AMD ROCm Release Notes v2.10

This page describes the features, fixed issues, and information about downloading and installing the ROCm software. It also covers known issues and deprecated features in the ROCm v2.10 release.

What Is ROCm?

ROCm is designed to be a universal platform for gpu-accelerated computing. This modular design allows hardware vendors to build drivers that support the ROCm framework. ROCm is also designed to integrate multiple programming languages and makes it easy to add support for other languages.

ROCm is built from open source software. Subject to the applicable license, you can download the source code, modify and rebuild the ROCm components. To ensure you are downloading the correct source code versions, the ROCm repository provides a repo manifest file called default.xml.

Note: You can also clone the source code for individual ROCm components from the GitHub repositories.

ROCm Components

The following components for the ROCm platform are released and available for the v2.10 release: • Drivers • Tools • Libraries • Source Code

You can access the latest supported version of drivers, tools, libraries, and source code for the ROCm platform at the following location: https://github.com/RadeonOpenCompute/ROCm

Supported Operating Systems

The ROCm v2.10.x platform is designed to support the following operating systems:

• SLES 15 SP1

• Ubuntu 16.04.6(Kernel 4.15) and 18.04.3(Kernel 5.0)

• CentOS 7.6 (Using devtoolset-7 runtime support)

• RHEL 7.6 (Using devtoolset-7 runtime support)

For details about deploying the ROCm v2.10.x on these operating systems, see the Deploying ROCm section later in the document.

Important ROCm Links

Access the following links for more information on: • ROCm documentation, see https://rocm-documentation.readthedocs.io/en/latest/index.html

• ROCm binary structure, see https://github.com/RadeonOpenCompute/ROCm/blob/master/README.md#rocm-binary-package-structure

• Common ROCm installation issues, see https://rocm.github.io/install_issues.html

• Instructions to install PyTorch after ROCm is installed – https://rocm-documentation.readthedocs.io/en/latest/Deep_learning/Deep-learning.html#pytorch

Note: These instructions reference the rocm/pytorch:rocm2.9_ubuntu16.04_py2.7_pytorch image. However, you can substitute the Ubuntu 18.04 image listed at https://hub.docker.com/r/rocm/pytorch/tags

What’s New in Version 2.10

rocBLAS - Support for Complex GEMM in AMD Radeon Pro Vega 20

The rocBLAS library is a gpu-accelerated implementation of the standard Basic Linear Algebra Subroutines (BLAS). rocBLAS is designed to enable you to develop algorithms, including high performance computing, image analysis, and machine learning. In the AMD ROCm release v2.10, support is extended to the General Matrix Multiply (GEMM) routine for multiple small matrices processed simultaneously for rocBLAS in AMD Radeon Pro Vega 20. Both single and double precision, CGEMM and ZGEMM, are now supported in rocBLAS.

Support for SLES 15 SP1

In the AMD ROCm v2.10 release, support is added for SUSE Linux® Enterprise Server (SLES) 15 SP1. SLES is a modular operating system for both multimodal and traditional IT. Note: The SUSE Linux® Enterprise Server is a licensed platform. Ensure you have registered and have a license key prior to installation. Use the following SUSE command line to apply your license: SUSEConnect -r < Key>

SLES 15 SP1

The following section tells you how to perform an install and uninstall ROCm on SLES 15 SP 1. Run the following commands once for a fresh install on the operating system: sudo usermod -a -G video $LOGNAME sudo usermod -a -G sudo $LOGNAME sudo reboot

Code Marker Support for rocProfiler and rocTracer Libraries

Code markers provide the external correlation ID for the calling thread. This function indicates that the calling thread is entering and leaving an external API region. • The rocProfiler library enables you to profile performance counters and derived metrics. This library supports GFX8/GFX9 and provides a hardware-specific low-level performance analysis interface for profiling of GPU compute applications. The profiling includes hardware performance counters with complex performance metrics.

• The rocTracer library provides a specific runtime profiler to trace API and asynchronous activity. The API provides functionality for registering the runtimes API callbacks and the asynchronous activity records pool support.

• rocTX provides a C API for code markup for performance profiling and supports annotation of code ranges and ASCII markers.

Fixed Issues

Fixed Issues in the v2.10 Release

Running TensorFlow and PyTorch Frameworks Consecutively Results in the Memory Access Fault error

Issue: Running the TensorFlow and PyTorch frameworks in quick succession results in a Memory Access Fault error.

Resolution: This issue is resolved, and the error no longer appears.

Issue with the Docker Container Environment Variable Setting

Issue: Applications fail when the docker container is launched on the NUMA system without the ‘security-opt seccomp=unconfined’ setting.

