scott.linder added a comment. In D53153#1315109 <https://reviews.llvm.org/D53153#1315109>, @rjmccall wrote:
> Okay. So it's still the case that all symbols will be defined within the > linkage unit; it's just that some things might need to get exposed outside of > it. > > LLVM does provide a `dso_local` attribute which you could use unconditionally > on every symbol without actually changing visibility. It seems like `dso_local` is an IR concept, and is a guarantee of non-preemptability? Marking e.g. a global variable as `dso_local` in IR doesn't seem to affect what LLD infers about it;`computeIsPreemptible` in the ELF writer still believes it is preemptable, and so LLD complains when it attempts to generate a dynamic relocation in a read-only section (i.e. text). Is `dso_local` meant to convey anything to the linker? If this is the intended behavior we could still tell LLD to assume all defined symbols are not preemptable with something like `-Bsymbolic`. If we support dynamic linking/preemption in the future we will have to revisit this, but at that time visibility will be meaningful. I suppose one approach is then: - Remove the implicit `-fvisibility=hidden` in the AMDGPU Clang toolchain - Add a module pass to mark all global values with `dso_local` - Add an implicit `-Bsymbolic` to the linker commandline CHANGES SINCE LAST ACTION https://reviews.llvm.org/D53153/new/ https://reviews.llvm.org/D53153 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
