Open psychocoderHPC opened 1 month ago
@psychocoderHPC Can you please split up the PR. At least in two PR: remove boost::atomic_ref
and the compiler detection.
At the moment it is nearly impossible to review it and I also saw, that you didn't modify the CI, which is definitively useful.
I think also a third PR with the demangling function names is useful.
Yes, I can do this after we discuss it in general.
Yes, I can do this after we discuss it in general.
You got me. I didn't read the description until the end :grimacing:
@psychocoderHPC Can you please short summarize the main changes?
If I'm not wrong the following boost features are used:
boost::atomic_ref
if std::atomic_ref
is not avaibleDid I missed something?
@psychocoderHPC Can you please short summarize the main changes?
If I'm not wrong the following boost features are used:
- preprocessor defines for compiler detection
boost::atomic_ref
ifstd::atomic_ref
is not avaible- demangle type names
Did I missed something?
no thats a good summery
fix https://github.com/alpaka-group/alpaka/issues/481
Make boost an optional dependency.
alpaka::core::demangled<TYPE>
returns a mangled name in case boost is not available. https://github.com/alpaka-group/alpaka/issues/2405 can solve this but should not be mixed into this PR. This PR removes the mixing of alpakas own defines e.gBOOST_LANG_HIP
which looked like boost but was exposed by alpaka under BOOST names.This PR is a base for discussion at the next developer meeting. In a separate PR we could reduce the number of tested boost versions. We should also run tests without boost at all.