Re: fixing AC_OPENMP for Apple clang
"Andrew W. Nosenko" <[email protected]> Tue, 3 Mar 2026 00:45:48 +0200
| Newsgroups | gmane.comp.sysutils.autoconf.general |
|---|---|
| Message-ID | <CALa-7vykPTgcefq=qkQUxv1xmieveR-5Tyy8Vc+tOe5dSs6pSw@mail.gmail.com> |
On Tue, 3 Mar 2026 at 00:39, Andrew W. Nosenko <[email protected]> wrote: > > On Mon, 2 Mar 2026 at 22:48, Dima Pasechnik <[email protected]> wrote: > >> On Mon, Mar 02, 2026 at 09:45:35PM +0200, Andrew W. Nosenko wrote: >> > >> > Does modern Xcode includes libomp =E2=80=9Cout of the box=E2=80=9D or = it need to be >> built >> > separately from Homebrew/MacPorts/Git? (I have access to Catalina bot >> not >> > to more modern version, and on Catalina it doesn=E2=80=99t) >> No, Xcode has no libomp. But Homebrew has a libomp package: >> https://formulae.brew.sh/formula/libomp >> >> It's also provided by R Project: >> https://mac.r-project.org/openmp/ >> >> And, yes, naturally, one can build libomp as a part of the llvm >> installation. >> > > > But then I=E2=80=99m confused again: if I need to build libomp separately= from > LLVM sources, then why I would to prefer XCode=E2=80=99s Clang instead of= Clang > from the same LLVM package, from which I=E2=80=99m going to build libomp = anyway? > OpenMP is about compiler=E2=80=99s #pragmas plus runtime support library. = So I see risks in de-synchronisation between frontend (compiler) and backend (library). But I see no benefits in using de-synchronised toolchain (Xcode clang + non-Xcode llvm libomp). >