ROCm 4.1.1 default.xml

add hipFFT to default.xml

README.md

@@ -1,41 +1,67 @@
-# AMD ROCm v3.5.1 Patch Release 
+# AMD ROCm™ v4.1.1 Patch Release Notes 
 
-AMD ROCm released a maintenance patch release v3.5.1. For more information about the release see, 
+The ROCm v4.1.1 release consists of the following updates:
 
-https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_ROCm_Release_Notes_v3.5.1.pdf
+* Changed Environment Variables for HIP
+
+* Updated HIP Instructions for ROCm Installation
+
+## Changed Environment Variables for HIP 
+
+In the ROCm v3.5 release, the Heterogeneous Compute Compiler (HCC) compiler was deprecated, and the HIP-Clang compiler was introduced for compiling Heterogeneous-Compute Interface for Portability (HIP) programs. In addition, the HIP runtime API was implemented on top of Radeon Open Compute Common Language Runtime (ROCclr). ROCclr is an abstraction layer that provides the ability to interact with different runtime backends such as ROCr. 
+
+While the *HIP_PLATFORM=hcc* environment variable was functional in subsequent releases, in the ROCm v4.1 release, the following environment variables were changed: 
+
+* *HIP_PLATFORM=hcc to HIP_PLATFORM=amd*
+
+* *HIP_PLATFORM=nvcc to HIP_PLATFORM=nvidia*
+
+Therefore, any applications continuing to use the *HIP_PLATFORM=hcc* variable will fail. You must update the environment variables to reflect the changes as mentioned above.
 
 
+## Updated HIP Instructions for ROCm Installation
 
-# AMD ROCm Release Notes v3.5.0
-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 v3.5.0 release.
+The hip-base package has a dependency on Perl modules that some operating systems may not have in their default package repositories.  Use the following commands to add repositories that have the required Perl packages:
+
+* For SLES 15 SP2
+
+		sudo zypper addrepo 
+
+  https://download.opensuse.org/repositories/devel:languages:perl/SLE_15/devel:languages:perl.repo
+
+* For CentOS8.3 
+
+		sudo yum config-manager --set-enabled powertools
+
+* For RHEL8.3
+
+		sudo subscription-manager repos --enable codeready-builder-for-rhel-8-x86_64-rpms
+
+
+# AMD ROCm™ v4.1 Release Notes 
+This document describes the features, fixed issues, and information about downloading and installing the AMD ROCm™ software. It also covers known issues and deprecations in the AMD ROCm v4.1 release.
 
 - [Supported Operating Systems and Documentation Updates](#Supported-Operating-Systems-and-Documentation-Updates)
   * [Supported Operating Systems](#Supported-Operating-Systems)
-  * [Documentation Updates](#Documentation-Updates)
+  * [ROCm Installation Updates](#ROCm-Installation-Updates)
+  * [AMD ROCm Documentation Updates](#AMD-ROCm-Documentation-Updates)
+
+- [Driver Compatibility Issue in This Release](#Driver-Compatibility-Issue-in-This-Release)
    
 - [What\'s New in This Release](#Whats-New-in-This-Release)
-  * [Upgrading to This Release](#Upgrading-to-This-Release)
-  * [Heterogeneous-Compute Interface for Portability](#Heterogeneous-Compute-Interface-for-Portability)
-  * [Radeon Open Compute Common Language Runtime](#Radeon-Open-Compute-Common-Language-Runtime)
-  * [OpenCL Runtime](#OpenCL-Runtime)
-  * [AMD ROCm GNU Debugger ROCgdb](#AMD-ROCm-GNU-Debugger-ROCgdb)
-  * [AMD ROCm Debugger API Library](#AMD-ROCm-Debugger-API-Library)
-  * [rocProfiler Dispatch Callbacks Start/Stop API](#rocProfiler-Dispatch-Callbacks-Start-Stop-API)
-  * [ROCm Communications Collective Library](#ROCm-Communications-Collective-Library)
-  * [NVIDIA Communications Collective Library Version Compatibility](#NVIDIA-Communications-Collective-Library-Version-Compatibility)
-  * [MIOpen Optional Kernel Package Installation](#MIOpen-Optional-Kernel-Package-Installation)
-  * [New SMI Event Interface and Library](#New-SMI-Event-Interface-and-Library)
-  * [API for CPU Affinity](#API-for-CPU-Affinity)
-  * [Radeon Performance Primitives Library](#Radeon-Performance-Primitives-Library)
-  
-  
-- [Fixed Issues](#Fixed-Issues)
+  * [TargetID for Multiple Configurations](#TargetID-for-Multiple-Configurations)
+  * [ROCm Data Center Tool](#ROCm-Data-Center-Tool)
+  * [ROCm Math and Communication Libraries](#ROCm-Math-and-Communication-Libraries)
+  * [HIP Enhancements](#HIP-Enhancements)
+  * [OpenMP Enhancements and Fixes](#OpenMP-Enhancements-and-Fixes)
+  * [MIOpen Tensile Integration](#MIOpen-Tensile-Integration)
 
+    
 - [Known Issues](#Known-Issues)
 
 - [Deprecations](#Deprecations)
-  * [Heterogeneous Compute Compiler](#Heterogeneous-Compute-Compiler)
+
+  * [Compiler Generated Code Object Version 2 Deprecation ](#Compiler-Generated-Code-Object-Version-2-Deprecation)
 
 - [Deploying ROCm](#Deploying-ROCm)
  
@@ -47,340 +73,563 @@ It also covers known issues and deprecated features in the ROCm v3.5.0 release.
   
 
 
-# Supported Operating Systems and Documentation Updates
 
-## Supported Operating Systems 
+## ROCm Installation Updates 
 
-The AMD ROCm v3.5.x platform is designed to support the following operating systems:
+### List of Supported Operating Systems
 
-* Ubuntu 16.04.6(Kernel 4.15) and 18.04.4(Kernel 5.3)
-* CentOS 7.7 (Kernel 3.10-1062) and RHEL 7.8(Kernel 3.10.0-1127)(Using devtoolset-7 runtime support)
-* SLES 15 SP1
-* CentOS and RHEL 8.1(Kernel 4.18.0-147)
+The AMD ROCm platform is designed to support the following operating systems:
+
+* Ubuntu 20.04.1 (5.4 and 5.6-oem) and 18.04.5 (Kernel 5.4)	
+* CentOS 7.9 (3.10.0-1127) & RHEL 7.9 (3.10.0-1160.6.1.el7) (Using devtoolset-7 runtime support)
+* CentOS 8.3 (4.18.0-193.el8) and RHEL 8.3 (4.18.0-193.1.1.el8) (devtoolset is not required)
+* SLES 15 SP2
+
+### Fresh Installation of AMD ROCM V4.1 Recommended
+
+A complete uninstallation of previous ROCm versions is required before installing a new version of ROCm. An upgrade from previous releases to AMD ROCm v4.1 is not supported. For more information, refer to the AMD ROCm Installation Guide at
+
+https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html
+
+**Note**: AMD ROCm release v3.3 or prior releases are not fully compatible with AMD ROCm v3.5 and higher versions. You must perform a fresh ROCm installation if you want to upgrade from AMD ROCm v3.3 or older to 3.5 or higher versions and vice-versa.
+
+**Note**: *render* group is required only for Ubuntu v20.04. For all other ROCm supported operating systems, continue to use video group. 
+
+* For ROCm v3.5 and releases thereafter, the clinfo path is changed to /opt/rocm/opencl/bin/clinfo. 
+
+* For ROCm v3.3 and older releases, the clinfo path remains /opt/rocm/opencl/bin/x86_64/clinfo. 
+ 
+### ROCm Multi-Version Installation Update
+
+With the AMD ROCm v4.1 release, the following ROCm multi-version installation changes apply:
+
+The meta packages rocm-dkms<version> are now deprecated for multi-version ROCm installs.  For example, rocm-dkms3.7.0, rocm-dkms3.8.0.
+ 
+* Multi-version installation of ROCm should be performed by installing rocm-dev<version> using each of the desired ROCm versions. For example, rocm-dev3.7.0, rocm-dev3.8.0, rocm-dev3.9.0.   
+
+* Version files must be created for each multi-version rocm <= 4.1.0
+
+ * Command: echo <version> | sudo tee /opt/rocm-<version>/.info/version
+
+ * Example: echo 4.1.0 | sudo tee /opt/rocm-4.1.0/.info/version
+
+* The rock-dkms loadable kernel modules should be installed using a single rock-dkms package. 
+
+* ROCm v3.9 and above will not set any ldconfig entries for ROCm libraries for multi-version installation.  Users must set LD_LIBRARY_PATH to load the ROCm library version of choice.
+
+**NOTE**: The single version installation of the ROCm stack remains the same. The rocm-dkms package can be used for single version installs and is not deprecated at this time.
 
