CI (GitHub Actions) - Githubissues

e-kwsm commented 3 weeks ago

At the moment there are no CI in the repository. Matrix would be:

[ ] compiler: GCC, AOCC
[ ] precision: single, double, long double, quad
[ ] parallelization: sequential, thread, OpenMP
[ ] parallelization: sequential, MPI (OpenMPI, MPICH, etc)
[ ] SIMD: none, SSE, AVX
[ ] AMD-specific optimization

I am not sure about their conflicts, though.

jeroen-mostert commented 3 weeks ago

There would be no point to building without SIMD; this repo is a fork specifically to include optimizations for Zen CPUs, all of which support at least AVX2. If you wanted downlevel support, you'd just build vanilla FFTW. Even as a reference implementation, comparing against vanilla FFTW is probably saner. AVX-512 vs AVX2 is pretty much the only interesting difference, but then it depends on your CPU what you'll be building. If the dynamic dispatcher support is enabled, the proper instruction set will be selected at runtime, so you can basically always leave all SIMD options enabled (except for long double and quad, where there's no acceleration).

BiplabRaut commented 2 weeks ago

We are planning to add a CI pipeline in the near future. Until then a guide for contributors will be published for the pending patches.

amd / amd-fftw

CI (GitHub Actions) #15