Benchmarks
optimization for benchmarking
If you like to run benchmarks you should set at least the following CMake variables.
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS="-ftree-vectorize -march=native" --build ../alpaka3
# with CUDA enabled
# -DCMAKE_CUDA_FLAGS="-ftree-vectorize -march=native"
# with HIP enabled
# -DCMAKE_HIP_FLAGS="-ftree-vectorize -march=native"
You can benchmark bableStream for different number of elements e.g. with a simple loop
for((i=1;i<10;++i)) ; do
./benchmark/babelstream/babelstream --array-size=$((33554432 * $i)) --number-runs=100
done