Slicing variable in parfor loop

Question

Arnab Samaddar-Chaudhuri 2022-11-17

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1853818-slicing-variable-in-parfor-loop

评论： Arnab Samaddar-Chaudhuri 2022-11-18

I have a problem with parfor. If I run the code, ofcourse I get an error " The PARFOR loop cannot run due to the way the variable 'completeCellPositions' and 'cellPos' is used ", since there is dependency of the value count from previous loop run.

My code so far:

count = 1;
xRange = [-2000,2000];
yRange = [-500,500];
parfor cellCOMX = xRange(1,1):5:xRange(1,2)
    for cellCOMY = yRange(1,1):5:yRange(1,2)
        
        [completeCellPositions{1,count}, cellPos{1,count}] = doesSomething(cellCOMX, cellCOMY);
        count = count+1;
    end
end

I am not sure, how to place sliced variable in this scenario. I cannot simply write

 [completeCellPositions{1,cellCOMX}, cellPos{1,cellCOMX}] = doesSomething(cellCOMX, cellCOMY);

Any suggestion, how to solve this issue?

PS: Ideally, I would like to use parfor for both for loops, but I can settle for even one parfor.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Edric Ellis 2022-11-17

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1853818-slicing-variable-in-parfor-loop#answer_1102983

在 MATLAB Online 中打开

In this case, you don't actually have a data dependency between the loop iterations. You can use a parfor reduction to do this, like so:

out1 = {};
out2 = {};
parfor i = 1:4
    for j = 1:3
        % Some calculation
        [tmp1, tmp2] = deal(i + j, i * j);
        % Build up out1 and out2 using concatenation
        out1 = [out1, tmp1];
        out2 = [out2, tmp2];
    end
end
Starting parallel pool (parpool) using the 'Processes' profile ...
Connected to the parallel pool (number of workers: 2).
celldisp(out1)
 
out1{1} =
 
     2     3     4     3     4     5     4     5     6     5     6     7

 
celldisp(out2)
 
out2{1} =
 
     1     2     3     2     4     6     3     6     9     4     8    12

 

There are additional restrictions you need to overcome - firstly, the range of your parfor loop must be consecutive integers. You could also combine the loops using ind2sub. Something a bit like this:

xVec = -2000:5:2000;
yVec = -500:5:500;
nX = numel(xVec);
nY = numel(yVec);
out1 = {};
out2 = {};
parfor idx = 1:(nX*nY)
    [i, j] = ind2sub([nX, nY], idx);
    xVal = xVec(i);
    yVal = yVec(j);
    % Some calculation
    [tmp1, tmp2] = deal(xVal + yVal, xVal * yVal);
    % Build up out1 and out2 using concatenation
    out1 = [out1, tmp1];
    out2 = [out2, tmp2];
end