ROCm/rocm-libraries

super repo for rocm libraries

What it solves

This repository acts as a "super-repo" that consolidates multiple ROCm-related libraries and shared components into a single location. It solves the problem of fragmented development by streamlining build and test workflows, unifying CI (Continuous Integration), and improving collaboration across different library teams.

How it works

The project organizes various ROCm libraries into two main categories: projects/ (libraries that are released as distinct packages, such as rocblas and miopen) and shared/ (components used as dependencies by multiple libraries, such as tensile and rocroller). It leverages a unified build system called "TheRock" to perform multi-component testing and integration.

Who it’s for

It is primarily for developers and contributors working on the ROCm ecosystem who need to build, test, and integrate high-performance math and compute libraries for AMD GPUs.

Highlights

  • Unified Workflow: Provides a single source of truth for a wide array of ROCm libraries, reducing the need to manage dozens of separate repositories.
  • Comprehensive Library Set: Includes critical components like hipblas, rocfft, miopen, and rocsolver.
  • Integrated CI: Uses TheRock CI for multi-component testing to ensure stability across the ecosystem.
  • Standardized Nomenclature: Standardizes project naming to match released packages, removing inconsistent casing and underscores.

Related

  • Project
    ROCm/rocm-systems

    A consolidated super-repository for ROCm systems projects that streamlines the development, CI, and integration of libraries and tools required for building PyTorch.

  • Project
    ROCm/legacy-rocm-build

    An open-source software stack of drivers, tools, and APIs that enables GPU programming and AI acceleration on AMD hardware.

  • Project
    ROCm/rccl

    A library of standard collective communication routines for GPUs that enables high-bandwidth data synchronization across multiple GPUs and nodes.

  • Project
    ROCm/rocm-examples

    A collection of examples and tutorials for the ROCm platform, providing code for HIP runtime functionality, GPU-accelerated applications, and AI model execution.

  • Project
    NVIDIA/cccl

    A unified collection of core CUDA C++ libraries (Thrust, CUB, and libcudacxx) providing high-performance parallel algorithms and standard library abstractions for GPU development.