[Openmp-dev] Error during build of CUDA offloading plugin / device RTL

[Lukas Sommer via Openmp-dev]

Hi everyone,

I just checked out branch release_70 of LLVM, Clang and OpenMP from git
and tried to build using cmake and make.

Unfortunately I experience an error during the build of the Nvidia CUDA
offloading device RTL:

> [ 10%] Building NVCC (Device) object
> projects/openmp/libomptarget/deviceRTLs/nvptx/CMakeFiles/omptarget-nvptx.dir/src/omptarget-nvptx_generated_cancel.cu.o
> /usr/include/c++/8/type_traits(1049): error: type name is not allowed
>
> /usr/include/c++/8/type_traits(1049): error: type name is not allowed
>
> /usr/include/c++/8/type_traits(1049): error: identifier
> "__is_assignable" is undefined
>
> 3 errors detected in the compilation of
> "/tmp/tmpxft_00005a09_00000000-6_cancel.cpp1.ii".
> CMake Error at omptarget-nvptx_generated_cancel.cu.o.Debug.cmake:275
> (message):
>   Error generating file
>  
> /scratch/llvm/llvm-latest/llvm-bin/projects/openmp/libomptarget/deviceRTLs/nvptx/CMakeFiles/omptarget-nvptx.dir/src/./omptarget-nvptx_generated_cancel.cu.o
I also see a similar error during the compilation of
omptarget-nvptx_generated_omp_data.cu.o.

I'm running Fedora 28 with kernel version 4.17.7, gcc version 8.1.1 and
nvcc version 9.1.85. 

My cmake-configuration is the default configuration, apart from
CMAKE_BUILD_TYPE=Debug, BUILD_SHARED_LIBS=ON and
LIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES=35,61 (6.1 is the capability of
my device).

Does anyone have some suggestions on how to fix this problem? Do I need
to further change my cmake configuration?

Thanks in advance,

Regards

Lukas