Skip to content

Latest commit

 

History

History
24 lines (16 loc) · 1.6 KB

install-cuda.md

File metadata and controls

24 lines (16 loc) · 1.6 KB

AdaptiveCpp installation instructions for CUDA

If using clang

Please install CUDA 10.0 or later.

clang usually produces CUDA programs with very competitive performance compared to nvcc or nvc++. For more information on compiling CUDA with clang, please read the LLVM documentation on CUDA support. Note that the requirements on the CUDA installation described there.

If you use a very recent CUDA version, you might get a warning when compiling with AdaptiveCpp that clang does not support your CUDA version and treats like an older version. This warning can usually safely be ignored.

CMake variables:

  • -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda to point AdaptiveCpp to the CUDA root installation directory (e.g. /usr/local/cuda), if cmake doesn't find the right CUDA installation.
  • -DWITH_CUDA_BACKEND=ON if AdaptiveCpp does not automatically enable the CUDA backend

If using nvc++

Please install the latest release of the NVIDIA HPC SDK and make sure to point AdaptiveCpp to nvc++ (see below). Please install CUDA 10.0 or later. You can also rely on the CUDA bundled with the NVIDIA HPC SDK

CMake variables:

  • -DNVCXX_COMPILER=/path/to/nvc++
  • You can use the CUDA bundled with nvc++. Make sure to point AdaptiveCpp to the right CUDA installation using -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda.
  • -DWITH_CUDA_BACKEND=ON if AdaptiveCpp does not automatically enable the CUDA backend
  • -DWITH_CUDA_NVCXX_ONLY=ON enable if you want to use the CUDA backend exclusively with nvc++ and not clang. This will allow you to use nvc++ without having to install LLVM.