Interesting puzzle for some openmp guru

Maybe here we could build argument for switching from gcc to clang?

I will test it on clang soonish and ask if its fixable if it happens too. Maybe it should be fixed in R or gcc or intel mkl. Idk yet.

Would some openmp in clang guru wanna have a peek at this “issue” with openmp around r and intel mkl lib on gcc?

Best regards,

CC'ing the people that can probably tell you. My money is on the
interaction of the GNU/OpenMP compatibility layer in LLVM/OpenMP
and some calls getting resolved to the wrong place while others
go to "the right place".

I would assume simply telling gcc to use the LLVM/OpenMP runtime
might be enough here. Having two OpenMP runtimes used is probably
not a good idea even if it "works".

~ Johannes

Our MKL team informed me that we do document this here