ZGESVD returning wrong singular values when lwork is large

John_Young — Mon, 07 Mar 2016 14:51:22 GMT

Hi,

Attached is a test case showing the ZGESVD returning wrong singular values when lwork is large. The test matrix is an identity matrix so that all singular values should be 1. If I choose lwork to be the minimum required value specified by the documenation, then I see no issues. When I query ZGESVD for the optimum lwork using lwork = -1, then the singular values are no longer all 1. Furthermore, the efficiency of the ZGESVD routine drops drastically. For some values of lwork, the routine seems to run forever and I have to kill it.

I'm using MKL 11.3.1 Product Build 20151021 under linux with 8-byte integer default (ILP64 library). I have observed the same issue under windows as well. My compile line is

ifort -i8 -module . test.F90 -L$INTEL_MKL_LIBS -liomp5 -lmkl_intel_ilp64 -lmkl_core -lmkl_intel_thread -lpthread -lm

The output of the test routine is

OpenMP: 0
Intel(R) Math Kernel Library Version 11.3.1 Product Build 20151021 for Intel(R)
64 architecture applications

SUM: (5868.00000000000,0.000000000000000E+000)
Starting SVD TRY 1                  5868                  5868
LWORK :                 17604
Condition #:    1.00000000000000
Done with SVD TRY 1
Starting SVD TRY 2                  5868                  5868
LWORK :                645480
Condition #:    11.9444068944964
Done with SVD TRY 2
Starting SVD TRY 3                  5868                  5868
LWORK :                 58680
^Cforrtl: error (69): process interrupted (SIGINT)

Hi John,

Ying_H_Intel — Tue, 08 Mar 2016 02:47:54 GMT

Hi John,

Thank for let us know. We will look into it and get back to you if any news.

Thanks

Ying .

topic Hi John, in Intel® oneAPI Math Kernel Library

ZGESVD returning wrong singular values when lwork is large

Hi John,