WARNING: SU2 was not compiled for an AVX-capable architecture.

[aditya12398]

I apologize in advanced if this issue seems to be trivial or stupid but I genuinely need help.

Describe the issue

Trying to run the Onera-M6 Newton-Krylov method when the solver starts with aforementioned warning.

I have a AMD Ryzen 5 3600 CPU which supports avx and avx2 architectures so this warning seems a little off somehow.

I did check out this CFD-Online Post. However, this page on Gentoo wiki suggests against using -march=native as a CXXFLAG for compiling.

gcc -c -Q -march=native --help=target shows -mavx and -mavx2 enabled but when I do gcc -c -Q -march=ryzen --help=target those two options are disabled. I can use x86-64-v3 instead which, according to this answer/script, is supported by my CPU but at this point I am just extremely confused as to what to do and how to avoid this warning.

Shall I recompile SU2 with this -march flag? If yes, which flag should I use out of these ryzen, x86-64, native, or x86-64-v3? Or is it something else altogether

Desktop information:

• OS: Arch Linux
• CPU: AMD Ryzen 5 3600 (12)
• C++ compiler and version: g++ (GCC) 12.2.0
• MPI implementation and version: MPICH 4.0.3
• SU2 Version: 7.4.0

Kindly let me know if you need any other details. Thanks!

[pcarruscag]

This is what I use to run locally
export CXXFLAGS="-march=native -funroll-loops -ffast-math -fno-finite-math-only" with --optimization=2

[aditya12398]

Oh, thanks a lot!

Will this option help get rid of the warning?

Also, what is the reason for using --optimization=2 instead of 3?