Resolution: Configure the “–security-opt seccomp=unconfined” variable setting to avoid this issue.

Deprecated Features

The following features are deprecated in the AMD ROCm v2.10 release.

ROCm OpenCL Driver

The AMD ROCm-OpenCL-Driver is now deprecated. Users should migrate to ROCm-CompilerSupport, which provides more comprehensive functionality. The compiler support repository provides various lightning compiler-related services. It currently contains a single library, the Code Object Manager (Comgr) at lib/comgr. ROCm-OpenCL-Driver will no longer be actively maintained after the v2.10 release. If your application was developed with the ROCm-OpenCL-Driver, we would encourage you to switch to the ROCm-CompilerSupport repository.

Peer-to-Peer Bridge Driver for PeerDirect

The Peer-to-Peer bridge driver for the PeerDirect feature still works in the current release, however, it is now included as part of the ROCk kernel driver. ROCmRDMA allows third-party kernel drivers to utilize DMA access to the GPU memory. It allows a direct path for data exchange (peer-to-peer) using the standard features of PCI Express.

Currently, ROCmRDMA provides the following benefits:

• Direct access to ROCm memory for 3rd party PCIe devices

• Support for PeerDirect(c) interface to offloads the CPU when dealing with ROCm memory for RDMA network stacks

Deploying ROCm

AMD hosts both Debian and RPM repositories for the ROCm v2.10x packages.

The following directions show how to install ROCm on supported Debian-based systems such as Ubuntu 18.04.

Note: These directions may not work as written on unsupported Debian-based distributions. For example, newer versions of Ubuntu may not be compatible with the rock-dkms kernel driver. In this case, you can exclude the rocm-dkms and rock-dkms packages.

For more information on the ROCm binary structure, see https://github.com/RadeonOpenCompute/ROCm/blob/master/README.md#rocm-binary-package-structure

For information about upstream kernel drivers, see the Using Debian-based ROCm with Upstream Kernel Drivers section.

Ubuntu

Installing a ROCm Package from a Debian Repository

To install from a Debian Repository:

1. Run the following code to ensure that your system is up to date:

   sudo apt update
   
   sudo apt dist-upgrade
   
   sudo apt install libnuma-dev
   
   sudo reboot 
   

2. Add the ROCm apt repository.

   For Debian-based systems like Ubuntu, configure the Debian ROCm repository as follows:

   wget -q0 –
    http://repo.radeon.com/rocm/apt/debian/rocm.gpg.key | 
   
   sudo apt-key add -echo 'deb [arch=amd64] 
   http://repo.radeon.com/rocm/apt/debian/ xenial main' | 
   
   sudo tee /etc/apt/sources.list.d/rocm.list
   

The gpg key may change; ensure it is updated when installing a new release. If the key signature verification fails while updating, re-add the key from the ROCm apt repository.

The current rocm.gpg.key is not available in a standard key ring distribution, but has the following sha1sum hash:

e85a40d1a43453fe37d63aa6899bc96e08f2817a rocm.gpg.key

3. Install the ROCm meta-package. Update the appropriate repository list and install the rocm-dkms meta-package:

    sudo apt update
   
    sudo apt install rocm-dkms
   

4. Set permissions. To access the GPU, you must be a user in the video group. Ensure your user account is a member of the video group prior to using ROCm. To identify the groups you are a member of, use the following command:

    groups
   

5. To add your user to the video group, use the following command for the sudo password:

    sudo usermod -a -G video $LOGNAME
   

6. By default, add any future users to the video group. Run the following command to add users to the video group:

    echo 'ADD_EXTRA_GROUPS=1' | sudo tee -a /etc/adduser.conf
   
    echo 'EXTRA_GROUPS=video' | sudo tee -a /etc/adduser.conf
   

7. Restart the system.

8. Test the basic ROCm installation.

9. After restarting the system, run the following commands to verify that the ROCm installation is successful. If you see your GPUs listed by both commands, the installation is considered successful.

    /opt/rocm/bin/rocminfo
    /opt/rocm/opencl/bin/x86_64/clinfo
   

Note: To run the ROCm programs more efficiently, add the ROCm binaries in your PATH.

	echo 'export PATH=$PATH:/opt/rocm/bin:/opt/rocm/profiler/bin:/opt/rocm/opencl/bin/x86_64' | 
	sudo tee -a /etc/profile.d/rocm.sh

If you have an installation issue, refer the FAQ at: https://rocm.github.io/install_issues.html

Uninstalling ROCm Packages from Ubuntu

To uninstall the ROCm packages from Ubuntu 1v6.04 or Ubuntu v18.04, run the following command:

sudo apt autoremove rocm-dkms rocm-dev rocm-utils

Installing Development Packages for Cross Compilation

It is recommended that you develop and test development packages on different systems. For example, some development or build systems may not have an AMD GPU installed. In this scenario, you must avoid installing the ROCk kernel driver on the development system.

