Requires hipSYCL - https://github.com/illuhad/hipSYCL
export HIPSYCL_TARGETS=omp
mkdir build
cd build
cmake -DSYCL_IMPL=hipSYCL -DSYCL_TARGET=cpu ..
make
export HIPSYCL_TARGETS=cuda:sm_XX
export CUDA_PATH=/path/to/cuda
mkdir build
cd build
cmake -DSYCL_IMPL=hipSYCL -DSYCL_TARGET=GPU ..
make
Requires Intell LLVM/SYCL fork - https://github.com/intel/llvm .
Instructions to install Intel's LLVM-Based SYCL can be found here.
export CXX=dpcpp
mkdir build_llvm
cd build_llvm
cmake -DSYCL_IMPL=LLVM -DSYCL_TARGET=CPU ..
make
export CUDA_PATH=/path/to/cuda
export CXX=clang++
mkdir build
cd build
cmake -DSYCL_IMPL=LLVM -DSYCL_TARGET=GPU ..
make
Output results on screen:
./sycl
Save benchmarks to file (eg: output.txt
):
./sycl --benchmark_format=csv --benchmark_out={output.txt}
Running axpby only:
./sycl --benchmark_filter=axpby --benchmark_format=csv --benchmark_out={output.txt}