Subscription assignment dimension mismatch when simulations > a certain number of times in parallel computing

Hi everyone,

I've another question. I'm new to parallel computing. In fact, I've only started trying to use it today. I've the following in a parallel computing environment:

parfor z=1:16;

g=[1 2 3 4 no. of simulations];
[x(:,:,z),y(:,:,z),a(:,:,z),b(:,:,z)]=function(w,g,z);

end

When no. of simulations is set to 10, the parallel computing works perfectly. But, ideally, I want to run 1000 simulations in this parfor loop. However, when I tried anything above 10 loops, it stops halfway, and gives me an error of "Subscripted assignment dimension mismatch". The z number of cases are independent of each other.

Would anyone be able to point out what is possibly wrong with my code?

Would appreciate any help. Thanks :)

Answer 1

[Matt J]

Does it happen in a regular for-loop (not parfor)? If so, it should be pretty easy to trap with DBSTOP.

[Matt J]

In any case, the reason for the error is clear. At some point function(w,g,z) is returning something non-scalar. Why this happens for g>10 is buried inside the definition of the function.

[jj]

You are right. It does not work even with a normal for loop beyond g>10. I've tried dbstop if error. It doesn't seem to tell me the line the error has occurred, which is strange.