parfor loop throws error for bigger matrices : Index must not exceed 0

2 次查看（过去 30 天）

Neelabh Jyoti Saharia 2023-7-28

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2001852-parfor-loop-throws-error-for-bigger-matrices-index-must-not-exceed-0

编辑： Neelabh Jyoti Saharia 2023-7-28

Hello, I am a newbie in parallel computing.

I am doing some calculation to genereate a 4D data. It worked fined while calculating a 4D matrix of small size(for e.g 2 × 3 × 4 × 100). But it then threw an error when the matrix sized was increased to 5 × 101 × 1001 × 101.

Below is my minimal code script.

'A' is the 4D matrix to be generated. The calculations necessary are done by my function 'bem4DParallel'.

So, at the innermost loop, a 1D array A_L is calculated. It is then stored in the 2D matrix A_K. This is done parallely for different values of k.

Then it is stored in 3D matix A_J and then finally in the 4D matrix A.

Computations for input_3 are parallelized. I didn't parallelize the innemost loop because to there is a dependency of the variable. That is, calculation of A_L(l) in the innermonst loop required the value of A_L(l-1). That's why I am passing the whole A_L array in the function while calculating.

% script name: main4DParallel.m
% inputs
input_1 = 0:5:20;
input_2 = 0:100;
input_3 = 500:1500;
input_4 = 0:0.01:2;
% 4D matrix to be calculated
A = zeros( length(input_1), length(input_2), length(input_3), length(input_4) ); 
for i = 1:length(input_1)
    
    arg_1 = input_1(i);
    A_J = zeros(length(input_2), length(input_3), length(input_4));
    
        for j = 1:length(input_2)
            
             arg_2 = input_2(j);
             A_K = zeros(length(input_3), length(input_4));
                     
             % parallel computing
             parfor k = 1:length(input_3)
                 
                 arg_3 = input_3(k);
                 A_L = zeros(1,length(input_4));
                 
                 for l = 1:length(input_4)                                                    
                     a = bem4DParallel(arg_1, arg_2, arg_3, input_4(l), A_L, l); 
                     A_L(l) = a; 
                 end
                     
                 A_K(k,:) = A_L;
                                         
             end
             A_J(j,:,:) = A_K;
                     
        end
        
        A(i,:,:,:) = A_J;
        
end

The script runs for some time(about 12 hours) and then gives the following error:

'main4DParallel' is the name of this script. And line 103 corresponds to the line where we have parfor k = 1:length(input_3)

After stopping, the values of the looping variables in the workspace are:

i = 1, j = 3. And there exists no k and l values.

Matrices:

A_L does not exist

2D matrix 'A_K' exist. And also data are present upto a row of 24. (Is it something to do with number of matlab workers?)

3D Matrix 'A_J' exist and contains value upto

A_J(2,:,:)

Make sense, because after stopping, j is at a value of 3

4D matrix A is completely empty.

Please let me know what I am doing wrong.

Thanks in advance

4 个评论
显示 2更早的评论隐藏 2更早的评论

Neelabh Jyoti Saharia 2023-7-28

编辑：Neelabh Jyoti Saharia 2023-7-28

在 MATLAB Online 中打开

If you are running the code, you can replace the funciton 'bem4DParallel' with some simple function like:

function value = doSuM(arg_1, arg_2, arg_3, input_4, Matrix, loopVar)
    m = Matrix(loopVar-1);
    value = arg_1 + arg_2 + arg_3 + input_4 + m + loopVar;  
end

and used it in the script in the innermost loop as:

a = doSum(arg_1, arg_2, arg_3, input_4(l), A_L, l)

Thanks :)

Harald 2023-7-28

Hi,

the next problem is that l does not exist in the function doSum. Let's assume that you want to use loopVar, so

m = Matrix(loopVar-1);

The next problem is that the loop over l starts at 1, meaning that you are falsely going to extract the 0-Element of Matrix. So, let me drop the "-1". Then, the provided example runs fine for me.

If that does not resolve the problem, I am afraid that it will be in bem4DParallel, and I recommend the following:

Before returning the output, verify that it has the expected data type and dimension (double and scalar, I suppose).
Wrap a try/catch - statement around the body of the function. In the catch portion, save the workspace of the function to a file. This will let you debug what the inputs were that led to the erroneous output, and thus fix the problem.

Best wishes,

Harald

请先登录，再进行评论。

请先登录，再回答此问题。