ROCm/HIPIFY

What is HIPIFY?

HIPIFY is a collection of tools that automatically convert CUDA C++ source code into HIP (Heterogeneous‑Compute Interface for Portability) C++. By translating CUDA code to HIP, developers can more easily run their GPU‑accelerated applications on AMD GPUs as well as NVIDIA GPUs.

Key Capabilities

CUDA‑to‑HIP translation – The main tool, hipify-clang, parses CUDA source files and emits equivalent HIP code.
Documentation generation – hipify-clang can also produce Markdown or CSV documentation of the CUDA APIs that are supported by HIP, with different output formats (full, strict, compact).
Command‑line options – Use --help or --help‑hidden to list all flags; common flags include --md for Markdown output and --doc-format/--doc-roc to control the style of the generated docs.

How to Build the Docs Locally

cd docs
pip3 install -r .sphinx/requirements.txt
python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html

This produces a browsable HTML version of the HIPIFY documentation.

Generating the CUDA‑to‑HIP API Reference

# Full documentation (single combined file)
hipify-clang --md --doc-format=full --doc-roc=joint

# Or separate files per API group
hipify-clang --md --doc-format=full --doc-roc=separate

Replace --md with --csv to get CSV output, and you can switch --doc-format to strict or compact for a more concise listing.

Where to Find More Information

Official online docs: https://rocm.docs.amd.com/projects/HIPIFY/en/latest/index.html
Source code for the docs lives in the HIPIFY/docs directory of the repository.
Contribution guidelines for ROCm documentation are linked from the online docs.

Bottom line: HIPIFY helps developers port CUDA‑based GPU code to the portable HIP framework, easing cross‑vendor GPU support for AI, ML, and other high‑performance computing workloads.