Reference Number : DPD200335246
Product : Intel® Math Kernel Library (Intel® MKL)
Version : Intel MKL 11.0 Update 3, 11.1 beta
Operating System : Windows* OS, Linux * OS
Problem Description :
Intel MKL Customers using SVD (computes the singular value decomposition of a general rectangular matrix) routines can get wrong results or even the segmentation faults
in the case number of threads is 5,6,7,9,10,11, or larger not a multiple of 4.
The bug caused by the wrong splitting workload on BW functions across different cores.
The bug affects many functions from LAPACK. The list of these functions you can find into the “function.txt” file attached to this article.
The temporarily workaround: using the number of threads which is multiple of 4.
Particularly following number of threads are working well: 1, 2, 3, 4, 8, 12, 16, ..., 4*i, where i is an integer
Resolution Status:
The bug has been already fixed and the fix of the problem will be available the next update 11.0 update 4.