topic Bug report: floating divide by zero error in dgeev and possibly also others in Intel® oneAPI Math Kernel Library

Bug report: floating divide by zero error in dgeev and possibly also others

yelh2015 — Thu, 08 May 2025 23:57:36 GMT

Hi friends,

I am using the OneAPI 2025.0 and 2015.1 HPC bundles to compile the following Fortran code:

solv.f:

program main
implicit none
integer(kind=4) :: info,lwork,i
real(kind=8), dimension(12) :: work
real(kind=8), dimension(3) :: er,ei
real(kind=8), dimension(3,3) :: mat,vl,vvr

lwork=12
mat=0.d0
do i=1,3
mat(i,i)=1.d0
enddo
call dgeev('N','V',3,mat,3,er,ei,vl,3,vvr,3,work,12,info)
end program main

My Makefile is:

solv:
ifx -O0 -g -fpe0 -gdwarf-2 -c solv.f
ifx -O0 -g -fpe0 -gdwarf-2 -o solv solv.o -lmkl_lapack95_lp64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential -lpthread
clean:
rm -f solv solv.o

Running the program produces 'floating divide by zero' error. The program runs well if compiled by OneAPI 2024.2, or gfortran, or other Intel compiler versions dated 2020,2018, etc. Also, the error does not show up if the code is compiled by other options like "-O2".

Similar error is also found for zhegvx (I do not have a testing program to reproduce it). The dependencies on the versions of the Intel compilers & MKL libraries, as well as the compiler options, are all the same as dgeev.

I am using Xubuntu 22.04 operating system.

I would be grateful if you can look into it. Thank you very much.

Dr. Lin-Hui Ye

Peking University.

Re:Bug report: floating divide by zero error in dgeev and possibly also others

Ruqiu_C_Intel — Fri, 16 May 2025 06:59:25 GMT

Hello Dr. Lin-Hui Ye,

Thank you for posting your issue.

Unfortunately, we didn't reproduce the issue with IFX 2025.1.1

rcao8@pvc01:~/issues/mkl$ ifx -O0 -g -fpe0 -gdwarf-2 -c solv.f

rcao8@pvc01:~/issues/mkl$ ifx -O0 -g -fpe0 -gdwarf-2 -o solv solv.o -lmkl_lapack95_lp64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential -lpthread

rcao8@pvc01:~/issues/mkl$ ./solv

rcao8@pvc01:~/issues/mkl$

rcao8@pvc01:~/issues/mkl$ ifx -v

ifx version 2025.1.1

Can you try again with ifx 2025.1.1?

Regards,

Ruqiu

Re:Bug report: floating divide by zero error in dgeev and possibly also others

Ruqiu_C_Intel — Thu, 22 May 2025 06:15:28 GMT

Thank you again for posting the issue.

We can run out error with oneMKL 2025.0 + ifx 2025.1.1. But it works well with oneMKL 2025.1 + ifx 2025.1.1. Please update to oneMKL 2025.1.