check non-Jenkins compilers work
Szilard has observed Apple Clang from XCode fail one of the C++11 tests (the library one I think). We ought to expect e.g. distro compiler maintainers to provide some way for a standard toolchain to get hooked up to an installed standard library. We should test it works out of the box, and/or fix and/or document.
Further thought: it's easy to accidentally name clang as the compiler, and even with
clang -std=c++11 you get mysterious looking errors that go away when you do
clang++ -std=c++11, so perhaps we should add a hint when that test fails that naming the compiler clang++ is necessary.
#2 Updated by Szilárd Páll over 2 years ago
Mark Abraham wrote:
Another thought - clang now supports OpenMP, but e.g. the package from the LLVM project running on bs_mac does not detect OpenMP support. Find out what is the issue here.
Perhaps it was build without OpenMP?
In my experience, I always had to set up clang with some external OpenMP lib (libomp or libgomp) and did not find any way to set a default such that I would not need to pass extra
-fopenmp=libomp to CMake nor set the
LD_LIBRARY_PATH every time when using a separate libomp.
#4 Updated by Mark Abraham about 2 years ago
- Status changed from New to Closed
- Assignee set to Mark Abraham
Post submit now tests linux clang with openmp. Since Apple XCode clang still doesn't support OpenMP, it's not yet an interesting performance target.
We have done some testing of cray wrapper compilers, and gcc for power 8 compilers, so I regard this as complete enough.