Troubleshoot DPC++ targeting GPU with and without CUDA - nvidia

Question

I followed the instructions at the below website to test DPC++ on my computer.
https://github.com/intel/llvm/blob/sycl/sycl/doc/GetStartedGuide.md#build-dpc-toolchain
$ ./simple-sycle-app.exe
The results are correct!
I then modified line 23 to target my gpu
// Creating SYCL queue
sycl::queue Queue( sycl::gpu_selector{} );
[18:49]ec2-user$ ./simple-sycl-app-gpu.exe
terminate called after throwing an instance of 'cl::sycl::runtime_error' what(): No device of requested type available. -1 (PI_ERROR_DEVICE_NOT_FOUND) Aborted (core dumped)
did i need to build my dpc environment with CUDA enabled to access my K80 Nvidia GPU? I didn't enable it because CUDA support is apparently still experimental. I thought the whole point of using DPC++ was for me not to need to use CUDA. Should i be able to use my Nvidia gpu with just an nvidia driver and not CUDA?

Answer 1

You need to build llvm compiler with CUDA support. Brief instruction is to
install CUDA development kit, e.g. from here
Build and install Intel's llvm compiler from sources: https://github.com/intel/llvm passing --cuda param to configure command (see instructions on project's README)
use compiler according to this instruction, e.g. like this:
export DPCPP_HOME=~/sycl_workspace
export PATH=$DPCPP_HOME/llvm/build/bin:$PATH
export LD_LIBRARY_PATH=$DPCPP_HOME/llvm/build/lib:$LD_LIBRARY_PATH
clang++ -fsycl -fsycl-targets=nvptx64-nvidia-cuda \
simple-sycl-app.cpp -o simple-sycl-app-cuda.exe