Can anyone duplicate my matlab crash?

I am running a somewhat older version of matlab:

8.3.0.532 (R2014a)

When I run these commands, Matlab crashes:

a=zeros(19,1053);
tmp=min(a,[],2);

On the other hand, these commands do not cause a crash:

a=zeros(19,1052);
tmp=min(a,[],2);

In fact, there are a number of matrix sizes for "a" that can cause a crash, but for me, they all seem to have between 17 and 23 rows. In fact, the total number of elements in the matrix also has to be bigger than 20000.

I am running on:

Ubuntu 22.04.2 LTS

Intel(R) Xeon(R) Gold 6126 CPU @ 2.60GHz

I have tried this on two machines and it crashes on both.

The other machine is:

Ubuntu 22.04.1 LTS

Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz

Answer 1

[Steven Lord]

I'm not certain but it sounds like this may be Bug Report 1076354. One test you could do would be to transpose your array and take the max along dimension 1 as per the Workaround section of that bug report. If that doesn't crash that supports my hypothesis that it is that bug.

[Jed]

Thank you!

Transposing the matrix was, in fact, exactly the workaround I chose (see comment above in response to @John D'Errico). Based on the helpful link you provided I was also able to verify that using a single compute thread avoids the crash:

matlab -singleCompThread

So I think you are exactly right that I am seeing that bug you referenced.

Answer 2

[John D'Errico]

No problem encountered on two machines, but I don't have that release on my machine, and I think it is probably too old to run on my computer anyway. (I checked, and it appears I cannot even run that release on my computer. At least not without some work.)

It is entirely possible you have written a function called zeros, or perhaps a function named min. DON'T DO THAT! If not, then you might try contacting support. But a far better solution is to just upgrade your 9 year old release. This is not an existing MATLAB bug in current releases.

[Jed]

You are right that I have an old release. I also have 2016a and it does not crash, so you are also right that it seems to be specific to this release. The min and zeros functions are the builtin functions.

[Jed]

I have other (mex-related) constraints that make it so I need to be using the old release, so in my code, I decided to just use this instead as a workaround:

tmp = min(a')';

Thanks for checking on this... perhaps this post will be useful to someone else running an old matlab someday.