PARDISO Crash

[Emil_J_]

PARDISO crashes with the following exception:

When the same system of equations is solved using an in-house solver, it returns a negative pivot error. We expect the same from PARDISO call. 

Test project is attached. It is a VS2019 project, using Intel® Fortran Compiler Classic 2021.10.0 [Intel(R) 64]. I understand this is the latest version, older versions have the same issue.

[mecej4]

What are the values of nn and na in your problem?

Those values be read from your data file, but that data file is big ( ~ 1.4 Gb), and a forum reader should not have to download, extract and read a file of that size just to know the values of nn and na.

[Emil_J_]

nn = 24341
na = 36555537
According to our in-house solver, the zero or negative pivot is in equation 24220. What could be causing PARDISO to crash instead of returning error=-4 (or some other error)? Is there any way to catch this type of issue before the crash? I have tried using phase=11, then phase=22, but the crash still occurs. 

[VarshaS_Intel]

Hi,

Thanks for posting in Intel Communities and Thanks for providing the details.

Could you please let us know if you are facing the same exception/issue with the small data file or only with the larger data file?

Thanks & Regards,

Varsha

[VarshaS_Intel]

Hi,

We have not heard back from you. Also, could you please provide us with the complete error you getting with the in-house solver?

Thanks & Regards,

Varsha

[Emil_J_]

Varsha,

I have not had a chance to try and reduce the size of the problem. I have seen the same error on another file of a similar scale, so I do not believe it is unique to this matrix. As stated above, our solver flags a zero or negative pivot error in equation 24220. I am not sure what else could be useful, but can give more info if you can tell me what you need exactly. 

The issue is not as much finding where the error is, but figuring out why the PARDISO call crashes (repeatable with the example I have provided) instead of exiting with an error. 

I should add that we have also submitted this issue as a support ticket to Intel, which might be more appropriate. 

[Emil_J_]

I have managed to create a smaller file, but it is still quite large. When I try to reduce it further, the crash disappears and the PARDISO call returns an error=-4 as expected. 

[VarshaS_Intel]

Hi,

Thanks for providing all the details and information.

When we tried with the file provided by you, we could still see the same exception as before even after replacing the file.

Could you please provide us with the complete sample reproducer project file you are using where you are getting pardiso error=-4?

Thanks & Regards,

Varsha

[Emil_J_]

Varsha,

Yes, this is correct. The second datafile I have posted is smaller, while still producing the same exception. I was not able to reduce the file more while still triggering it. 

Sorry, I'm not sure how attaching a file where we can see error=-4 would help us here. In cases when I see error=-4, the error is caught properly. The issue here is the crash, which as I understand is reproducible on your end too. Again, based on the output of our in-house solver, this is triggered by a zero or negative pivot. Is there a check or a setting I have missed that might help catch the crash?

[VarshaS_Intel]

Hi,

Thanks for reporting the issue.

We are able to reproduce your issue and we are discussing your issue internally with priority, we will get back to you soon.

Thanks & Regards,

Varsha

[VarshaS_Intel]

Hi,

Could you please check the value you have provided for array ja as the array ja ends with many zeros?

Thanks & Regards,

Varsha

[VarshaS_Intel]

Hi,

We have not heard back from you. Could you please provide us with an update on your issue?

Thanks & Regards,

Varsha

[VarshaS_Intel]

Hi,

We have not heard back from you. Could you please check the value you have provided for array ja as the array ja ends with many zeros in the input?

Thanks & Regards,

Varsha