AMD ROCm Open GPU Compute Platform | Generated by AI

Home 2026.05

Here’s what this project is:

AMD ROCm (Radeon Open Compute) – the official open-source GPU compute platform

Cloned from: git@github.com:ROCm/ROCm.git

It is AMD’s counterpart to NVIDIA’s CUDA – an open-source software stack for GPU computation, targeting HPC, AI/ML, scientific computing, and CAD workloads.

Key components of this repo:

  1. Documentation (docs/) – The main ROCm documentation site (Sphinx-based), covering installation guides, API references, compatibility matrices, conceptual overviews, how-to guides, and release notes. This is actually the primary content of this repository.

  2. Build system (tools/rocm-build/) – 97 shell scripts and configs for building the various ROCm sub-projects from source. These include build scripts for:
    • Math libraries: rocBLAS, rocFFT, rocRAND, rocSOLVER, rocSPARSE, hipBLAS, etc.
    • ML/CV: MIOpen, MIVisionX, AMDMIGraphX
    • Communication: RCCL (collective communication)
    • Compilers/runtimes: HIP, LLVM, ROCclr, HSA runtime
    • Profiling: ROCm Profiler, Omniperf, Omnitrace
    • Docker build environments for Ubuntu 22/24

  3. Repo manifest (default.xml) – A repo-tool manifest that pins ~30+ sub-projects to rocm-7.2.0 tags (ROCK kernel driver, HIPIFY, LLVM, ROCgdb, composable kernels, etc.). This is how you’d repo sync the full source tree.

  4. CMakeLists.txt – A thin CMake wrapper (v5.7.1) that primarily builds the documentation.

  5. CHANGELOG.md (490KB) – Detailed release history across all ROCm versions.

In summary: This repo is the umbrella/manifest repository for the entire ROCm platform. It doesn’t contain the actual runtime or library source code (those live in separate repos like ROCm/rocm-systems, ROCm/llvm-project, etc.). Instead, it provides: the documentation, the build orchestration scripts, and the repo manifest to assemble all sub-projects into a coherent release.

Back Donate