Compiler features and run time support

[dbermond]

When at build time we enable a compiler feature like AVX-512 with the corresponding CMake option (-DOJPH_DISABLE_AVX512=OFF in this case, which is the default), will it detect at run time if the CPU supports AVX-512 and run fine on non-AVX-512 CPUs? Or will it cause run time problems in such non-AVX-512 CPUs?

It other words, is it safe to enable all compiler features (SSSE3, SSE4, AVX, AVX2 and AVX-512) in distributed builds?

By examining the source code, my understanding is that we can safely enable all compiler features in distributed builds, as it will detect at run time if a given feature is supported by the running CPU. But I'm asking just to make sure.

[aous72]

Hi Daniel,

Thank you for contacting me.

The code performs run-time detection of CPU features, and only uses code paths that are supported by the available features.
You can enable AVX-512 all the time, and the code should run fine on any machine.
The CMake configuration flags are mainly for testing and should a compiler not support them.

Feature detection code is in
https://github.com/aous72/OpenJPH/blob/master/src/core/others/ojph_arch.cpp

Kind regards,
Aous.