 
-## Documentation Updates
+### Updated HIP Instructions for ROCm Installation
 
-### HIP-Clang Compile
+The hip-base package has a dependency on Perl modules that some operating systems may not have in their default package repositories.  Use the following commands to add repositories that have the required Perl packages:
 
-* [HIP FAQ - Transition from HCC to HIP-Clang](https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-FAQ.html#hip-faq)
-* [HIP-Clang Porting Guide](https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-porting-guide.html#hip-porting-guide)
-* [HIP - Glossary of Terms](https://rocmdocs.amd.com/en/latest/ROCm_Glossary/ROCm-Glossary.html)
 
-### AMD ROCDebugger (ROCgdb)
+#### For SLES 15 SP2
 
-* [ROCgdb User Guide](https://github.com/RadeonOpenCompute/ROCm/blob/master/gdb.pdf)
-* [ROCgdbapi Guide](https://github.com/RadeonOpenCompute/ROCm/blob/master/amd-dbgapi.pdf)
+		sudo zypper addrepo 
 
-### AMD ROCm Systems Management Interface
 
-* [System Management Interface Event API Guide](https://github.com/RadeonOpenCompute/ROCm/blob/master/ROCm_SMI_Manual.pdf)
+For more information, see
 
-### AMD ROCm Deep Learning
+https://download.opensuse.org/repositories/devel:languages:perl/SLE_15/devel:languages:perl.repo
 
-* [MIOpen API](https://github.com/ROCmSoftwarePlatform/MIOpen)
 
-### AMD ROCm Glossary of Terms
 
-* [Updated Glossary of Terms and Definitions](https://rocmdocs.amd.com/en/latest/ROCm_Glossary/ROCm-Glossary.html)
+#### For CentOS8.3
 
-### General AMD ROCm Documentatin Links
+		sudo yum config-manager --set-enabled powertools
+	
 
-Access the following links for more information on:
+#### For RHEL8.3
+
+		sudo subscription-manager repos --enable codeready-builder-for-rhel-8-x86_64-rpms
+ 
+
+# Driver Compatibility Issue in This Release
+
+In certain scenarios, the ROCm 4.1 run-time and userspace environment are not compatible with ROCm v4.0 and older driver implementations for 7nm-based (Vega 20) hardware (MI50 and MI60). 
+
+To mitigate issues, the ROCm v4.1 or newer userspace prevents running older drivers for these GPUs.
+
+Users are notified in the following scenarios:
+
+* Bare Metal 
+* Containers
+ 
+## Bare Metal
+
+In the bare-metal environment, the following error message displays in the console: 
+
+*“HSA Error: Incompatible kernel and userspace, Vega 20 disabled. Upgrade amdgpu.”*
+
+To test the compatibility, run the ROCm v4.1 version of rocminfo using the following instruction: 
+
+*/opt/rocm-4.1.0/bin/rocminfo 2>&1 | less*
+
+## Containers
+
+A container (built with error detection for this issue) using a ROCm v4.1 or newer run-time is initiated to execute on an older kernel. The container fails to start and the following warning appears:
+
+*Error: Incompatible ROCm environment. The Docker container requires the latest kernel driver to operate correctly.
+Upgrade the ROCm kernel to v4.1 or newer, or use a container tagged for v4.0.1 or older.*
+
+To inspect the version of the installed kernel driver,  run either: 
+
+* dpkg --status rock-dkms [Debian-based]
+
+or
+
+* rpm -ql rock-dkms [RHEL, SUSE, and others]
+
+To install or update the driver, follow the installation instructions at:
+
+https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html
+
+
+# AMD ROCm Documentation Updates
+
+## AMD ROCm Installation Guide 
+
+The AMD ROCm Installation Guide in this release includes:
+
+* Supported Environments
+
+* Installation Instructions for v4.1
+
+* HIP Installation Instructions 
+
+For more information, refer to the ROCm documentation website at:
+
+https://rocmdocs.amd.com/en/latest/
+
+
+## AMD ROCm - HIP Documentation Updates
+
+* HIP Programming Guide v4.1 
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_HIP_Programming_Guide_v4.1.pdf
+
+* HIP API Guide v4.1
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_HIP_API_Guide_v4.1.pdf
+  
+
+* HIP-Supported CUDA API Reference Guide v4.1
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/HIP_Supported_CUDA_API_Reference_Guide_v4.1.pdf
+
+* HIP FAQ  
+
+  For more information, refer to
+
+  https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-FAQ.html#hip-faq
+
+
+## ROCm Data Center User and API Guide
+
+* ROCm Data Center Tool User Guide
+
+   - Grafana Plugin Integration
+
+  For more information, refer to the ROCm Data Center User Guide at,
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_ROCm_DataCenter_Tool_User_Guide_v4.1.pdf
+  
+
+* ROCm Data Center Tool API Guide
+
+  For more information, refer to the ROCm Data Center API Guide at,
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/ROCm_Data_Center_Tool_API_Manual_4.1.pdf
+
+
+## ROCm SMI API Documentation Updates 
+
+* ROCm SMI API Guide 
+
+  For more information, refer to the ROCm SMI API Guide at,
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/ROCm_SMI_API_GUIDE_v4.1.pdf
+  
+## ROC Debugger User and API Guide 
+
+* ROC Debugger User Guide 
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/Debugging%20with%20ROCGDB%20User%20Guide%20v4.1.pdf
+  
+
+* Debugger API Guide 
+
+  https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD-Debugger%20API%20Guide%20v4.1.pdf
+
+
+## General AMD ROCm Documentation Links
+
+Access the following links for more information:
 
 * For AMD ROCm documentation, see 
 
   https://rocmdocs.amd.com/en/latest/
 
-* For installation instructions on supported platforms, see
+* For installation instructions on supped platforms, see
 
   https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html
 
 * For AMD ROCm binary structure, see
 
-  https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#machine-learning-and-high-performance-computing-software-stack-for-amd-gpu-v3-3-0
+  https://rocmdocs.amd.com/en/latest/Installation_Guide/Software-Stack-for-AMD-GPU.html
+  
 
 * For AMD ROCm Release History, see
 
-  https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#amd-rocm-version-history
+ https://rocmdocs.amd.com/en/latest/Current_Release_Notes/ROCm-Version-History.html
+
 
 
 # What\'s New in This Release
 
-## Upgrading to This Release
+## TargetID for Multiple Configurations
 
-You must perform a fresh and a clean AMD ROCm install to successfully upgrade from v3.3 to v3.5. The following changes apply in this release: 
+The new TargetID functionality allows compilations to specify various configurations of the supported hardware. 
 
-* HCC is deprecated and replaced with the HIP-Clang compiler
-* HIP-HCC runtime is changed to Radeon Open Compute Common Language Runtime (HIP-ROCClr)
-* In the v3.5 release, the firmware is separated from the kernel package. The difference is as    follows:
-     * v3.5 release has two separate rock-dkms and rock-dkms-firmware packages
-     * v3.3 release had the firmware as part of the rock-dkms package
+Previously, ROCm supported only a single configuration per target. 
+
+With the TargetID enhancement, ROCm supports configurations for Linux, PAL and associated configurations such as XNACK. This feature addresses configurations for the same target in different modes and allows applications to build executables that specify the supported configurations, including the option to be agnostic for the desired setting.
+
+
+### New Code Object Format Version for TargetID
+
+* A new clang option -mcode-object-version can be used to request the legacy code object version 3 or code object version 2. For more information, refer to
+
+  https://llvm.org/docs/AMDGPUUsage.html#elf-code-object
+
+* A new clang --offload-arch= option is introduced to specify the offload target architecture(s) for the HIP language.
+
+* The clang --offload-arch= and -mcpu options accept a new Target ID syntax. This allows both the processor and target feature settings to be specified. For more details, refer to
+
+  https://llvm.org/docs/AMDGPUUsage.html#amdgpu-target-id
+
+    - If a target feature is not specified, it defaults to a new concept of "any". The compiler, then, produces code, 
+      which executes on a target configured for either value of the setting impacting the overall performance. 
+      
+      It is recommended to explicitly specify the setting for more efficient performance. 
+
+    - In particular, the setting for XNACK now defaults to produce less performant code than previous ROCm releases.
+
+    - The legacy clang -mxnack, -mno-xnack, -msram-ecc, and -mno-sram-ecc options are deprecated. They are still 
+      supported, however, they will be removed in a future release. 
+
+    - The new Target ID syntax renames the SRAM ECC feature from sram-ecc to sramecc.
+
+* The clang offload bundler uses the new offload hipv4 for HIP code object version 4. For more information, see 
+https://clang.llvm.org/docs/ClangOffloadBundler.html
+
+* ROCm v4.1 corrects code object loading to enforce target feature settings of the code object to match the setting of the agent. It also corrects the recording of target feature settings in the code object. As a consequence, the legacy code objects may no longer load due to mismatches.
+
+* gfx802, gfx803, and gfx805 do not support the XNACK target feature in the ROCm v4.1 release.
+
+
+### New Code Object Tools
+
+AMD ROCm v4.1 provides new code object tools *roc-obj-ls* and *roc-obj-extract*. These tools allow for the listing and extraction of AMD GPU ROCm code objects that are embedded in HIP executables and shared objects. Each tool supports a --help option that provides more information. 
+
+Refer to the HIP Programming Guide v4.1 for additional information and examples.
+
+https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_HIP_Programming_Guide_v4.1.pdf
+
+**Note**
+
+The extractkernel tool in previous AMD ROCm releases has been removed from the AMD ROCm v4.1 release 
+and will no longer be  supported.
+
+**Note**
+ 
+ The roc-obj-ls and roc-obj-extract tools may generate an error about the following missing Perl modules: 
+
+* File::Which
+* File::BaseDir
+* File::Copy
+* URI::Encode
+
+This error is due to the missing dependencies in the hip-base installer package.  As a workaround, you may use the 
+following instructions to install the Perl modules:  
+
+*Ubuntu*
      
