stp

Introduction

stp is a Semi-Tensor Product (STP) engine for Electronic Design Automation (EDA). The default implementation uses a compressed-vector representation for logic simulation.

Optional backends serve distinct purposes:

• Eigen: reference backend for validating the compressed-vector implementation.
• CUDA: accelerated backend for larger workloads.

Read the full documentation.

Build

The default build does not require Eigen or CUDA.

git clone https://github.com/nbulsi/stp.git
cd stp
cmake -S . -B build
cmake --build build -j
./build/bin/stp

Eigen reference backend

Enable Eigen only when building or validating the Eigen-based reference backend:

cmake -S . -B build/eigen -DSTP_ENABLE_EIGEN=ON
cmake --build build/eigen -j

If CMake cannot locate Eigen3, install it and provide its installation prefix:

cmake -S . -B build/eigen -DSTP_ENABLE_EIGEN=ON \
  -DCMAKE_PREFIX_PATH=/path/to/eigen-install-prefix

CUDA acceleration backend

Enable CUDA only when the host has an NVIDIA GPU, a compatible NVIDIA driver, and a CUDA Toolkit whose nvcc compiler is available on PATH. Verify the toolchain with nvcc --version.

cmake -S . -B build/cuda -DSTP_ENABLE_CUDA=ON
cmake --build build/cuda -j

The Eigen and CUDA options are independent and can be enabled together when needed.

Use as a command-line tool

Start the interactive shell:

./build/bin/stp

Simulate a LUT BENCH circuit. The default report prints each output truth table in hexadecimal and the simulation runtime:

sim -l yourcase.bench

Print the detailed input/output truth table as well:

sim -l --verbose yourcase.bench