muDock — Molecular Docking Microapp

muDock is a compact, Autodock-style docking engine that uses a genetic algorithm and the Autodock 4.0 energy model. It was born as a benchmarking tool for Autodock-style workflows, and today it remains a small, focused codebase for experimenting with performance techniques (kernel porting, vectorization, accelerator backends, and approximation strategies) while staying usable as a docking tool.

The pipeline is intentionally split into clean stages (input parsing, scoring, search, and output), so individual pieces can be swapped or extended. That structure makes it practical to prototype new scoring functions, docking algorithms, or search strategies without rewriting the rest of the system.

Repository layout

• application — CLI entry point and executable sources
• mudock — core library and domain logic
• chem — chemical knowledge (JSON) used for code generation
• cmake — CMake helpers and dependency setup
• script — utility scripts (formatting, code generation, etc.)
• share — icons and non-code assets
• test — tests (disabled by default)

Dependencies

Required:

• CMake 3.25+
• A C++20-capable compiler (GCC/Clang/IntelLLVM)
• Boost (components: program_options, graph, fiber)
• OpenBabel3

Fetched automatically:

• Microsoft GSL (via CMake FetchContent)

Optional (enabled via build flags):

• OpenMP (CPU parallelism)
• CUDA Toolkit (with curand) for CUDA backend
• HIP + hiprand for HIP backend
• SYCL toolchain (oneAPI/LLVM)
• Google Highway (HWY) and/or xsimd for CPU vectorization
• LIKWID for profiling

Build and configuration

The project uses CMake and builds like a standard CMake package.

Basic build:

cmake -S /path/to/muDock -B /path/to/muDock/build -DCMAKE_INSTALL_PREFIX=/install/path
cmake --build /path/to/muDock/build

Key configuration options:

• CMAKE_BUILD_TYPE — Release (default), Debug, or RelWithDebInfo
• MUDOCK_ENABLE_FAST — enables aggressive optimizations (auto-ON for Release)
• MUDOCK_ENABLE_OMP — OpenMP CPU parallelism
• MUDOCK_ENABLE_CUDA — CUDA backend
• MUDOCK_ENABLE_HIP — HIP backend
• MUDOCK_ENABLE_SYCL — SYCL backend
• MUDOCK_ENABLE_GH — Google Highway vectorization
• MUDOCK_ENABLE_XSIMD — xsimd vectorization
• MUDOCK_ENABLE_LIKWID — LIKWID profiling
• MUDOCK_ENABLE_TEST — enable tests (not allowed in Release)

GPU/accelerator target configuration:

• MUDOCK_GPU_ARCHITECTURES — format platform:arch
  • Examples: cuda:sm_80, amd:gfx90a, intel:gen12
• MUDOCK_CPU_ARCHITECTURES, MUDOCK_CPU_TARGET, MUDOCK_CPU_TUNE — fine-tune CPU codegen

Example: CUDA build targeting SM80:

cmake -S /path/to/muDock -B /path/to/muDock/build \
  -DMUDOCK_ENABLE_CUDA=ON \
  -DMUDOCK_GPU_ARCHITECTURES=cuda:sm_80 \
  -DCMAKE_BUILD_TYPE=Release
cmake --build /path/to/muDock/build

Example: OpenMP CPU build:

cmake -S /path/to/muDock -B /path/to/muDock/build \
  -DMUDOCK_ENABLE_OMP=ON \
  -DCMAKE_BUILD_TYPE=Release
cmake --build /path/to/muDock/build

Running

Main application:

./build/application/muDock --protein /path/to/protein.pdb --ligand /path/to/ligands.mol2 --use CPP:CPU:0

The --use flag maps implementations to devices using IMPLEMENTATION:DEVICE:IDS (for example, CPP:CPU:0).

Converter:

./build/application/converter --input input.mol2 --output output.pdbqt

Supported formats (by file extension):

• mol2
• pdbqt
• pdb
• adtmol2

Note on adtmol2: the converter can produce adtmol2, and muDock can parse it for Autodock-like scoring with ligands parsed in parallel. Other formats are currently parsed sequentially.

Tests

Enable tests at configure time (Release is not allowed):

cmake -S /path/to/muDock -B /path/to/muDock/build -DMUDOCK_ENABLE_TEST=ON -DCMAKE_BUILD_TYPE=Debug
cmake --build /path/to/muDock/build
ctest --test-dir /path/to/muDock/build

References

• Gianmarco Accordi, Jens Domke, Theresa Pollinger, Davide Gadioli, Gianluca Palermo. "Towards High-Performance and Portable Molecular Docking on CPUs Through Vectorization." IEEE Cluster 2025. DOI: https://doi.org/10.1109/CLUSTER59342.2025.11186493