+     apt-get install libfile-which-perl libfile-basedir-perl libfile-copy-recursive-perl liburi-encode-perl
+   
+*CentOS*
+            
+    “sudo yum install perl-File-Which perl-File-BaseDir perl-File-Copy-Recursive perl-URI-Encode” 
+    
 
-## rocProf Command Line Tool Python Requirement
-SQLite3 is a required Python module for the rocprof command-line tool.  You can install the SQLite3 Python module using the pip utility and set env var ROCP_PYTHON_VERSION to the Python version, which includes the SQLite3 module.
+Repo for CentOS8.3: 
+
+    sudo yum config-manager --set-enabled powertools
+
+
+*RHEL*
+
+    sudo yum install perl-File-Which perl-File-BaseDir perl-File-Copy-Recursive perl-URI-Encode 
+    
+
+Repo for RHEL8.3: 
+
+    sudo subscription-manager repos --enable codeready-builder-for-rhel-8-x86_64-rpms
+
+
+*SLES 15 SP2*
+
+       sudo zypper addrepo https://download.opensuse.org/repositories/devel:languages:perl/SLE_15/devel:languages:perl.repo
+
+
+## ROCm Data Center Tool 
+
+### Grafana Integration
+
+The ROCm Data Center (RDC) Tool is enhanced with the Grafana plugin. Grafana is a common monitoring stack used for storing and visualizing time series data. Prometheus acts as the storage backend, and Grafana is used as the interface for analysis and visualization. Grafana has a plethora of visualization options and can be integrated with Prometheus for the ROCm Data Center (RDC) dashboard. 
+
+For more information about Grafana integration and installation, refer to the ROCm Data Center Tool User guide at:
+
+https://github.com/RadeonOpenCompute/ROCm/blob/master/AMD_ROCm_DataCenter_Tool_User_Guide_v4.1.pdf
 
 
 
-## Heterogeneous-Compute Interface for Portability
+## ROCm Math and Communication Libraries 
 
-In this release, the Heterogeneous Compute Compiler (HCC) compiler is deprecated and the HIP-Clang compiler is introduced for compiling Heterogeneous-Compute Interface for Portability (HIP) programs.
+### rocSPARSE
 
-NOTE: The HCC environment variables will be gradually deprecated in subsequent releases.
+rocSPARSE extends support for:
 
-The majority of the codebase for the HIP-Clang compiler has been upstreamed to the Clang trunk. The HIP-Clang implementation has undergone a strict code review by the LLVM/Clang community and comprehensive tests consisting of LLVM/Clang build bots. These reviews and tests resulted in higher productivity, code quality, and lower cost of maintenance.
+* gebsrmm
+* gebsrmv
+* gebsrsv
+* coo2dense and dense2coo
+* generic API including axpby, gather, scatter, rot, spvv, spmv, spgemm, sparsetodense, densetosparse
+* mixed indexing types in matrix formats
 
-![ScreenShot](HIPClang2.png)
+For more information, see 
 
-For most HIP applications, the transition from HCC to HIP-Clang is transparent and efficient as the HIPCC and HIP cmake files automatically choose compilation options for HIP-Clang and hide the difference between the HCC and HIP-Clang code. However, minor changes may be required as HIP-Clang has a stricter syntax and semantic checks compared to HCC.
-
-NOTE: Native HCC language features are no longer supported.
-
-## Radeon Open Compute Common Language Runtime
-In this release,  the HIP runtime API is implemented on top of Radeon Open Compute Common Language Runtime (ROCclr). ROCclr is an abstraction layer that provides the ability to interact with different runtime backends such as ROCr.
-
-## OpenCL Runtime
-The following OpenCL runtime changes are made in this release:
-
-* AMD ROCm OpenCL Runtime extends support to OpenCL2.2
-* The developer branch is changed from master to master-next
-
-## AMD ROCm GNU Debugger (ROCgdb)
-The AMD ROCm Debugger (ROCgdb) is the AMD ROCm source-level debugger for Linux based on the GNU Debugger (GDB). It enables heterogeneous debugging on the AMD ROCm platform of an x86-based host architecture along with AMD GPU architectures and supported by the AMD Debugger API Library (ROCdbgapi).
-
-The AMD ROCm Debugger is installed by the rocm-gdb package. The rocm-gdb package is part of the rocm-dev meta-package, which is in the rocm-dkms package.
-
-The current AMD ROCm Debugger (ROCgdb) is an initial prototype that focuses on source line debugging. Note, symbolic variable debugging capabilities are not currently supported.
-
-You can use the standard GDB commands for both CPU and GPU code debugging. For more information about ROCgdb, refer to the ROCgdb User Guide, which is installed at:
-
-* /opt/rocm/share/info/gdb.info as a texinfo file
-* /opt/rocm/share/doc/gdb/gdb.pdf as a PDF file
-
-The AMD ROCm Debugger User Guide is available as a PDF at:
-
-https://github.com/RadeonOpenCompute/ROCm/blob/master/gdb.pdf
- 
-For more information about GNU Debugger (GDB), refer to the GNU Debugger (GDB) web site at: http://www.gnu.org/software/gdb
+https://rocsparse.readthedocs.io/en/latest/
 
 
-## AMD ROCm Debugger API Library 
+### rocSOLVER
 
-The AMD ROCm Debugger API Library (ROCdbgapi) implements an AMD GPU debugger application programming interface (API) that provides the support necessary for a client of the library to control the execution and inspect the state of AMD GPU devices.
+rocSOLVER extends support for:
 
-The following AMD GPU architectures are supported:
-* Vega 10 
-* Vega 7nm
+* Eigensolver routines for symmetric/hermitian matrices:
+  - STERF, STEQR
+  
+* Linear solvers for general non-square systems:
+  - GELS (API added with batched and strided_batched versions. Only the overdetermined non-transpose case is implemented in 
+    this release. Other cases will return rocblas_status_not_implemented status for now.)
+    
+* Extended test coverage for functions returning information
 
-The AMD ROCm Debugger API Library is installed by the rocm-dbgapi package. The rocm-gdb package is part of the rocm-dev meta-package, which is in the rocm-dkms package.
-The AMD ROCm Debugger API Specification is available as a PDF at:
+* Changelog file
 
-https://github.com/RadeonOpenCompute/ROCm/blob/master/amd-dbgapi.pdf
+* Tridiagonalization routines for symmetric and hermitian matrices:
+  - LATRD
+  - SYTD2, SYTRD (with batched and strided_batched versions)
+  - HETD2, HETRD (with batched and strided_batched versions)
+  
+* Sample code and unit test for unified memory model/Heterogeneous Memory Management (HMM)
 
-## rocProfiler Dispatch Callbacks Start Stop API
+For more information, see
 
-In this release, a new rocprofiler start/stop API is added to enable/disable GPU kernel HSA dispatch callbacks. The callback can be registered with the 'rocprofiler_set_hsa_callbacks' API. The API helps you eliminate some profiling performance impact by invoking the profiler only for kernel dispatches of interest. This optimization will result in significant performance gains.
+https://rocsolver.readthedocs.io/en/latest/
 
-The API provides the following functions:
-* *hsa_status_t rocprofiler_start_queue_callbacks();* is used to start profiling
-* *hsa_status_t rocprofiler_stop_queue_callbacks();* is used to stop profiling. 
+### hipCUB
 
-For more information on kernel dispatches, see the HSA Platform System Architecture Specification guide at http://www.hsafoundation.com/standards/.
+The new iterator DiscardOutputIterator in hipCUB represents a special kind of pointer that ignores values written to it upon dereference.  It is useful for ignoring the output of certain algorithms without wasting memory capacity or bandwidth.  DiscardOutputIterator may also be used to count the size of an algorithm's output, which was not known previously.
 
-## ROCm Communications Collective Library 
-The ROCm Communications Collective Library (RCCL) consists of the following enhancements:
-* Re-enable target 0x803
-* Build time improvements for the HIP-Clang compiler
+For more information, see
 
-### NVIDIA Communications Collective Library Version Compatibility
-AMD RCCL is now compatible with NVIDIA Communications Collective Library (NCCL) v2.6.4 and provides the following features: 
-* Network interface improvements with API v3
-* Network topology detection 
-* Improved CPU type detection
-* Infiniband adaptive routing support
+https://hipcub.readthedocs.io/en/latest/
 
-## MIOpen Optional Kernel Package Installation
-MIOpen provides an optional pre-compiled kernel package to reduce startup latency. 
 
-NOTE: The installation of this package is optional. MIOpen will continue to function as expected even if you choose to not install the pre-compiled kernel package. This is because MIOpen compiles the kernels on the target machine once the kernel is run. However, the compilation step may significantly increase the startup time for different operations.
+## HIP Enhancements
 
-To install the kernel package for your GPU architecture, use the following command:
+### Support for hipEventDisableTiming Flag
 
-*apt-get install miopen-kernels-<arch>-<num cu>*
- 
-* <arch> is the GPU architecture. For example, gfx900, gfx906
-* <num cu> is the number of CUs available in the GPU. For example, 56 or 64 
- 
+HIP now supports the hipEventDisableTiming flag for hipEventCreateWithFlags. Note, events created with this flag do not record profiling data and provide optimal performance when used for synchronization.
 
