MKL/Pardiso crash

[Peter_K_3]

We have been using the Intel MKL and its Pardiso solver for quite some time, and did suddenly run into a problem where the solver would crash randomly when trying to solve one particular problem.

We were able to reproduce this in a small test application where the same problem (a 2D Poisson problem actually) is solved over and over again, and after a few thousand iterations, the solver would crash at this location:

     MKLCrash.exe!mkl_pds_lp64_blkslvs1_omp_pardiso()  + 0x1b26 bytes    
     MKLCrash.exe!__kmp_invoke_microtask()  + 0x8c bytes    Asm
     MKLCrash.exe!__kmp_fork_call()  + 0x9e8 bytes    
     MKLCrash.exe!__kmpc_fork_call()  + 0x6c bytes    
     MKLCrash.exe!mkl_pds_lp64_blkslvs1_omp_pardiso()  + 0x71f bytes    
     MKLCrash.exe!mkl_pds_lp64_sssslv_pardiso()  + 0x117c bytes    
     MKLCrash.exe!mkl_pds_lp64_solve_pardiso()  + 0x422 bytes    
     MKLCrash.exe!mkl_pds_lp64_do_all_pardiso_fc()  + 0x15a2 bytes    
     MKLCrash.exe!mkl_pds_lp64_pardiso_c()  + 0x18cd bytes    
     MKLCrash.exe!mkl_pds_lp64_pardiso()  + 0x4a8 bytes    
     MKLCrash.exe!mkl_pds_lp64_dss_solve_real()  + 0x2bb bytes    
     MKLCrash.exe!dss_solve_real()  + 0x3f bytes    
     MKLCrash.exe!solver_crash_test()  Line 358    C++

We were abel to reproduce this on Windows (using MKL 10.3) as well as OS X (using MKL 11). The crash location seems to be the same on both platforms. I've attached the source code for our test application as well as crash logs.

Our test application is really quite minimal, so I suspect a concurrency bug in the solver. Can anyone confirm this?

Thanks,

Peter

[Alexander_K_Intel2]

Hi Peter, I will check your example. With best regards, Alexander Kalinkin

[Gennady_F_Intel]

Peter, I checked your code with 11.0 version, win7, lp64, threaded I couldn't reproduce the problem on my side. --Gennady

[Peter_K_3]

Thank you for having a look at this! Gennady, which exact version of the MKL for Windows were you using? I tried the following ones (all 64-bit): - 10.3 update 8, Win7, machine A --> crashes early (~ 20 iterations) - 10.3 update 3, Win7, machine B --> no crash after several thousand iterations - 11.0, Win7, machine B -> crashes early (~ 20 iterations) So on the same machine, it didn't crash with 10.3 update 3, but now crashes after upgrading to 11.0. Also, it only crashes with the release build, and when not attaching a debugger to the process. Thanks Peter

[Alexander_K_Intel2]

Hi Peter, I've reproduced your issue on Win7 only, on linux testcase work correctly. With best regards, Alexander Kalinkin

[Gennady_F_Intel]

Hello Peter, on my side the problem reproduced only when I launced the example from command line and it happened on #7 iteration. mkl 11.0, 64 bit, Win7. so, the problem is escalated and we will let you know if any update asap. --Gennady

[Peter_K_3]

Thanks a lot, would be great if this could be fixed. We currently cannot use the IntelMKL sparse solver in our application because of this bug. Peter

[Gennady_F_Intel]

Peter, please see my private email to you

[Jihuan_T_]

Gennady Fedorov (Intel) wrote:

Peter, please see my private email to you

Hello Gennady, I've also met with similar problems when I'm using Pardiso. When it runs after several iterations, there are completely no responses from the workstation. I'm using Intel Composer XE 2013 & VS2010. Would you please also tell me the possible causes you've sent to Peter? Thanks for your help!

[Gennady_F_Intel]

Peter, 

This issue has been fixed in MKL v.11.0 update 2 released yesterday. You can download this update from intel registration center and check the problem on your side.

regards, Gennady