Any known issues with dcopy ( MKL 11.2.4) in multithread environment?

AndrewC · ‎05-26-2016

My software seems to be 'randomly' deadlocking inside dcopy. I have lots of threads running. Is this a known issue? Running Intel Inspector XE always shows a litany of data race issues inside MKL but I have always been told 'don't worry, they are ok'.

     ntdll.dll!000000007750d3fa()
    [Frames below may be incorrect and/or missing, no symbols loaded for ntdll.dll]
    KernelBase.dll!000007fefd4110ac()
    libiomp5md.dll!000007fed0c747a0()
    libiomp5md.dll!000007fed0c08977()
    libiomp5md.dll!000007fed0c0ce40()
    libiomp5md.dll!000007fed0c3d7d1()
    libiomp5md.dll!000007fed0c0e736()
    mkl_intel_thread.dll!000007fea7a9baa5()
    mkl_intel_thread.dll!000007fea7a0b611()
     xxxxx.dll!dcopy() + 0x7a bytes

Ying_H_Intel · ‎05-29-2016

Hi vasci_i,

Could you build a small test case to reproduce the problem?

I tried one small case, (with MKL dcopy call in MKL example folder). the inspector reports "No problem detected".

Best Regards,
Ying

AndrewC · ‎05-29-2016

Intel inspector does not give specific warnings on DCOPY. I do find a number of warnings about data races on various LAPACK functions.

Anyway,this error occurred under very heavy threading conditions. Nested OMP threading , when called from separate threads. By reducing the # of threads the problem resolved itself. Upgrading to MKL 11.3.4 also corrected the issue.

Ying_H_Intel · ‎05-31-2016

Hi vasci_

Thanks for sharing the result.

There are some work between intel inspector and us to remove the MKL's warning in their report. So it was improved along later version.

Thanks

Ying

Here is MKL bug fix.

https://software.intel.com/en-us/articles/intel-mkl-113-bug-fixes-list