-## New SMI Event Interface and Library
+### Cooperative Group Functions
 
-A Systems Management Interface (SMI) event interface is added to the kernel and ROCm SMI  lib for system administrators to get notified when specific events occur. On the kernel side, AMDKFD_IOC_SMI_EVENTS input/output control is enhanced to allow notifications propagation to user mode through the event channel. 
+Cooperative Groups defines, synchronizes, and communicates between groups of threads and blocks for efficiency and ease of management. HIP now supports the following kernel language Cooperative Groups types and functions:
 
-On the ROCm SMI lib side, APIs are added to set an event mask and receive event notifications with a timeout option. Further, ROCm SMI API details can be found in the PDF generated by Doxygen from source or by referring to the rocm_smi.h header file (see the rsmi_event_notification_* functions).
+![Screenshot](images/CGMain.PNG)
 
-For the more details about ROCm SMI API, see 
 
-https://github.com/RadeonOpenCompute/ROCm/blob/master/ROCm_SMI_Manual.pdf
 
-## API for CPU Affinity
-A new API is introduced for aiding applications to select the appropriate memory node for a given accelerator(GPU). 
+###  Support for Extern Shared Declarations
 
-The API for CPU affinity has the following signature:
+Previously, it was required to declare dynamic shared memory using the HIP_DYNAMIC_SHARED macro for accuracy as using static shared memory in the same kernel could result in overlapping memory ranges and data-races.
+Now, the HIP-Clang compiler provides support for extern shared declarations, and the HIP_DYNAMIC_SHARED option is no longer required. 
 
-*rsmi_status_t rsmi_topo_numa_affinity_get(uint32_t dv_ind, uint32_t *numa_node);*
+You may use the standard extern definition:
 
-This API takes as input, device index (dv_ind), and returns the NUMA node (CPU affinity), stored at the location pointed by numa_node pointer, associated with the device.
+    extern __shared__ type var[];
 
-Non-Uniform Memory Access (NUMA) is a computer memory design used in multiprocessing, where the memory access time depends on the memory location relative to the processor. 
 
-## Radeon Performance Primitives Library
-The new Radeon Performance Primitives (RPP) library is a comprehensive high-performance computer vision library for AMD (CPU and GPU) with the HIP and OpenCL backend. The target operating system is Linux.
+## OpenMP Enhancements and Fixes
 
-![ScreenShot](RPP.png)
+This release includes the following OpenMP changes:
 
-For more information about prerequisites and library functions, see 
+* Usability Enhancements
+* Fixes to Internal Clang Math Headers
+* OpenMP Defect Fixes
 
-https://github.com/GPUOpen-ProfessionalCompute-Libraries/MIVisionX/tree/master/docs
+### Usability Enhancements
 
-# Fixed Issues
+* OMPD updates for flang
+* To support OpenMP debugging, the selected OpenMP runtime sources are included in lib-debug/src/openmp. The ROCgdb debugger 
+  will find these automatically.
+* Threadsafe hsa plugin for libomptarget
+* Support multiple devices with malloc and hostrpc
+* Improve hostrpc version check
+* Add max reduction offload feature to flang
+* Integration of changes to support HPC Toolkit
+* Support for fprintf
+* Initial support for GPU malloc and Free. The internal (device rtl) is required for GPU malloc and Free for nested parallelism.  
+  GPU malloc and Free are now replaced, which improves the device memory footprint.
+* Increase detail of debug printing controlled by LIBOMPTARGET_KERNEL_TRACE environment variable
+* Add support for -gpubnames in Flang Driver
+* Increase detail of debug printing controlled by LIBOMPTARGET_KERNEL_TRACE environment variable
+* Add support for -gpubnames in Flang Driver
 
-## Device printf Support for HIP-Clang
-HIP now supports the use of printf in the device code. The parameters and return value for the device-side printf follow the POSIX.1 standard, with the exception that the "%n" specifier is not supported. A call to printf blocks the calling wavefront until the operation is completely processed by the host. 
+###  Fixes to Internal Clang Math Headers
 
-No host-side runtime calls by the application are needed to cause the output to appear. There is also no limit on the number of device-side calls to printf or the amount of data that is printed.
+This release includes a set of changes applied to Clang internal headers to support OpenMP C, C++, FORTRAN, and HIP C. This establishes consistency between NVPTX and AMDGCN offloading, and OpenMP, HIP, and CUDA. OpenMP uses function variants and header overlays to define device versions of functions. This causes Clang LLVM IR codegen to mangle names of variants in both the definition and callsites of functions defined in the internal Clang headers. The changes apply to headers found in the installation subdirectory lib/clang/11.0.0/include.
 
-For more details, refer the HIP Programming Guide at:
-https://rocmdocs.amd.com/en/latest/Programming_Guides/HIP-GUIDE.html#hip-guide
+The changes also temporarily eliminate the use of the libm bitcode libraries for C and C++. Although math functions are now defined with internal clang headers, a bitcode library of the C functions defined in the headers is still built for the FORTRAN toolchain linking. This is because FORTRAN cannot use C math headers. This bitcode library is installed in lib/libdevice/libm-.bc. The source build of the bitcode library is implemented with the aomp-extras repository and the component-built script build_extras.sh. 
 
-## Assertions in HIP Device Code  
-Previously, a failing assertion caused early termination of kernels and the application to exit with a line number, file, and failing condition printed to the screen.
-This issue is now fixed and the assert() and abort() functions are implemented for HIP device code. 
-NOTE: There may be a performance impact in the use of device assertions in its current form. 
+### OpenMP Defect Fixes
 
-You may choose to disable the assertion in the production code. For example, to disable an assertion of:
+The following OpenMP defects are fixed in this release:
 
-*assert(foo != 0);*    
+* Openmpi configuration issue with real16. 
+* [flang] The AOMP 11.7-1 Fortran compiler claims to support the -isystem flag, but ignores it.
+* [flang] producing internal compiler error when the character is used with KIND.
+* [flang] openmp map clause on complex allocatable expressions !$omp target data map( chunk%tiles(1)%field%density0).
+* Add a fatal error if missing -Xopenmp-target or -march options when -fopenmp-targets is specified. However, this requirement is not 
+  applicable for offloading to the host when there is only a single target and that target is the host.
+* Openmp error message output for no_rocm_device_lib was asserting.
+* Linkage on constant per-kernel symbols from external to weaklinkageonly to prevent duplicate symbols when building kokkos.
+* Add environment variables ROCM_LLD_ARGS ROCM_LINK_ARGS ROCM_SELECT_ARGS to test driver options without compiler rebuild.  
+* Fix problems with device math functions being ambiguous, especially the pow function.ix aompcc to accept file type cxx. 
+* Fix a latent race between host runtime and devicertl.
 
-you may comment it out as:  
+## MIOpen Tensile Integration
 
-*//assert(foo != 0);*
+MIOpenTensile provides host-callable interfaces to the Tensile library and supports the HIP programming model. You may use the Tensile feature in the HIP backend by setting the building environment variable value to ON.
+
+    MIOPEN_USE_MIOPENTENSILE=ON
+
+MIOpenTensile is an open-source collaboration tool where external entities can submit source pull requests (PRs) for updates. MIOpenTensile maintainers review and approve the PRs using standard open-source practices. 
+
+For more information about the sources and the build system, see
+
+https://github.com/ROCmSoftwarePlatform/MIOpenTensile
 
-NOTE: Assertions are currently enabled by default. 
 
 # Known Issues 
-The following are the known issues in the v3.5 release.
 
-## HIPify-Clang Installation Failure on CentOS/RHEL 
+The following are the known issues in this release.
 
-HIPify-Clang fails to install on CentOS/RHEL with the following error:
+## Upgrade to AMD ROCm v4.1 Not Supported
 
-*file from install of hipify-clang conflicts with file from package hip-base*
+An upgrade from previous releases to AMD ROCm v4.1 is not supported. A complete uninstallation of previous ROCm versions is required before installing a new version of ROCm.
 
-**Workaround**: This is a known issue and the following workaround is recommended for a successful installation of HIPify-Clang on CentOS/RHEL:
+## Performance Impact for Kernel Launch Bound Attribute
 
-* Download HIPify-Clang RPM. For example, *hipify-clang-11.0.0.x86_64.rpm*
-* Perform a force install using the following command: 
+Kernels without the *__launch_bounds__* attribute assume the default maximum threads per block value. In the previous ROCm release, this value was 256. In the ROCm v4.1 release, it is changed to 1024. The objective of this change ensures the actual threads per block value used to launch a kernel, by default, are always within the launch bounds, thus, establishing the correctness of HIP programs. 
 
