Matrix-free CG SYCL benchmarks

Building with hiPSYCL

Requires hipSYCL - https://github.com/illuhad/hipSYCL

CPUs

export HIPSYCL_TARGETS=omp
mkdir build
cd build
cmake -DSYCL_IMPL=hipSYCL -DSYCL_TARGET=cpu ..
make 

NVIDIA GPUS

export HIPSYCL_TARGETS=cuda:sm_XX
export CUDA_PATH=/path/to/cuda
mkdir build
cd build
cmake -DSYCL_IMPL=hipSYCL -DSYCL_TARGET=GPU ..
make

---

Building with Intel LLVM

Requires Intell LLVM/SYCL fork - https://github.com/intel/llvm .

Instructions to install Intel's LLVM-Based SYCL can be found here.

CPUs

export CXX=dpcpp
mkdir build_llvm
cd build_llvm
cmake -DSYCL_IMPL=LLVM -DSYCL_TARGET=CPU ..
make 

NVIDIA GPUS

export CUDA_PATH=/path/to/cuda
export CXX=clang++
mkdir build
cd build
cmake -DSYCL_IMPL=LLVM -DSYCL_TARGET=GPU ..
make

---

Running benchmarks

Running all benchmarks

Output results on screen:

./sycl

Save benchmarks to file (eg: output.txt):

./sycl --benchmark_format=csv --benchmark_out={output.txt}

Running a specific benchmark

Running axpby only:

./sycl --benchmark_filter=axpby --benchmark_format=csv --benchmark_out={output.txt}