How gams-dev/cuoptlink-builder works
Project type and purpose
• This is an open‐source “solver link” project whose sole purpose is to build, package, and distribute a GAMS/GAMSPy plugin that allows users to call NVIDIA’s cuOpt solver from within the GAMS modeling environment.
• Key indicators: a single C source file (gmscuopt.c) under the root, an assets folder containing configuration for GAMS, a GitHub Actions workflow, and examples demonstrating usage in a Jupyter notebook.
File structure
• Top-level artifacts:
• .github/workflows/main.yml → CI build & packaging pipeline
• gmscuopt.c → the core native code that implements the solver link API for GAMS
• assets/ → GAMS configuration files (gamsconfig.yaml, .def, .run) needed at runtime
• examples/ → demonstration code (Python/Jupyter) using GAMSPy + cuOpt
• No “frontend” or “backend”—this is a build‐and‐integration tool.
• Packaging relies on GitHub Actions to compile the C code, assemble assets, zip them into cuopt-link-release.zip, and tag a GitHub Release.
README for additional insights
• Describes OS and dependency requirements (Linux, GAMS ≥49, GAMSPy ≥1.12.1, CUDA ≥12.0, NVIDIA GPU).
• Spell out installation: unzip the released ZIP into the GAMS system directory (or GAMSPy base).
• Runtime test: model → GAMS “trnsport” example uses solver “cuopt.”
Elements
• Source → Build → Artifact → Unzip → GAMS System Dir → Runtime loads → cuOpt → GPU → Results → GAMS
• Blue for build-time components
• Green for runtime plugin components
• Orange for external dependencies (CUDA/cuOpt)
• Parallelogram for configuration files
Detail without over-splitting
• Call out the five specific CUDA libraries under assets to show exactly what runtime binaries are needed.