-  *sudo rpm -ivh --force hipify-clang-11.0.0.x86_64.rpm*
+**NOTE**: Using the above-mentioned approach may incur performance degradation in certain cases. Users must add a minimum launch bound to each kernel, which covers all possible threads per block values used to launch that kernel for correctness and performance. 
+
+The recommended workaround to recover the performance is to add *--gpu-max-threads-per-block=256* to the compilation options for HIP programs.
+
+## Issue with Passing a Subset of GPUs in a Multi-GPU System
+
+ROCm support for passing individual GPUs via the docker --device flag in a Docker run command has a known issue when passing a subset of GPUs in a multi-GPU system. The command runs without any warning or error notification. However, all GPU executable run outputs are randomly corrupted. 
+
+Using GPU targeting via the Docker command is not recommended for users of ROCm 4.1. There is no workaround for this issue currently. 
+
+## Performance Impact for LDS-Bound Kernels
+
+The compiler in ROCm v4.1 generates LDS load and stores instructions that incorrectly assume equal performance between aligned and misaligned accesses. While this does not impact code correctness, it may result in sub-optimal performance.
+
+This issue is under investigation, and there is no known workaround at this time.
+
+## Changed HIP Environment Variables in ROCm v4.1 Release
+
+In the ROCm v3.5 release, the Heterogeneous Compute Compiler (HCC) compiler was deprecated, and the HIP-Clang compiler was introduced for compiling Heterogeneous-Compute Interface for Portability (HIP) programs. Also, the HIP runtime API was implemented on top of the Radeon Open Compute Common Language runtime (ROCclr). ROCclr is an abstraction layer that provides the ability to interact with different runtime backends such as ROCr. 
+
+While the *HIP_PLATFORM=hcc* environment variable was functional in subsequent releases after ROCm v3.5, in the ROCm v4.1 release, changes to the following environment variables were implemented: 
+
+* *HIP_PLATFORM=hcc was changed to HIP_PLATFORM=amd*
+
+* *HIP_PLATFORM=nvcc was changed to HIP_PLATFORM=nvidia*
+
+Therefore, any applications continuing to use the HIP_PLATFORM=hcc environment variable will fail.
+
+**Workaround:**  Update the environment variables to reflect the changes mentioned above.
 
 
-## Failure to Process Breakpoint before Queue Destroy Results in ROCm Debugger Error
-When ROCgdb is in non-stop mode with an application that rapidly creates and destroys queues, a breakpoint may be reported that is not processed by the debugger before the queue is deleted. In some cases, this can result in the following error that prevents further debugging:
+# Deprecations
 
-*[amd-dbgapi]: fatal error: kfd_queue_id 2 should have been reported as a NEW_QUEUE before next_pending_event failed (rc=-2)*
+This section describes deprecations and removals in AMD ROCm.
 
-There are no known workarounds at this time.
+## Compiler Generated Code Object Version 2 Deprecation 
 
-## Failure to Process Breakpoint before Queue Destroy Results in ROCm Debugger API Error
-
-When the ROCdbgapi library is used with an application that rapidly creates and destroys queues, a breakpoint may be reported that is not processed by the client before the queue is deleted. In some cases, this can result in a fatal error and the following error log message is produced:
-
-*[amd-dbgapi]: fatal error: kfd_queue_id 2 should have been reported as a NEW_QUEUE before next_pending_event failed (rc=-2)*
-
-There are no known workarounds at this time.
-
-## rocThrust and hipCUB Unit Test Failures 
-
-The following unit test failures have been observed due to known issues in the ROCclr runtime. 
-
-rocThrust
-* sort 
-* sort_by_key
-
-hipCUB
-* BlockDiscontinuity 
-* BlockExchange 
-* BlockHistogram 
-* BlockRadixSort
-* BlockReduce 
-* BlockScan
-
-There are no known workarounds in the current release. 
+Compiler-generated code object version 2 is no longer supported and has been completely removed. Support for loading code object version 2 is also deprecated with no announced removal release.
 
 
-## Multiple GPU Configuration Freezes with Imagenet Training and tf_cnn_benchmark on TensorFlow 
+# Deploying ROCm
 
-A random freeze has been observed with Imagenet training and tf_cnn_benchmark on TensorFlow when multiple GPU configurations are involved. 
-
-There is no freeze observed with single GPUs.  
-
-There are no known workarounds at this time.
-
-## Issue with Running AMD ROCm v3.3 User Mode with AMD ROCm v3.5 DKMS Kernel Module
-
-Running AMD ROCm v3.3 in the user mode with the AMD ROCm v3.5 DKMS kernel module will cause the following features to be broken:
-
-* IPC import/export, cross memory copy (used by UCX and MPI)
-* Experimental GDB support
-
-**Resolution**: Install ROCm v3.5 Thunk (*Hsakmt*) when using ROCm 3.5 Kernel Fusion Driver (KFD).
-
-
-## SQLite3 Library Not Found in ROCProfiler
-
-The ROCProfiler tool appears to be broken when the SQLite3 library is not found.
-
-**Resolution**: Install the SQLite3 Python module separately and ensure the environment variable is set to ROCP_PYTHON_VERSION to confirm the Python version, which includes the SQLite3 module.
-
-
-
-# Deprecations 
-Install ROCm v3.5 Thunk (Hsakmt) when using ROCm 3.5 Kernel Fusion Driver (KFD). You can access the Thunk package at:
-
-https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface
-## Heterogeneous Compute Compiler
-In this release, the Heterogeneous Compute Compiler (HCC) compiler is deprecated and the HIP-Clang compiler is introduced for compiling Heterogeneous-Compute Interface for Portability (HIP) programs.
-
-For more information, see HIP documentation at:
-https://rocmdocs.amd.com/en/latest/Programming_Guides/Programming-Guides.html
-
-
-## Deploying ROCm
-AMD hosts both Debian and RPM repositories for the ROCm v3.5.x packages. 
+AMD hosts both Debian and RPM repositories for the ROCm packages. 
 
 For more information on ROCM installation on all platforms, see
 
 https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html
 
-## Hardware and Software Support
+
+# Machine Learning and High Performance Computing Software Stack for AMD GPU
+
+For an updated version of the software stack for AMD GPU, see
+
+https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#software-stack-for-amd-gpu
+
+
+
+# Hardware and Software Support
 ROCm is focused on using AMD GPUs to accelerate computational tasks such as machine learning, engineering workloads, and scientific computing.
 In order to focus our development efforts on these domains of interest, ROCm supports a targeted set of hardware configurations which are detailed further in this section.
 
+**Note:** The AMD ROCm™ open software platform is a compute stack for headless system deployments. GUI-based software applications are currently not supported.
+
 #### Supported GPUs
 Because the ROCm Platform has a focus on particular computational domains, we offer official support for a selection of AMD GPUs that are designed to offer good performance and price in these domains.
 
+**Note:** The integrated GPUs of Ryzen are not officially supported targets for ROCm.
+
 ROCm officially supports AMD GPUs that use following chips:
 
-  * GFX8 GPUs
-    * "Fiji" chips, such as on the AMD Radeon R9 Fury X and Radeon Instinct MI8
-    * "Polaris 10" chips, such as on the AMD Radeon RX 580 and Radeon Instinct MI6
-  * GFX9 GPUs
-    * "Vega 10" chips, such as on the AMD Radeon RX Vega 64 and Radeon Instinct MI25
-    * "Vega 7nm" chips, such as on the Radeon Instinct MI50, Radeon Instinct MI60 or AMD Radeon VII
+* GFX9 GPUs
+
+  - "Vega 10" chips, such as on the AMD Radeon RX Vega 64 and Radeon Instinct MI25
+  
+  - "Vega 7nm" chips, such as on the Radeon Instinct MI50, Radeon Instinct MI60 or AMD Radeon VII, Radeon Pro VII
+
+* CDNA GPUs
+
+  - MI100 chips such as on the AMD Instinct™ MI100
+
 
 ROCm is a collection of software ranging from drivers and runtimes to libraries and developer tools.
 Some of this software may work with more GPUs than the "officially supported" list above, though AMD does not make any official claims of support for these devices on the ROCm software platform.
+
 The following list of GPUs are enabled in the ROCm software, though full support is not guaranteed:
 
   * GFX8 GPUs
@@ -393,10 +642,10 @@ As described in the next section, GFX8 GPUs require PCI Express 3.0 (PCIe 3.0) w
 
 The integrated GPUs in AMD APUs are not officially supported targets for ROCm.
 As described [below](#limited-support), "Carrizo", "Bristol Ridge", and "Raven Ridge" APUs are enabled in our upstream drivers and the ROCm OpenCL runtime.
-However, they are not enabled in our HCC or HIP runtimes, and may not work due to motherboard or OEM hardware limitations.
+However, they are not enabled in the HIP runtime, and may not work due to motherboard or OEM hardware limitations.
 As such, they are not yet officially supported targets for ROCm.
 
-For a more detailed list of hardware support, please see [the following documentation](https://rocm.github.io/hardware.html).
+For a more detailed list of hardware support, please see [the following documentation](https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units).
 
 #### Supported CPUs
 As described above, GFX8 GPUs require PCIe 3.0 with PCIe atomics in order to run ROCm.
@@ -436,26 +685,29 @@ does not require or take advantage of PCIe Atomics. However, we still recommend
 from the list provided above for compatibility purposes.
 
