[Openmp-dev] AMDGPU and clang-tools-extra

Jon Chesterfield via Openmp-dev openmp-dev at lists.llvm.org
Wed May 19 14:51:09 PDT 2021


Via enable projects will build openmp using the compiler you use for llvm.
Via enable runtimes will build clang, and then use that clang to build
openmp. You might want either.

Shilei was/is looking at making the compiler checking more precise than 'is
it called clang' which will convert "doesn't build" problems into "looked
like it worked but there is no openmp".

Thanks!


On Wed, 19 May 2021, 22:34 Itaru Kitayama, <itaru.kitayama at gmail.com> wrote:

> Jon, Ye,
> Is this
>
> -DLLVM_ENABLE_PROJECTS="clang" -DLLVM_ENABLE_RUNTIMES="openmp"
>
> correct? I used, and Spack as well, combined them (clang and openmp
> projects) together indeed in ENABLE_PROJECTS.
> On Thu, May 20, 2021 at 5:18 AM Ye Luo via Openmp-dev
> <openmp-dev at lists.llvm.org> wrote:
> >
> > spack install llvm at main does work on my workstation with ubuntu 20.04.
> rocm is installed and amd pieces for OpenMP offload are rebuilt properly. I
> have limited ideas about the experience on other OSes.
> >
> > yeluo at epyc-server:~/opt/llvm-clang/llvm-project$ ls
> /home/packages/spack/opt/spack/linux-ubuntu20.04-zen2/gcc-10.2.0/llvm-main-6n25gihrin6rc727fbothf44yq6zbp5a/lib/libomp
> > libomp.so                      libomptarget-amdgcn-gfx803.bc
> libomptarget.rtl.cuda.so
> > libomptarget-amdgcn-gfx700.bc  libomptarget-amdgcn-gfx900.bc
> libomptarget.rtl.x86_64.so
> > libomptarget-amdgcn-gfx701.bc  libomptarget-amdgcn-gfx906.bc
> libomptarget.so
> > libomptarget-amdgcn-gfx801.bc  libomptarget.rtl.amdgpu.so
> >
> > yeluo at epyc-server:~/opt/llvm-clang/llvm-project$ which clang-tidy
> >
> /home/packages/spack/opt/spack/linux-ubuntu20.04-zen2/gcc-10.2.0/llvm-main-6n25gihrin6rc727fbothf44yq6zbp5a/bin/clang-tidy
> >
> > spack uses ENABLE_PROJECTS and doesn't use ENABLE_RUNTIMES right now.
> > So the output from building clang-tools-extra and openmp can be blended.
> >
> > Best,
> > Ye
> > ===================
> > Ye Luo, Ph.D.
> > Computational Science Division & Leadership Computing Facility
> > Argonne National Laboratory
> >
> >
> > On Wed, May 19, 2021 at 2:22 PM Jon Chesterfield via Openmp-dev <
> openmp-dev at lists.llvm.org> wrote:
> >>>
> >>> Date: Wed, 19 May 2021 15:09:18 +0900
> >>> From: Itaru Kitayama via Openmp-dev <openmp-dev at lists.llvm.org>
> >>> To: openmp-dev <openmp-dev at lists.llvm.org>
> >>> Subject: [Openmp-dev] AMDGPU and clang-tools-extra
> >>> Message-ID:
> >>>         <CANW9uysd4uCQ7j3Jr1qmn2=RPo5a5OGU9e8-vPuuGv=
> P0nAXDQ at mail.gmail.com>
> >>> Content-Type: text/plain; charset="UTF-8"
> >>>
> >>> I am trying to build llvm on main branch with Spack with help from
> Tom, however
> >>> the AMDGPU code does not seem to work well with clang-tools-extra
> >>> project, which
> >>> is added together with the clang project when a Spack build initiated.
> >>> Has anyone
> >>> observed an issue with AMDGPU code with tools in upstream?
> >>
> >>
> >> None reported. I don't build clang-tools-extra, seems totally
> independent.
> >>
> >> Are you building openmp with ENABLE_PROJECTS and a clang that can't
> build amdgpu code, instead of ENABLE_RUNTIMES? If so, use ENABLE_RUNTIMES,
> a working clang, or disable the amdgpu library.
> >>
> >> Otherwise, please expand on 'does not seem to work well'.
> >>
> >> Jon
> >> _______________________________________________
> >> Openmp-dev mailing list
> >> Openmp-dev at lists.llvm.org
> >> https://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev
> >
> > _______________________________________________
> > Openmp-dev mailing list
> > Openmp-dev at lists.llvm.org
> > https://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/openmp-dev/attachments/20210519/6eebf717/attachment-0001.html>


More information about the Openmp-dev mailing list