IFORT bug report: wrong result (optimisation problem?) when using -fPIC -qopenmp together

[raf3]

Hello,

I encountered a strange behavior when using -fPIC and -qopenmp together with ifort (ifort (IFORT) 2021.9.0 20230302 and ifort (IFORT) 2021.10.0 20230609 at least).

I have attached a minimal reproducible example to this message.

A .sh file is used to compile and execute the test with different compiler options:

#!/bin/bash

\rm -f *.o *.mod *.exe
ifort -O2 my_test.f90 -o test.exe
./test.exe

\rm -f *.o *.mod *.exe
ifort -O2 -fPIC my_test.f90 -o test.exe
./test.exe

\rm -f *.o *.mod *.exe
ifort -O2 -qopenmp my_test.f90 -o test.exe
./test.exe

\rm -f *.o *.mod *.exe
ifort -O2 -fPIC -qopenmp my_test.f90 -o test.exe
./test.exe

The expected result is 1.0, but when the two options are used together, the result is 5.0.

Adding a "magic print" works around the problem.

My questions are:

• Are you able to reproduce the behavior?
• Can you confirm that this is a real bug in the compiler (or have I missed something)?
• Is it possible to fix it in the next release?

Best regards,

Raphaël

[Mark_Lewy]

I note that this still outputs 5.0 with "ifort -O2 -fPIC -qopenmp my_test.f90", but 1.0 with ifx and the same options in the current OneAPI:  ifort 2021.12.0 20240211 / ifx 2024.1.2 20240508.  Given that ifort is deprecated, it's unlikely that Intel will fix this with ifort; you are better off switching to ifx, if possible.

[raf3]

Ok, thank you for this answer. I will look to switch my code to ifx.