 #### Not supported or limited support under ROCm
+
 ##### Limited support
 
-* ROCm 2.9.x should support PCIe 2.0 enabled CPUs such as the AMD Opteron, Phenom, Phenom II, Athlon, Athlon X2, Athlon II and older Intel Xeon and Intel Core Architecture and Pentium CPUs. However, we have done very limited testing on these configurations, since our test farm has been catering to CPUs listed above. This is where we need community support. _If you find problems on such setups, please report these issues_.
+* ROCm 4.x should support PCIe 2.0 enabled CPUs such as the AMD Opteron, Phenom, Phenom II, Athlon, Athlon X2, Athlon II and older Intel Xeon and Intel Core Architecture and Pentium CPUs. However, we have done very limited testing on these configurations, since our test farm has been catering to CPUs listed above. This is where we need community support. _If you find problems on such setups, please report these issues_.
 * Thunderbolt 1, 2, and 3 enabled breakout boxes should now be able to work with ROCm. Thunderbolt 1 and 2 are PCIe 2.0 based, and thus are only supported with GPUs that do not require PCIe 3.1.0 atomics (e.g. Vega 10). However, we have done no testing on this configuration and would need community support due to limited access to this type of equipment.
-* AMD "Carrizo" and "Bristol Ridge" APUs are enabled to run OpenCL, but do not yet support HCC, HIP, or our libraries built on top of these compilers and runtimes.
+* AMD "Carrizo" and "Bristol Ridge" APUs are enabled to run OpenCL, but do not yet support HIP or our libraries built on top of these compilers and runtimes.
   * As of ROCm 2.1, "Carrizo" and "Bristol Ridge" require the use of upstream kernel drivers.
   * In addition, various "Carrizo" and "Bristol Ridge" platforms may not work due to OEM and ODM choices when it comes to key configurations parameters such as inclusion of the required CRAT tables and IOMMU configuration parameters in the system BIOS.
   * Before purchasing such a system for ROCm, please verify that the BIOS provides an option for enabling IOMMUv2 and that the system BIOS properly exposes the correct CRAT table. Inquire with your vendor about the latter.
-* AMD "Raven Ridge" APUs are enabled to run OpenCL, but do not yet support HCC, HIP, or our libraries built on top of these compilers and runtimes.
+* AMD "Raven Ridge" APUs are enabled to run OpenCL, but do not yet support HIP or our libraries built on top of these compilers and runtimes.
   * As of ROCm 2.1, "Raven Ridge" requires the use of upstream kernel drivers.
   * In addition, various "Raven Ridge" platforms may not work due to OEM and ODM choices when it comes to key configurations parameters such as inclusion of the required CRAT tables and IOMMU configuration parameters in the system BIOS.
   * Before purchasing such a system for ROCm, please verify that the BIOS provides an option for enabling IOMMUv2 and that the system BIOS properly exposes the correct CRAT table. Inquire with your vendor about the latter.
 
 ##### Not supported
 
-* "Tonga", "Iceland", "Vega M", and "Vega 12" GPUs are not supported in ROCm 2.9.x
+* "Tonga", "Iceland", "Vega M", and "Vega 12" GPUs are not supported.
 * We do not support GFX8-class GPUs (Fiji, Polaris, etc.) on CPUs that do not have PCIe 3.0 with PCIe atomics.
   * As such, we do not support AMD Carrizo and Kaveri APUs as hosts for such GPUs.
   * Thunderbolt 1 and 2 enabled GPUs are not supported by GFX8 GPUs on ROCm. Thunderbolt 1 & 2 are based on PCIe 2.0.
 
+In the default ROCm configuration, GFX8 and GFX9 GPUs require PCI Express 3.0 with PCIe atomics. The ROCm platform leverages these advanced capabilities to allow features such as user-level submission of work from the host to the GPU. This includes PCIe atomic Fetch and Add, Compare and Swap, Unconditional Swap, and AtomicOp Completion.
+
 #### ROCm support in upstream Linux kernels
 
 As of ROCm 1.9.0, the ROCm user-level software is compatible with the AMD drivers in certain upstream Linux kernels.
@@ -482,9 +734,17 @@ For users that have the option of using either AMD's or the upstreamed driver, t
 |      |   | Does not include most up-to-date firmware |
 
 
+# Disclaimer
 
-## Machine Learning and High Performance Computing Software Stack for AMD GPU
+AMD®, the AMD Arrow logo, AMD Instinct™, Radeon™, ROCm® and combinations thereof are trademarks of Advanced Micro Devices, Inc.
 
-For an updated version of the software stack for AMD GPU, see
+Linux® is the registered trademark of Linus Torvalds in the U.S. and other countries.
+
+PCIe® is a registered trademark of PCI-SIG Corporation. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.  
+
+Google®  is a registered trademark of Google LLC.
+
+Ubuntu and the Ubuntu logo are registered trademarks of Canonical Ltd.
+
+Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.
 
-https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation-Guide.html#machine-learning-and-high-performance-computing-software-stack-for-amd-gpu-v3-5-0

default.xml

@@ -1,80 +1,75 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <manifest>
     <remote name="roc-github"
-            fetch="http://github.com/RadeonOpenCompute/" />
+fetch="http://github.com/RadeonOpenCompute/" />
     <remote name="rocm-devtools"
-            fetch="https://github.com/ROCm-Developer-Tools/" />
+fetch="https://github.com/ROCm-Developer-Tools/" />
     <remote name="rocm-swplat"
-            fetch="https://github.com/ROCmSoftwarePlatform/" />
+fetch="https://github.com/ROCmSoftwarePlatform/" />
     <remote name="gpuopen-libs"
-            fetch="https://github.com/GPUOpen-ProfessionalCompute-Libraries/" />
+fetch="https://github.com/GPUOpen-ProfessionalCompute-Libraries/" />
     <remote name="gpuopen-tools"
-            fetch="https://github.com/GPUOpen-Tools/" />
-    <remote name="KhronosGroup" 
-            fetch="https://github.com/KhronosGroup/" />
-    <default revision="refs/tags/rocm-3.5.0"
-             remote="roc-github"
-             sync-c="true"
-             sync-j="4" />
-    <!--list of projects for ROCM-->
-    <project name="ROCK-Kernel-Driver" revision="refs/tags/rocm-3.5.1" />
+fetch="https://github.com/GPUOpen-Tools/" />
+    <remote name="KhronosGroup"
+fetch="https://github.com/KhronosGroup/" />
+    <default revision="refs/tags/rocm-4.1.0"
+     remote="roc-github"
+     sync-c="true"
+     sync-j="4" />
+<!--list of projects for ROCM-->
+    <project name="ROCK-Kernel-Driver" />
     <project name="ROCT-Thunk-Interface" />
     <project name="ROCR-Runtime" />
     <project name="ROC-smi" />
-    <project name="rocm_smi_lib" remote="roc-github" />
+    <project name="rocm_smi_lib" />
     <project name="rocm-cmake" />
     <project name="rocminfo" />
     <project name="rocprofiler" remote="rocm-devtools" />
     <project name="roctracer" remote="rocm-devtools" />
-    <project name="ROCm-OpenCL-Runtime" revision="refs/tags/roc-3.5.0" />
+    <project name="ROCm-OpenCL-Runtime" />
     <project path="ROCm-OpenCL-Runtime/api/opencl/khronos/icd" name="OpenCL-ICD-Loader" remote="KhronosGroup" revision="6c03f8b58fafd9dd693eaac826749a5cfad515f8" />
     <project name="clang-ocl" />
-    <!--HIP Projects-->
-    <project name="HCC-Example-Application" remote="rocm-devtools" revision="ffd6533305e79eed667badd3c4cdb7879a1281b8" />
-    <project name="HIP" remote="rocm-devtools" revision="refs/tags/rocm-3.5.1" />
+<!--HIP Projects-->
+    <project name="HIP" remote="rocm-devtools" revision="refs/tags/rocm-4.1.1" />
     <project name="HIP-Examples" remote="rocm-devtools" />
-    <project name="ROCclr" remote="rocm-devtools" revision="refs/tags/roc-3.5.0" />
-    <project name="HIPIFY" remote="rocm-devtools" /> 
-    <!-- The following projects are all associated with the AMDGPU LLVM compiler -->
-    <project name="llvm-project" path="llvm_amd-stg-open" /> 
+    <project name="ROCclr" remote="rocm-devtools" />
+    <project name="HIPIFY" remote="rocm-devtools" />
+<!-- The following projects are all associated with the AMDGPU LLVM compiler -->
+    <project name="llvm-project" revision="refs/tags/rocm-4.1.1" />
     <project name="ROCm-Device-Libs" />
     <project name="atmi" />
     <project name="ROCm-CompilerSupport" />
-    <project name="rocr_debug_agent" remote="rocm-devtools" revision="refs/tags/roc-3.5.0" />
+    <project name="rocr_debug_agent" remote="rocm-devtools" />
     <project name="rocm_bandwidth_test" />
+    <project name="half" remote="rocm-swplat" revision="37742ce15b76b44e4b271c1e66d13d2fa7bd003e" />
     <project name="RCP" remote="gpuopen-tools" revision="3a49405a1500067c49d181844ec90aea606055bb" />
-    <!-- gdb projects -->
+<!-- gdb projects -->
     <project name="ROCgdb" remote="rocm-devtools" />
     <project name="ROCdbgapi" remote="rocm-devtools" />
-    <!-- ROCm Libraries -->
+<!-- ROCm Libraries -->
+    <project name="rdc" remote="roc-github" />
     <project name="rocBLAS" remote="rocm-swplat" />
+    <project name="Tensile" remote="rocm-swplat" />
     <project name="hipBLAS" remote="rocm-swplat" />
     <project name="rocFFT" remote="rocm-swplat" />
+    <project name="hipFFT" remote="rocm-swplat" />
     <project name="rocRAND" remote="rocm-swplat" />
     <project name="rocSPARSE" remote="rocm-swplat" />
     <project name="rocSOLVER" remote="rocm-swplat" />
     <project name="hipSPARSE" remote="rocm-swplat" />
     <project name="rocALUTION" remote="rocm-swplat" />
-    <project name="MIOpenGEMM" remote="rocm-swplat" revision="refs/tags/1.1.6" />
+    <project name="MIOpenGEMM" remote="rocm-swplat" />
     <project name="MIOpen" remote="rocm-swplat" />
     <project name="rccl" remote="rocm-swplat" />
-    <project name="MIVisionX" remote="gpuopen-libs" revision="refs/tags/1.7" />
+    <project name="MIVisionX" remote="gpuopen-libs" />
     <project name="rocThrust" remote="rocm-swplat" />
     <project name="hipCUB" remote="rocm-swplat" />
-    <project name="rocPRIM" remote="rocm-swplat" revision="refs/tags/rocm-3.5.1" />
+    <project name="rocPRIM" remote="rocm-swplat" />
+    <project name="hipfort" remote="rocm-swplat" />
     <project name="AMDMIGraphX" remote="rocm-swplat" />
     <project name="ROCmValidationSuite" remote="rocm-devtools" />
-    <!-- Projects for AOMP -->
-    <project name="ROCT-Thunk-Interface" path="aomp/roct-thunk-interface" remote="roc-github" />
-    <project name="ROCR-Runtime" path="aomp/rocr-runtime" remote="roc-github" />
-    <project name="ROCm-Device-Libs" path="aomp/rocm-device-libs" remote="roc-github" />
-    <project name="ROCm-CompilerSupport" path="aomp/rocm-compilersupport" remote="roc-github" />
-    <project name="rocminfo" path="aomp/rocminfo" remote="roc-github" />
-    <project name="HIP" path="aomp/hip-on-vdi" remote="rocm-devtools" revision="ffcbd7e63395f8a4d3ccb7e4d5133f8d2dde793e"  />
-    <project name="aomp" path="aomp/aomp" remote="rocm-devtools" />
-    <project name="aomp-extras" path="aomp/aomp-extras" remote="rocm-devtools" />
-    <project name="flang" path="aomp/flang" remote="rocm-devtools" />
-    <project name="amd-llvm-project" path="aomp/amd-llvm-project" remote="rocm-devtools" />
-    <project name="ROCclr" path="aomp/vdi" remote="rocm-devtools" revision="72ce2c9783d514fc7da94db40f9f420320df098d" />
-    <project name="ROCm-OpenCL-Runtime" path="aomp/opencl-on-vdi" remote="roc-github" revision="12fb33212c99cb4b596b0f34691e7d044218e3e9" />
+<!-- Projects for OpenMP-Extras -->
+    <project name="aomp" path="openmp-extras/aomp" remote="rocm-devtools" />
+    <project name="aomp-extras" path="openmp-extras/aomp-extras" remote="rocm-devtools" />
+    <project name="flang" path="openmp-extras/flang" remote="rocm-devtools" />
 </manifest>

