User error or bug in p?geevx ?

[MOZ]

Dear MKL Community,

I have encountered a problem while converting C code that uses LAPACK into code that will use ScaLAPACK. I need to find the right eigenvectors of a non-symmetric real matrix and both sets of eigenvectors (along with eigenvalues) of a non-symmetric, non-hermitian complex matrix.

It is tempting to jump to the conclusion that there is a bug in the relatively newly released p?geevx routines. But I realize that it probably is more likely that I misunderstood the documentation and am sending one or more incorrect parameters to routines or mis-initializing arrays.

I would greatly appreciate it if a more experienced user could take a look at my code to help me find the problem (or problems).

As part of the process of debugging, I stripped the code to more-or-less the bare minimum needed to produce the undesirable behavior. I am posting the real version as pdtest.c and the complex version as pztest.c. The compiler switches I used to compile the code are provided in comments at the beginning of these files.

I am compiling using the gnu C compiler on Ubuntu 16.04 LTS. I have tried both 2020 releases of Intel MKL. I'm using the 2019.7.217 version of the Intel MPI.

The pdtest.c code appears to work with matrix size=12. Example output is provided in the file output01.txt. However, with size=14, the application aborts with calls to MPI_Abort (as seen in output02.txt).

The pztest.c code appears to get stuck in an endless loop, never returning from the call to pzgeevx.  (By never I mean about an hour.) Example output is provided in output03.txt. It may be helpful to know that top shows 100% CPU usage by the processes (see top-output.txt).

Thank you in advance for your help with this problem!

Yours sincerely,

Matthew

[Khang_N_Intel]

Error confirmed with oneMKL 2021.2. Escalated!

[Khang_N_Intel]

Hi Mathew,

The problem was in the compiler.

This issue is fixed in the next release of oneMKL, 2021.4.

Best,

Khang

[MOZ]

Dear Khang,

This is good news. I'm looking forward to the release of 2021.4, then!

Yours sincerely,
Matthew

[Khang_N_Intel]

Hi Mathew,

oneMKL 2021.4 will be released sometime in September this year, unless there are some kinds of delay.

Khang

[Khang_N_Intel]

Hi Matthew,

oneMKL 2021.4.will be released tomorrow.

Please install the 2021.4 version when it becomes available and let me know if the issue is fixed from your side.

Best regards,

Khang

[Gennady_F_Intel]

Intel oneAPI Math Kernel Library (oneMKL) version 2021.4 is now available

[Khang_N_Intel · Accepted solution]

Hi Matthew,

We haven't heard from you for a while.

Since the solution has been provided in oneMKL 2021.4, I would like to close this thread.

There will be no more communication on this thread.

Best regards,

Khang

[Khang_N_Intel]

Hi Matthew,

I apologize for the previous post.

Could you install oneMKL 2021.4 and check to see if the issue is fixed?

Please let me know as soon as possible so that we can close this thread?

Thanks,

Khang

[Khang_N_Intel]

Hi Matthew,

I haven't heard from you for a while.

Since the solution has been provided, I would like to close this thread.

There will be no more communication on this thread.

Best regards,

Khang