MKL access violation, only occur when my problem scale goes up.

Zard — Mon, 02 Nov 2020 13:04:56 GMT

Hi everyone,

I'm coding for a scientific problem, when my time step and space step becomes small(like 1e-5 for time step), my code always fail. But not for bigger step size. I'm evaluating my algorithm's convergence rate and the result is OK . In my code, I only use fgmres and few sparse matrix formation functions. For fgmres module, I modified official tutorial case to suit my problem, most part is used directly(like RCI_Request branch routines). Hope someone can give me some advise. Thank you.

error occurs in this line:

mkl_sparse_d_trsv(transA, 1.0, csrL, descrL, &tmp[ipar[21] - 1], trvec);

Problem Detail: Exception raised at 0x7A062D3A (mkl_intel_thread.dll) (located in difflim1d1order.exe): 0xC0000005: An access violation occurred while reading location 0xCCCCCCF0.

Re: MKL access violation, only occur when my problem scale goes up.

Gennady_F_Intel — Mon, 02 Nov 2020 16:23:42 GMT

This is an unknown issue with the current version of MKL. You may give us the reproducer or submit the official report to the Intel Online Service Center.

topic Re: MKL access violation, only occur when my problem scale goes up. in Intel® oneAPI Math Kernel Library

MKL access violation, only occur when my problem scale goes up.

Re: MKL access violation, only occur when my problem scale goes up.