ABataev added a comment. Some of the previous comments are not fixed.
================ Comment at: clang/include/clang/Basic/DiagnosticSemaKinds.td:9100 +def note_omp_defaultmap_attr_none : Note< + "explicit data sharing attribute, data mapping attribute, or is_devise_ptr clause requested here">; def err_omp_wrong_dsa : Error< ---------------- is_device_ptr, not is_devise_ptr ================ Comment at: clang/lib/Sema/SemaOpenMP.cpp:1969-1972 + (DSAStack->getDefaultDMIBAtLevel(Level, OMPC_DEFAULTMAP_scalar) == + DMIB_alloc || + DSAStack->getDefaultDMIBAtLevel(Level, OMPC_DEFAULTMAP_scalar) == + DMIB_to || ---------------- cchen wrote: > ABataev wrote: > > I think, alloc and to scalars also must be captured by value. Moreover, > > seems to me, alloc scalars should not be captured at all since their > > behavior is very similar to the behavior of private variables. > Shouldn't defaultmap(alloc/to:scalar) have the same behavior (copy by ref) as > map(alloc/to) for scalar? Also, I don't think defaultmap(alloc:scalar) should > be the same as private (not copy at all) since you still need to register the > scalar address on the device. Ah, yes, I forgot that we actually allocate variables on the host. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D69204/new/ https://reviews.llvm.org/D69204 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
