if size is a problem.

function ReadDAQTraces(Folder,NumberOfTraces,Time,Points)
    global SampleRate
    SampleRate=2e6;

    Data=zeros(NumberOfTraces,Points);
    parfor i=1:NumberOfTraces
        Data(i,:)=ReadInDAQ(Folder,i,Time,Points)
        i
    end
    a=1;%for a debug stop

end

function Data=ReadInDAQ(Folder,i,Time,Points)

    if ~ischar(i)
        filename = strcat(Folder,'\I_Trace',int2str(i),'.csv');
    else
        filename = strcat(Folder,'\',i,'I.csv');
    end
    delimiter = {''};
    formatSpec = '%f%[^\n\r]';
    fileID = fopen(filename,'r');
    dataArray = textscan(fileID, formatSpec, 'Delimiter', delimiter, 'TextType', 'string', 'EmptyValue', NaN,  'ReturnOnError', false);
    fclose(fileID);
    I = dataArray{:, 1};
    
    if ~ischar(i)
        filename = strcat(Folder,'\Q_Trace',int2str(i),'.csv');
    else
        filename = strcat(Folder,'\',i,'Q.csv');
    end
    delimiter = {''};
    formatSpec = '%f%[^\n\r]';
    fileID = fopen(filename,'r');
    dataArray = textscan(fileID, formatSpec, 'Delimiter', delimiter, 'TextType', 'string', 'EmptyValue', NaN,  'ReturnOnError', false);
    fclose(fileID);
    Q = dataArray{:, 1};
    
    Data=ApplyFilter(I,Q,Time,Points);
end

function Data=ApplyFilter(I,Q,Time,Points)
    global SampleRate
    SamplingRate=SampleRate;
    
    window=Time*SamplingRate;
    
    I=BoydFilter(I,window).^2+BoydFilter(Q,window).^2;
    clear Q
    
    Data=AverageDetector(I,Points);
end

function Data=BoydFilter(data,window)
    if(window~=floor(window))
        window
        window=floor(window);
    end
    if(mod(window,2)==0)
        -1*window
        error('Boyd Filter Window is not an Odd Number')
    end
    data=smooth(data,window);
    Data=data(1+floor(window/2):length(data)-ceil(window/2));
end

function Output=AverageDetector(Input,TotalPoints)
    Output=zeros(1,TotalPoints);
    AverageOver=floor(length(Input)/TotalPoints);
    
    for i=1:TotalPoints
        Output(i)=mean(Input((i-1)*AverageOver+1:i*AverageOver));
    end
end

Answer 1

[Edric Ellis]

I'm not sure if this is the only problem, but the fact that you have a global variable set on the client and used by the workers is definitely a problem. The workers will not see the value set at the client (doc link). I suggest you modify your code to accept the SampleRate as an input parameter to your functions.

[Tim Woodworth]

Yep, that was it. Works now.

Thanks Edric!