How-to guides

Short recipes for one task each. They assume you have built dynG (Install).

• Choose a backend
• Profile an update
• Read and write graph and batch files
• Use hypergraphs from Python
• Run the parity check
• Build the documentation
• Add an algorithm
• Port research code

Planned, written with the code they describe: streams and memory (with an RMM pool), reproduce a paper, use from PyTorch or CuPy (the CUDA plugins of 0.2; on the CPU, torch.from_dlpack of a result array works today, Python API), add an operator, add a backend, add a file format, benchmark and read the performance gates, and debug CUDA (sanitizers, NVTX). The Add an algorithm outline becomes a full guide later; the release process is Release process.