This is a bit of a hack when your default gcc is greater than 6.0.0 and you are compiling arrayfire from source on Arch

arrayfire.md

So when building arrayfire on Arch you kind of run into the error below if the default gcc version is more than 6.0.0

Make Error: NVCC does not support GCC version 6.0 or greater.

So here is a hack that might help you fix that particular snag.

• Make sure you intall an alternate version of gcc that is < 6 but > 4.3 in this case gcc-5

• Make sure you are in the arrayfire directory

   cd path/to/arrayfire
  

• Modify the CMake configuration file

   nano src/backend/cuda/CMakeLists.txt
  

• Add this line

   CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
   
   FIND_PACKAGE(CUDA 7.0 REQUIRED)
   set(CMAKE_CXX_COMPILER_VERSION /usr/bin/gcc-5) # this one
   INCLUDE(CLKernelToH)
   INCLUDE(FindNVVM)
   
   OPTION(USE_LIBDEVICE "Use libdevice for CUDA JIT" ON)
   SET(CUDA_LIBDEVICE_DIR "${CUDA_NVVM_HOME}/libdevice" CACHE PATH "Path where lib$
   
   MARK_AS_ADVANCED(
       CUDA_BUILD_CUBIN
       CUDA_BUILD_EMULATION
       CUDA_SDK_ROOT_DIR
       CUDA_VERBOSE_BUILD)
  

• So basically what that does is point arrayfire to the right compatible compiler. You should be good to go my flags to cmake were

    cmake .. -DCMAKE_BUILD_TYPE=Release BUILD_CUDA=ON