version_history.md

@@ -1,503 +0,0 @@
-## ROCm Version History
-This file contains archived version history information for the [ROCm project](https://github.com/RadeonOpenCompute/ROCm)
-
-### Current ROCm Version: 3.3
-- [New features and enhancements in ROCm v3.1](#new-features-and-enhancements-in-rocm-v31)
-- [New features and enhancements in ROCm v3.0](#new-features-and-enhancements-in-rocm-v30)
-- [New features and enhancements in ROCm v2.10](#new-features-and-enhancements-in-rocm-v210)
-- [New features and enhancements in ROCm 2.9](#new-features-and-enhancements-in-rocm-29)
-- [New features and enhancements in ROCm 2.8](#new-features-and-enhancements-in-rocm-28)
-- [New features and enhancements in ROCm 2.7.2](#new-features-and-enhancements-in-rocm-272)
-- [New features and enhancements in ROCm 2.7](#new-features-and-enhancements-in-rocm-27)
-- [New features and enhancements in ROCm 2.6](#new-features-and-enhancements-in-rocm-26)
-- [New features and enhancements in ROCm 2.5](#new-features-and-enhancements-in-rocm-25)
-- [New features and enhancements in ROCm 2.4](#new-features-and-enhancements-in-rocm-24)
-- [New features and enhancements in ROCm 2.3](#new-features-and-enhancements-in-rocm-23)
-- [New features and enhancements in ROCm 2.2](#new-features-and-enhancements-in-rocm-22)
-- [New features and enhancements in ROCm 2.1](#new-features-and-enhancements-in-rocm-21)
-- [New features and enhancements in ROCm 2.0](#new-features-and-enhancements-in-rocm-20)
-- [New features and enhancements in ROCm 1.9.2](#new-features-and-enhancements-in-rocm-192)
-- [New features and enhancements in ROCm 1.9.2](#new-features-and-enhancements-in-rocm-192-1)
-- [New features and enhancements in ROCm 1.9.1](#new-features-and-enhancements-in-rocm-191)
-- [New features and enhancements in ROCm 1.9.0](#new-features-and-enhancements-in-rocm-190)
-- [New features as of ROCm 1.8.3](#new-features-as-of-rocm-183)
-- [New features as of ROCm 1.8](#new-features-as-of-rocm-18)
-- [New Features as of ROCm 1.7](#new-features-as-of-rocm-17)
-- [New Features as of ROCm 1.5](#new-features-as-of-rocm-15)
-
-
-## New features and enhancements in ROCm v3.2
-The AMD ROCm v3.2 release was not productized.
-
-## New features and enhancements in ROCm v3.1
-### Change in ROCm Installation Directory Structure
-
-A fresh installation of the ROCm toolkit installs the packages in the /opt/rocm-<version> folder. Previously, ROCm toolkit packages were installed in the /opt/rocm folder.
-
-### Reliability, Accessibility, and Serviceability Support for Vega 7nm
-
-The Reliability, Accessibility, and Serviceability (RAS) support for Vega7nm is now available.
-
-### SLURM Support for AMD GPU
-
-SLURM (Simple Linux Utility for Resource Management) is an open source, fault-tolerant, and highly scalable cluster management and job scheduling system for large and small Linux clusters.
-
-
-## New features and enhancements in ROCm v3.0
-### Support for CentOS RHEL v7.7 <a id="centos-anchor"></a> 
-Support is extended for CentOS/RHEL v7.7 in the ROCm v3.0 release. For more information about the CentOS/RHEL v7.7 release, see:
-
-[CentOS/RHEL](https://centos.org/forums/viewtopic.php?t=71657)
-
-
-### Initial distribution of AOMP 0.7-5 in ROCm v3.0 <a id="aomp-anchor"></a> 
-The code base for this release of AOMP is the Clang/LLVM 9.0 sources as of October 8th, 2019. The LLVM-project branch used to build this release is AOMP-191008. It is now locked. With this release, an artifact tarball of the entire source tree is created. This tree includes a Makefile in the root directory used to build AOMP from the release tarball. You can use Spack to build AOMP from this source tarball or build manually without Spack.
-
-For more information about AOMP 0.7-5, see: [AOMP](https://github.com/ROCm-Developer-Tools/aomp/tree/roc-3.0.0)
-
-
-### Fast Fourier Transform Updates
-The Fast Fourier Transform (FFT) is an efficient algorithm for computing the Discrete Fourier Transform. Fast Fourier transforms are used in signal processing, image processing, and many other areas. The following real FFT performance change is made in the ROCm v3.0 release:
-
-•	Implement efficient real/complex 2D transforms for even lengths.
-
-Other improvements:
-
-•	More 2D test coverage sizes.
-
-•	Fix buffer allocation error for large 1D transforms.
-
-•	C++ compatibility improvements.
-
-### MemCopy Enhancement for rocProf
-In the v3.0 release, the rocProf tool is enhanced with an additional capability to dump asynchronous GPU memcopy information into a .csv file. You can use the '-hsa-trace' option to create the results_mcopy.csv file.
-Future enhancements will include column labels.
-
-
-### New features and enhancements in ROCm v2.10
-#### rocBLAS Support for Complex GEMM
-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 Instinct MI50. 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.
-
-#### 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.
-
-### New features and enhancements in ROCm 2.9
-
-#### Initial release for Radeon Augmentation Library(RALI)
-The AMD Radeon Augmentation Library (RALI) is designed to efficiently decode and process images from a variety of storage formats and modify them through a processing graph programmable by the user. RALI currently provides C API.
-
-#### Quantization in MIGraphX v0.4
-MIGraphX 0.4 introduces support for fp16 and int8 quantization. For additional details, as well as other new MIGraphX features, see [MIGraphX documentation](https://github.com/ROCmSoftwarePlatform/AMDMIGraphX/wiki/Getting-started:-using-the-new-features-of-MIGraphX-0.4).
-
-#### rocSparse csrgemm
-csrgemm enables the user to perform matrix-matrix multiplication with two sparse matrices in CSR format.
-
-#### Singularity Support
-ROCm 2.9 adds support for Singularity container version 2.5.2.
-
-#### Initial release of rocTX
-ROCm 2.9 introduces rocTX, which provides a C API for code markup for performance profiling.  This initial release of rocTX supports annotation of code ranges and ASCII markers.  For an example, see this [code](https://github.com/ROCm-Developer-Tools/roctracer/blob/amd-master/test/MatrixTranspose_test/MatrixTranspose.cpp).
-
-#### Added support for Ubuntu 18.04.3
-Ubuntu 18.04.3 is now supported in ROCm 2.9.
-
-
-
-### New features and enhancements in ROCm 2.8
-
-#### Support for NCCL2.4.8 API
-Implements ncclCommAbort() and ncclCommGetAsyncError() to match the NCCL 2.4.x API
-
-### New features and enhancements in ROCm 2.7.2
-
-This release is a hotfix for ROCm release 2.7.
-
-#### Issues fixed in ROCm 2.7.2
-
-##### A defect in upgrades from older ROCm releases has been fixed.
-
-##### rocprofiler --hiptrace and --hsatrace fails to load roctracer library
-In ROCm 2.7.2, rocprofiler --hiptrace and --hsatrace fails to load roctracer library defect has been fixed.  
-To generate traces, please provide directory path also using the parameter: -d <$directoryPath> for example:
-```shell
-/opt/rocm/bin/rocprof  --hsa-trace -d $PWD/traces /opt/rocm/hip/samples/0_Intro/bit_extract/bit_extract
-  ```
-All traces and results will be saved under $PWD/traces path
-
-#### Upgrading from ROCm 2.7 to 2.7.2
-
-To upgrade, please remove 2.7 completely as specified [for ubuntu](#how-to-uninstall-from-ubuntu-1604-or-Ubuntu-1804) or [for centos/rhel](#how-to-uninstall-rocm-from-centosrhel-76), and install 2.7.2 as per instructions [install instructions](#installing-from-amd-rocm-repositories)
-
-#### Other notes
-
