| ROCm | |
|---|---|
 | |
| Developer | AMD |
| Initial release | November 14, 2016; 9 years ago (2016-11-14) |
| Stable release | |
| Repository | Meta-repository github |
| Written in | C,C++,Python,Fortran,Julia |
| Middleware | HIP |
| Engine | AMDgpukernel driver,HIPCC, aLLVM-basedcompiler |
| Operating system | Linux,Windows[2] |
| Platform | Supported GPUs |
| Predecessor | Close to metal,Stream,HSA |
| Size | <2GiB |
| Type | GPGPU libraries and APIs |
| License | MIT License |
| Website | www |
ROCm[3] is anAdvanced Micro Devices (AMD) software stack forgraphics processing unit (GPU) programming. ROCm spans several domains, includinggeneral-purpose computing on graphics processing units (GPGPU),high performance computing (HPC), andheterogeneous computing. It offers several programming models:HIP (GPU-kernel-based programming),OpenMP (directive-based programming), andOpenCL.
ROCm is free, libre andopen-source software (except the GPUfirmware blobs[4]), and it is distributed under various licenses. ROCm initially stood forRadeonOpenCompute platform; however, due toOpen Compute being a registered trademark, ROCm is no longer an acronym — it is simply AMD's open-source stack designed for GPU compute.
The first GPGPU software stack fromATI/AMD wasClose to Metal, which becameStream.
ROCm was launched around 2016[5] with theBoltzmann Initiative.[6] ROCm stack builds upon previous AMD GPU stacks; some tools trace back toGPUOpen and others to theHeterogeneous System Architecture (HSA).
HSAIL[7] was aimed at producing a middle-level, hardware-agnostic intermediate representation that could be JIT-compiled to the eventual hardware (GPU, FPGA...) using the appropriate finalizer. This approach was dropped for ROCm: now it builds only GPU code, usingLLVM, and itsAMDGPU backend that was upstreamed,[8] although there is still research on such enhanced modularity with LLVM MLIR.[9]
![[icon]](/mt/?noimg=&dark=on&url=http%3a%2f%2fen.wikipedia.org%2fwiki%2fFile%3aWiki_letter_w_cropped.svg) | This sectionneeds expansion. You can help byadding to it.(January 2022) |
ROCm as a stack ranges from the kernel driver to the end-user applications.AMD has introductory videos about AMD GCN hardware,[10] and ROCm programming[11] via its learning portal.[12]
One of the best technical introductions about the stack and ROCm/HIP programming, remains, to date, to be found on Reddit.[13]
ROCm is primarily targeted at discrete professional GPUs,[14] but consumer GPUs and APUs of the same architecture as a supported professional GPU are known to work with ROCm. For example, all professional GPUs of theRDNA 2 architecture are officially supported by ROCm 5.x; users report that Consumer RDNA2 units such as the Radeon 6800M APU and the Radeon 6700XT GPU also work.[15]
| Name ofGPU series | Southern Islands | Sea Islands | Volcanic Islands | Arctic Islands/Polaris | Vega | Navi 1X | Navi 2X | Navi 3X | Navi 4X |
|---|---|---|---|---|---|---|---|---|---|
| Released | Jan 2012 | Sep 2013 | Jun 2015 | Jun 2016 | Jun 2017 | Jul 2019 | Nov 2020 | Dec 2022 | Jan 2025 |
| Marketing Name | Radeon HD 7000 | Radeon Rx 200 | Radeon Rx 300 | Radeon RX 400/500 | Radeon RX Vega/Radeon VII(7 nm) | Radeon RX 5000 | Radeon RX 6000 | Radeon RX 7000 | Radeon RX 9000 |
| AMD support |  |  | |||||||
| Instruction set | GCN instruction set | RDNA instruction set | |||||||
| Microarchitecture | GCN 1st gen | GCN 2nd gen | GCN 3rd gen | GCN 4th gen | GCN 5th gen | RDNA | RDNA 2 | RDNA 3 | RDNA 4 |
| Type | Unified shader model | ||||||||
| ROCm | | needenv var in 5.x[16] | dropped in 6.x[17] | [a] | | ||||
| OpenCL | 1.2 (onLinux: 1.1 (no Image support) with Mesa 3D) | 2.0 (Adrenalin driver onWin7+) (onLinux: 1.1 (no Image support) with Mesa 3D, 2.0 with AMD drivers or AMD ROCm) | 2.0 | 2.1[20] | |||||
| Vulkan | 1.0 (Win 7+ orMesa 17+) | 1.2 (Adrenalin 20.1, Linux Mesa 3D 20.0) | 1.3 | ||||||
| Shader model | 5.1 | 5.1 6.3 | 6.4 | 6.5 | 6.7 | ||||
| OpenGL | 4.6 (on Linux: 4.6 (Mesa 3D 20.0)) | ||||||||
| Direct3D | 11 (11_1) 12 (11_1) | 11 (12_0) 12 (12_0) | 11 (12_1) 12 (12_1) | 11 (12_1) 12 (12_2) | |||||
/drm/amdgpu[b] | Experimental[21] | | |||||||
| This sectionneeds expansion. You can help byadding to it.(January 2022) |
AMD ROCm product manager Terry Deem gave a tour of the stack.[22]
The main consumers of the stack are machine learning and high-performance computing/GPGPU applications.
Variousdeep learning frameworks have a ROCm backend:[23]
ROCm is gaining significant traction in thetop 500.[25]ROCm is used with the Exascale supercomputersEl Capitan[26][27] andFrontier.
Some related software is to be found atAMD Infinity hub.
As of version 3.0,Blender can now use HIP compute kernels for itsrenderer cycles.[28]
Julia has the AMDGPU.jl package,[29] which integrates with LLVM and selects components of the ROCm stack. Instead of compiling code through HIP, AMDGPU.jl uses Julia's compiler to generate LLVM IR directly, which is later consumed by LLVM to generate native device code. AMDGPU.jl uses ROCr's HSA implementation to upload native code onto the device and execute it, similar to how HIP loads its own generated device code.
AMDGPU.jl also supports integration with ROCm's rocBLAS (for BLAS), rocRAND (for random number generation), and rocFFT (for FFTs). Future integration with rocALUTION, rocSOLVER, MIOpen, and certain other ROCm libraries is planned.
Installation instructions are provided for Linux and Windows in theofficial AMD ROCm documentation. ROCm software is currently spread across several publicGitHub repositories. Within the main publicmeta-repository, there is anXML manifest for each official release: usinggit-repo, aversion control tool built on top ofGit, is the recommended way to synchronize with the stack locally.[30]
AMD starts distributing containerized applications for ROCm, notably scientific research applications gathered underAMD Infinity Hub.[31]
AMDdistributes itself packages tailored to various Linux distributions.
There is a growingthird-party ecosystem packaging ROCm.
Linux distributions are officially packaging (natively) ROCm, with various degrees of advancement:Arch Linux,[32] Gentoo,[33]Debian,Fedora,[34]GNU Guix, andNixOS.
| This sectionneeds expansion. You can help byadding to it.(January 2022) |
There is one kernel-space component, ROCk, and the rest - there is roughly a hundred components in the stack - is made ofuser-space modules.
The unofficial typographic policy is to use: uppercase ROC lowercase following for low-level libraries, i.e. ROCt, and the contrary for user-facing libraries, i.e. rocBLAS.[36]
AMD is active developing with the LLVM community, but upstreaming is not instantaneous, and as of January 2022, is still lagging.[37] AMD still officially packages various LLVM forks[38][39][9] for parts that are not yet upstreamed – compiler optimizations destined to remain proprietary, debug support, OpenMP offloading, etc.
Support libraries implemented as LLVM bitcode. These provide various utilities and functions for math operations, atomics, queries for launch parameters, on-device kernel launch, etc.
Thethunk is responsible for all the thinking and queuing that goes into the stack.
TheROC runtime is a set of APIs/libraries that allows the launch of compute kernels by host applications. It is AMD's implementation of the HSA runtime API.[40] It is different from the ROC Common Language Runtime.
ROCm code object manager is in charge of interacting with LLVMintermediate representation.
Thecommon language runtime is an indirection layer adapting calls to ROCr on Linux and PAL on windows.It used to be able to route between different compilers, like the HSAIL-compiler. It is now being absorbed by the upper indirection layers (HIP and OpenCL).
ROCm ships its installable client driver (ICD) loader and an OpenCL[41]implementation bundled together.As of January 2022, ROCm 4.5.2 ships OpenCL 2.2, and is lagging behind competition.[42]
The AMD implementation for its GPUs is calledHIPAMD. There is also aCPU implementation mostly for demonstration purposes.
HIP builds a `HIPCC` compiler that either wrapsClang and compiles with LLVM open AMDGPU backend, or redirects to theNVIDIA compiler.[43]
HIPIFY is a source-to-source compiling tool. It translates CUDA to HIP and reverse, either using a Clang-based tool, or a sed-likePerl script.
Like HIPIFY,GPUFORT is a tool compiling source code into other third-generation-language sources, allowing users to migrate from CUDA Fortran to HIP Fortran. It is also in the repertoire of research projects, even more so.[44]
ROCm high-level libraries are usually consumed directly by application software, such asmachine learning frameworks. Most of the following libraries are in theGeneral Matrix Multiply (GEMM) category, which GPU architecture excels at.
The majority of these user-facing libraries comes in dual-form:hip for the indirection layer that can route to Nvidia hardware, androc for the AMD implementation.[45]
rocBLAS andhipBLAS are central in high-level libraries, it is the AMD implementation forBasic Linear Algebra Subprograms.It uses the libraryTensile privately.
This pair of libraries constitutes theLAPACK implementation for ROCm and is strongly coupled to rocBLAS.
ROCm competes with other GPU computing stacks: NvidiaCUDA andIntel OneAPI.
Nvidia's CUDA is closed-source, whereas AMD ROCm is open source. There is open-source software built on top of the closed-source CUDA, for instanceRAPIDS.
CUDA is able to run on consumer GPUs, whereas ROCm support is mostly offered for professional hardware such asAMD Instinct andAMD Radeon Pro.
Nvidia provides a C/C++-centered frontend and itsParallel Thread Execution (PTX) LLVM GPU backend as theNvidia CUDA Compiler (NVCC).
All the oneAPI corresponding libraries are published on itsGitHub Page.
Unified Acceleration Foundation (UXL) is a new technology consortium that are working on the continuation of the OneAPI initiative, with the goal to create a new open standard accelerator software ecosystem, related open standards and specification projects through Working Groups and Special Interest Groups (SIGs). The goal will compete with Nvidia's CUDA. The main companies behind it are Intel, Google, Arm, Qualcomm, Samsung, Imagination, and VMware.[46]
ROC_ENABLE_PRE_VEGA
AMD Instinct MI50, Radeon Pro VII, and Radeon VII products (collectively gfx906 GPUs) enters maintenance mode in ROCm 6.0. [...] end of maintenance [EOM] will be aligned with the closest ROCm release [...] Distribution and operating system updates will continue per the ROCm release cadence for gfx906 GPUs until EOM.
| General | |
|---|---|
| Levels | |
| Multithreading | |
| Theory | |
| Elements | |
| Coordination | |
| Programming | |
| Hardware | |
| APIs | |
| Problems | |
