LibTorch should be built when you build Pytorch from source. If it is not, you can elso execute.
cd <pytorch_root>
# Make a new folder to build in to avoid polluting the source directories
mkdir build_libtorch && cd build_libtorch
# You might need to export some required environment variables here.
Normally setup.py sets good default env variables, but you'll have to do
that manually.
python ../tools/build_libtorch.py
Remember to delete your old build diectory, first.
./run_build.sh <Path to your Pytorch Root Directory>
mkdir build
cd build
cmake -DCMAKE_CXX_FLAGS="-I <Absolute Path to your Pytorch Root Directory>" -DCMAKE_PREFIX_PATH=<Absolute Path to your Pytorch Root Directory> ..
cmake --build . --config Release
The profiling scripts require the Python package Pandas.
pip install -r requirements.txt
Running profiling requires the use of sudo as nvprof requires root permissions for security. This can override your python path if you run from a virtual environment.
sudo <path to python>/python prof_codegen.py --block-x-start 10 --block-x-stop 320 --block-x-inc 10 --thread-x-start 128 --thread-x-stop 128 --thread-x-inc 128