Instead, install the following development subset of packages:

sudo apt update
sudo apt install rocm-dev

Note: To execute ROCm enabled applications, you must install the full ROCm driver stack on your system.

Using Debian-based ROCm with Upstream Kernel Drivers

You can install the ROCm user-level software without installing the AMD's custom ROCk kernel driver. To use the upstream kernels, run the following commands instead of installing rocm-dkms:

sudo apt update	
sudo apt install rocm-dev	
echo 'SUBSYSTEM=="kfd", KERNEL=="kfd", TAG+="uaccess", GROUP="video"' | sudo tee /etc/udev/rules.d/70-kfd.rules

CentOS/RHEL v7 (v7.6) Support

This section describes how to install ROCm on supported RPM-based systems such as CentOS v7.6.

Note: The following instructions may not work on unsupported RPM-based distributions. For example, Fedora may not be compatible with the rock-dkms kernel driver. You can exclude the rocm-dkms and rock-dkms packages and use the upstream kernel driver instead.

Note: Although support for CentOS/RHEL v7 was added in ROCm v1.8, ROCm requires a special runtime environment provided by the RHEL Software Collections and additional dkms support packages to install and run correctly.

For more details, refer: https://github.com/RadeonOpenCompute/ROCm/blob/master/README.md#rocm-binary-package-structure

Preparing RHEL v7 (7.6) for Installation

RHEL is a subscription-based operating system. You must enable the external repositories to install on the devtoolset-7 environment and the dkms support files.

Note: The following steps do not apply to the CentOS installation.

1.The subscription for RHEL must be enabled and attached to a pool ID. See the Obtaining an RHEL image and license page for instructions on registering your system with the RHEL subscription server and attaching to a pool id.

2.Enable the following repositories:

sudo subscription-manager repos --enable rhel-server-rhscl-7-rpms		
sudo subscription-manager repos --enable rhel-7-server-optional-rpms	
sudo subscription-manager repos --enable rhel-7-server-extras-rpms

3.Enable additional repositories by downloading and installing the epel-release-latest-7 repository RPM:

sudo rpm -ivh 

For more details, see https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm

4.Install and set up Devtoolset-7.

To setup the Devtoolset-7 environment, follow the instructions on this page: https://www.softwarecollections.org/en/scls/rhscl/devtoolset-7/

Note: devtoolset-7 is a software collections package and is not supported by AMD.

Installing CentOS/RHEL (v7.6) for DKMS

Use the dkms tool to install the kernel drivers on CentOS/RHEL v7.6:

sudo yum install -y epel-release
sudo yum install -y dkms kernel-headers-`uname -r` kernel-devel-`uname -r`

ROCm Installation

Installing ROCm

To install ROCm on your system, follow the instructions below:

1.Delete the previous versions of ROCm before installing the latest version. 2.Create a /etc/yum.repos.d/rocm.repo file with the following contents:

[ROCm]
name=ROCm
baseurl=http://repo.radeon.com/rocm/yum/rpm
enabled=1
gpgcheck=0

Note: The URL of the repository must point to the location of the repositories’ repodata database.

3.Install ROCm components using the following command:

sudo yum install rocm-dkms

4.Restart the system. The rock-dkms component is installed and the /dev/kfd device is now available.

Machine Learning and High Performance Computing Software Stack for AMD GPU

ROCm Version 2.10

ROCm Binary Package Structure

• ROCm Binary Package Structure

ROCm Platform - Links to Drivers, Tools, Libraries and Source Code

The latest supported version of the drivers, tools, libraries and source code for the ROCm platform have been released and are available from the following GitHub repositories:

ROCm Core Components

• ROCk Kernel Driver
• ROCr Runtime
• ROCt Thunk Interface

ROCm Support Software

• ROCm SMI
• ROCm cmake
• rocminfo
• ROCm Bandwidth Test

ROCm Development Tools

• HCC compiler
• HIP
• ROCm Device Libraries
• ROCm OpenCL, which is created from the following components:
  • ROCm OpenCL Runtime
  • ROCm OpenCL Driver
• The ROCm OpenCL compiler, which is created from the following components:
  • ROCm LLVM OCL
  • ROCm LLVM HCC
  • ROCm Clang
  • ROCm lld OCL
  • ROCm lld HCC
  • ROCm Device Libraries
• ROCM Clang-OCL Kernel Compiler
• Asynchronous Task and Memory Interface (ATMI)
• ROCr Debug Agent
• ROCm Code Object Manager
• ROC Profiler
• ROC Tracer
• Radeon Compute Profiler
• Example Applications:
  • HCC Examples
  • HIP Examples

