[llvm-dev] building with openmp target nvptx with llvm 12 versus 11

[KESTENER Pierre via llvm-dev]

Dear Johannes,

thanks for your answer. 
Yes gcc-multilib was necessary, silly of me !
The build is ok now.

Best regards.

________________________________________
De : Johannes Doerfert [johannesdoerfert at gmail.com]
Envoyé : vendredi 30 avril 2021 17:27
À : KESTENER Pierre; llvm-dev at lists.llvm.org; Luo, Ye
Objet : Re: [llvm-dev] building with openmp target nvptx with llvm 12 versus 11

This was an oversight and we'll make sure the runtime will be
freestanding in the future.
That said, Ye Luo fixed this by installing gcc-multilib, maybe that
helps you too.

If not, we'll have to come up with a hotfix.

~ Johannes


On 4/29/21 8:04 AM, KESTENER Pierre via llvm-dev wrote:
> Hello,
>
> I have succesfully built and used clang with OpenMP offload target for Nivida architecture (sm_80) using llvm-11.0.1 release sources.
>
> However, using release 12.0.0, I have a strange compile error.
>
> Here is how I configure the openmp support with cmake:
> ccmake  -DCLANG_OPENMP_NVPTX_DEFAULT_ARCH=sm_80 -DLIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES=80 -DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;libcxx;libcxxabi;lld;openmp" -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=/opt/local/llvm-12-omptarget -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ ../openmp
>
> here clang and clang++ are actually clang-12 and clang++-12 (built previously, I crossed check that I was able to compile cuda code, it is ok).
>
> The compile error in building openmp support is
>
> [ 24%] Building LLVM bitcode target_impl.cu-cuda_112-sm_80.bc
> cd /home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/build-openmp/libomptarget/deviceRTLs/nvptx && /opt/local/llvm-12-omptarget/bin/clang -S -x c++ -O1 -std=c++14 -target nvptx64 -Xclang -emit-llvm-bc -Xclang -aux-triple -Xclang x86_64-unknown-linux-gnu -fopenmp -fopenmp-cuda-mode -Xclang -fopenmp-is-device -D__CUDACC__ -I/home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/openmp/libomptarget/deviceRTLs -I/home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/openmp/libomptarget/deviceRTLs/nvptx/src -DOMPTARGET_NVPTX_DEBUG=0 -Xclang -target-cpu -Xclang sm_80 -D__CUDA_ARCH__=800 -Xclang -target-feature -Xclang +ptx70 -DCUDA_VERSION=11200 /home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/openmp/libomptarget/deviceRTLs/nvptx/src/target_impl.cu -o target_impl.cu-cuda_112-sm_80.bc
> In file included from /home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/openmp/libomptarget/deviceRTLs/nvptx/src/target_impl.cu:14:
> In file included from /home/kestener/install/llvm/llvm-project-llvmorg-12.0.0/openmp/libomptarget/deviceRTLs/nvptx/src/target_impl.h:15:
> In file included from /usr/include/assert.h:35:
> /usr/include/features.h:461:12: fatal error: 'sys/cdefs.h' file not found
> #  include <sys/cdefs.h>
>             ^~~~~~~~~~~~~
>
> I don't know why this header is not found, it is actually present in my system (Ubuntu 20.04)
>
> Did I miss something here ? It was building fine with llvm-11.0.1.
>
> Thanks for your help.
>
> Pierre
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev