mkl_blas_dcopy seems to be in an infinite recursion stack

[carlos-carbonera]

Has anyone experience any random crash in mkl_blas_dcopy?

When loading the core file in totalview, the call to mkl_blas_dcopy seems to be in an infinite recursion stack.

The symptoms seem to indicate memory override of a sort. I would like to validate that there are no particular installation or caveats associate to the new MKL 10.5 64 bit libraries Linux RedHat 5.

[Vamsi_S_Intel]

Hi Carlos,

I do not encounter any random crash/error when using MKL dcopy() routine. I tested with the latest version of MKL on a Linux (Fedora 13) machine. I repeated my test 100 times, and I see no crashes.

It would be helpful if you can provide more information such as what architecture you are running on, the size/increment parameters passed to dcopy().

Also, MKL does nothave any special installation requirementsassociated with RedHat Linux 5.

[Gennady_F_Intel]

Carlos,10.5? may be you mean 10.0 Update 5?

Would you please give us the test case to check the problem with the current ( the latest version is 10.3 Update 4) version?

[mecej4]

I don't see why there is any need for recursion at all, in an array copy subroutine. Secondly, there is no evidence in your descriptions that stack overflow occurred, so the thread title is misleading.

It would not be hard to make mkl_blas_dcopy crash, and I could arrange for the crash to appear random. Debuggers can make information confusing at times.

Since the routine in question is fairly simple, first run in single-thread mode, and check the argument list for correct type and then for reasonable values of input arguments. Then, look for array overruns and undefined values being used. Finally, look for threading errors if your code and/or the libraries are multithreaded.