ROCm/legacy-rocm-build

AMD ROCm™ Software - GitHub Home

What is ROCm / legacy‑rocm‑build?

ROCm (Radeon Open Compute) is AMD’s open‑source software stack that lets you run GPU‑accelerated code on AMD GPUs. This repository contains the legacy build scripts and packaging for the ROCm stack – essentially the glue that assembles the drivers, runtimes, compilers, libraries and tools into a usable distribution.

Why it matters for AI / ML

  • ROCm provides the low‑level runtime (HIP) that lets developers write CUDA‑like code that runs on AMD hardware.
  • The stack includes math libraries (hipBLAS, rocBLAS, MIOpen, etc.) and profiling/debugging tools that are required for training and inference with frameworks such as PyTorch and TensorFlow.
  • By offering a full‑featured, open‑source alternative to NVIDIA’s CUDA ecosystem, ROCm enables AI researchers and engineers to leverage AMD GPUs for high‑performance computing, deep‑learning workloads, and scientific simulations.

Core pieces shipped via this repo

Category Example components (linked)
Math & compute libraries hipBLAS / rocBLAS, hipFFT / rocFFT, MIOpen, rocPRIM, etc.
Communication libraries RCCL, rocSHMEM
Runtimes & compilers HIP, HIPIFY, LLVM‑based ROCr runtime, SPIRV‑LLVM‑Translator
Profiling & debugging rocprofiler‑compute, ROCgdb, ROCdbgapi
Control & monitoring AMD SMI, ROCm Data Center Tool, rocminfo
Media & storage rocDecode, rocJPEG, hipFile
Extras ROCm Validation Suite, TransferBench, ROCm Optiq

Getting started

  1. Install ROCm – follow the official installation guide linked in the README (rocm.docs.amd.com/en/latest/install/rocm.html).
  2. Add AI frameworks – after the base stack is installed, install PyTorch or JAX for ROCm using the provided docs.
  3. Develop – write GPU kernels in HIP (C++ API) or use existing libraries; compile with the ROCm‑provided LLVM toolchain.
  4. Profile / debug – use rocprofiler-compute or ROCgdb to tune performance.

Current status

  • The repository is deprecated and will be superseded by the ROCm/TheRock monorepo. Existing issues and discussions are being migrated there.
  • All documentation, release notes, and contribution guidelines point to the newer TheRock location.

Who should use it?

  • Developers who need a stable, reproducible ROCm environment on supported Linux distributions (Ubuntu, RHEL, SLES, etc.).
  • Researchers building or porting AI/ML workloads to AMD GPUs.
  • Contributors who want to help maintain the legacy build system while the transition to TheRock is underway.

TL;DR – legacy-rocm-build is the packaging backbone of AMD’s open‑source GPU compute stack, providing the drivers, runtimes, math libraries and tooling that power AI training and inference on AMD hardware. It is being phased out in favor of the newer ROCm/TheRock repository, but still serves as the reference build for existing ROCm installations.

Related

  • Project
    ROCm/rocm-examples

    A collection of code examples and tutorials for ROCm and HIP, enabling users to accelerate applications and port CUDA code to AMD GPUs.

  • Project
    JuliaGPU/AMDGPU.jl

    A Julia package that enables programming and executing computations on AMD GPUs using the ROCm platform.

  • Project
    ROCm/MIOpen

    AMD's library for high-performance machine learning primitives, providing optimized kernels for deep learning on ROCm-enabled platforms.

  • Project
    ROCm/rocBLAS

    rocBLAS is a Basic Linear Algebra Subprograms library implemented in HIP and optimized for AMD GPUs to provide high-performance linear algebra operations.

  • Project
    ROCm/aiter

    A high-performance AI operator library for AMD ROCm that provides optimized GPU kernels for inference and training workloads across various AI frameworks.