Repository for compilation and cycle-accurate simulator for scale-out systolic arrays.
Execute the following from the root directory of the repository:
bash ./install.sh
This command automatically install CMake and Boost.
Use CMake to build (assuming in the repository root directory):
mkdir build-Debug # directory for out-of-source build
cd build-Debug
cmake .. -DCMAKE_PREFIX_PATH="$HOME/.local"
make
If you want a highly optimized release build, use instead:
mkdir build-Release # directory for out-of-source build
cd build-Release
cmake .. -DCMAKE_PREFIX_PATH="$HOME/.local" -DCMAKE_BUILD_TYPE=Release
make
You can use CMake and CMake Tools extensions of vscode to facilitate development. Let CMake Tools configure IntelliSense.
Precompile a Keras model
python precompiler/precompile.py # --help for information about optional arguments
Run binaries
./build-Release/compiler_st # --help for information about optional arguments
To reproduce the results in the original paper or to see example execution, check run_experiments.py
To use the C-simulator in Python projects, we provide a Python binder for the C binaries. Please check pybinder_tester.py for example usage.
After installation finishes, make sure that you restart vscode first by running "Remote-SSH: kill VS Code Server on Host...".
We provide settings.json
and launch.json
files to ease debugging and development. Please take a look at these files to have an idea what is going on under the hood.