ROCm Libraries

• rocBLAS
• hipBLAS
• rocFFT
• rocRAND
• rocSPARSE
• hipSPARSE
• rocALUTION
• MIOpenGEMM
• MIOpen
• rocThrust
• ROCm SMI Lib
• RCCL
• MIVisionX
• hipCUB

ROCm Binary Package Structure

ROCm is a collection of software ranging from drivers and runtimes to libraries and developer tools. In AMD's package distributions, these software projects are provided as a separate packages. This allows users to install only the packages they need, if they do not wish to install all of ROCm. These packages will install most of the ROCm software into /opt/rocm/ by default.

The packages for each of the major ROCm components are:

• ROCm Core Components
  • ROCk Kernel Driver: rock-dkms
  • ROCr Runtime: hsa-rocr-dev, hsa-ext-rocr-dev
  • ROCt Thunk Interface: hsakmt-roct, hsakmt-roct-dev
• ROCm Support Software
  • ROCm SMI: rocm-smi
  • ROCm cmake: rocm-cmake
  • rocminfo: rocminfo
  • ROCm Bandwidth Test: rocm_bandwidth_test
• ROCm Development Tools
  • HCC compiler: hcc
  • HIP: hip_base, hip_doc, hip_hcc, hip_samples
  • ROCm Device Libraries: rocm-device-libs
  • ROCm OpenCL: rocm-opencl, rocm-opencl-devel (on RHEL/CentOS), rocm-opencl-dev (on Ubuntu)
  • ROCM Clang-OCL Kernel Compiler: rocm-clang-ocl
  • Asynchronous Task and Memory Interface (ATMI): atmi
  • ROCr Debug Agent: rocr_debug_agent
  • ROCm Code Object Manager: comgr
  • ROC Profiler: rocprofiler-dev
  • ROC Tracer: roctracer-dev
  • Radeon Compute Profiler: rocm-profiler
• ROCm Libraries
  • rocALUTION: rocalution
  • rocBLAS: rocblas
  • hipBLAS: hipblas
  • hipCUB: hipCUB
  • rocFFT: rocfft
  • rocRAND: rocrand
  • rocSPARSE: rocsparse
  • hipSPARSE: hipsparse
  • ROCm SMI Lib: rocm_smi_lib64
  • rocThrust: rocThrust
  • MIOpen: MIOpen-HIP (for the HIP version), MIOpen-OpenCL (for the OpenCL version)
  • MIOpenGEMM: miopengemm
  • MIVisionX: mivisionx
  • RCCL: rccl

To make it easier to install ROCm, the AMD binary repositories provide a number of meta-packages that will automatically install multiple other packages. For example, rocm-dkms is the primary meta-package that is used to install most of the base technology needed for ROCm to operate. It will install the rock-dkms kernel driver, and another meta-package (rocm-dev) which installs most of the user-land ROCm core components, support software, and development tools.

The rocm-utils meta-package will install useful utilities that, while not required for ROCm to operate, may still be beneficial to have. Finally, the rocm-libs meta-package will install some (but not all) of the libraries that are part of ROCm.

The chain of software installed by these meta-packages is illustrated below

  rocm-dkms
    |--rock-dkms
    \--rocm-dev
       |--hsa-rocr-dev
       |--hsa-ext-rocr-dev
       |--hsakmt-roct
       |--hsakmt-roct-dev
       |--rocm-cmake
       |--rocm-device-libs
       |--hcc
       |--hip_base
       |--hip_doc
       |--hip_hcc
       |--hip_samples
       |--rocm-smi
       |--hsa-amd-aqlprofile
       |--comgr
       |--rocr_debug_agent
       \--rocm-utils
          |--rocminfo
          \--rocm-clang-ocl # This will cause OpenCL to be installed

  rocm-libs
    |--rocalution
    |--hipblas
    |--rocblas
    |--rocfft
    |--rocrand
    |--hipsparse
    \--rocsparse

These meta-packages are not required but may be useful to make it easier to install ROCm on most systems.

Note:Some users may want to skip certain packages. For instance, a user that wants to use the upstream kernel drivers (rather than those supplied by AMD) may want to skip the rocm-dkms and rock-dkms packages, and instead directly install rocm-dev.

Similarly, a user that only wants to install OpenCL support instead of HCC and HIP may want to skip the rocm-dkms and rocm-dev packages. Instead, they could directly install rock-dkms, rocm-opencl, and rocm-opencl-dev and their dependencies.

Features and enhancements introduced in previous versions of ROCm can be found in version_history.md