-  To use rocprofiler features, the following steps need to be completed before using rocprofiler:
-
-  ##### Step-1: Install roctracer
-
-###### Ubuntu 16.04 or Ubuntu 18.04:
-
-  ```shell
-  sudo apt install roctracer-dev
-  ```
-
-######  CentOS/RHEL 7.6:
-
-  ```shell
-  sudo yum install roctracer-dev
-  ```
-  ##### Step-2: Add /opt/rocm/roctracer/lib to LD_LIBRARY_PATH
-  
-### New features and enhancements in ROCm 2.7
-
-#### [rocFFT] Real FFT Functional
-Improved real/complex 1D even-length transforms of unit stride. Performance improvements of up to 4.5x are observed. Large problem sizes should see approximately 2x.
-
-#### rocRand Enhancements and Optimizations
-- Added support for new datatypes: uchar, ushort, half.
-- Improved performance on "Vega 7nm" chips, such as on the Radeon Instinct MI50
-- mtgp32 uniform double performance changes due generation algorithm standardization. Better quality random numbers now generated with 30% decrease in performance
-- Up to 5% performance improvements for other algorithms
-
-#### RAS
-Added support for RAS on Radeon Instinct MI50, including:
-- Memory error detection
-- Memory error detection counter
-
-#### ROCm-SMI enhancements
-Added ROCm-SMI CLI and LIB support for FW version, compute running processes, utilization rates, utilization counter, link error counter, and unique ID.
-
-### New features and enhancements in ROCm 2.6
-
-#### ROCmInfo enhancements
-ROCmInfo was extended to do the following:
-For ROCr API call errors including initialization determine if the error could be explained by:
-- ROCk (driver) is not loaded / available
-- User does not have membership in appropriate group - "video"
-- If not above print the error string that is mapped to the returned error code
-- If no error string is available, print the error code in hex
-
-#### Thrust - Functional Support on Vega20
-ROCm2.6 contains the first official release of rocThrust and hipCUB. rocThrust is a port of thrust, a parallel algorithm library. hipCUB is a port of CUB, a reusable software component library. Thrust/CUB has been ported to the HIP/ROCm platform to use the rocPRIM library. The HIP ported library works on HIP/ROCm platforms.
-
-Note: rocThrust and hipCUB library replaces https://github.com/ROCmSoftwarePlatform/thrust (hip-thrust), i.e. hip-thrust has been separated into two libraries, rocThrust and hipCUB. Existing hip-thrust users are encouraged to port their code to rocThrust and/or hipCUB. Hip-thrust will be removed from official distribution later this year.
-
-#### MIGraphX v0.3
-MIGraphX optimizer adds support to read models frozen from Tensorflow framework. Further details and an example usage at https://github.com/ROCmSoftwarePlatform/AMDMIGraphX/wiki/Getting-started:-using-the-new-features-of-MIGraphX-0.3
-
-#### MIOpen 2.0
-- This release contains several new features including an immediate mode for selecting convolutions, bfloat16 support, new layers, modes, and algorithms.
-- MIOpenDriver, a tool for benchmarking and developing kernels is now shipped with MIOpen.
-BFloat16 now supported in HIP requires an updated rocBLAS as a GEMM backend.
-- Immediate mode API now provides the ability to quickly obtain a convolution kernel.
-- MIOpen now contains HIP source kernels and implements the ImplicitGEMM kernels. This is a new feature and is currently disabled by default. Use the environmental variable "MIOPEN_DEBUG_CONV_IMPLICIT_GEMM=1" to activation this feature. ImplicitGEMM requires an up to date HIP version of at least 1.5.9211.
-- A new "loss" catagory of layers has been added, of which, CTC loss is the first. See the API reference for more details.
-2.0 is the last release of active support for gfx803 architectures. In future releases, MIOpen will not actively debug and develop new features specifically for gfx803.
-- System Find-Db in memory cache is disabled by default. Please see build instructions to enable this feature.
-Additional documentation can be found here: https://rocmsoftwareplatform.github.io/MIOpen/doc/html/
-
-#### Bloat16 software support in rocBLAS/Tensile
-Added mixed precision bfloat16/IEEE f32 to gemm_ex. The input and output matrices are bfloat16. All arithmetic is in IEEE f32.
-
-#### AMD Infinity Fabric™ Link enablement
-The ability to connect four Radeon Instinct MI60 or Radeon Instinct MI50 boards in two hives or two Radeon Instinct MI60 or Radeon Instinct MI50 boards in four hives via AMD Infinity Fabric™ Link GPU interconnect technology has been added.
-
-#### ROCm-smi features and bug fixes
-  - mGPU & Vendor check
-  - Fix clock printout if DPM is disabled
-  - Fix finding marketing info on CentOS
-  - Clarify some error messages
-
-#### ROCm-smi-lib enhancements
-- Documentation updates
-- Improvements to *name_get functions
-
-#### RCCL2 Enablement
-RCCL2 supports collectives intranode communication using PCIe, Infinity Fabric™, and pinned host memory, as well as internode communication using Ethernet (TCP/IP sockets) and Infiniband/RoCE (Infiniband Verbs).  Note: For Infiniband/RoCE, RDMA is not currently supported.
-
-#### rocFFT enhancements
-- Added: Debian package with FFT test, benchmark, and sample programs
-- Improved: hipFFT interfaces
-- Improved: rocFFT CPU reference code, plan generation code and logging code
-
-### New features and enhancements in ROCm 2.5
-
-#### UCX 1.6 support
-Support for UCX version 1.6 has been added.
-
-#### BFloat16 GEMM in rocBLAS/Tensile
-Software support for BFloat16 on Radeon Instinct MI50, MI60 has been added.  This includes:
-- Mixed precision GEMM with BFloat16 input and output matrices, and all arithmetic in IEEE32 bit
-- Input matrix values are converted from BFloat16 to IEEE32 bit, all arithmetic and accumulation is IEEE32 bit. Output values are rounded from IEEE32 bit to BFloat16
-- Accuracy should be correct to 0.5 ULP
-
-#### ROCm-SMI enhancements
-CLI support for querying the memory size, driver version, and firmware version has been added to ROCm-smi.
-
-#### [PyTorch] multi-GPU functional support (CPU aggregation/Data Parallel)
-Multi-GPU support is enabled in PyTorch using Dataparallel path for versions of PyTorch built using the 06c8aa7a3bbd91cda2fd6255ec82aad21fa1c0d5 commit or later.
-
-#### rocSparse optimization on Radeon Instinct MI50 and MI60
-This release includes performance optimizations for csrsv routines in the rocSparse library.
-
-#### [Thrust] Preview
-Preview release for early adopters. rocThrust is a port of thrust, a parallel algorithm library. Thrust has been ported to the HIP/ROCm platform to use the rocPRIM library. The HIP ported library works on HIP/ROCm platforms.
-
-Note: This library will replace https://github.com/ROCmSoftwarePlatform/thrust in a future release. The package for rocThrust (this library) currently conflicts with version 2.5 package of thrust. They should not be installed together.
-
-#### Support overlapping kernel execution in same HIP stream
-HIP API has been enhanced to allow independent kernels to run in parallel on the same stream.
-
-#### AMD Infinity Fabric&#x2122; Link enablement
-The ability to connect four Radeon Instinct MI60 or Radeon Instinct MI50 boards in one hive via AMD Infinity Fabric™ Link GPU interconnect technology has been added.
-### New features and enhancements in ROCm 2.4
-
-#### TensorFlow 2.0 support
-ROCm 2.4 includes the enhanced compilation toolchain and a set of bug fixes to support TensorFlow 2.0 features natively
-
-#### AMD Infinity Fabric&#x2122; Link enablement
-ROCm 2.4 adds support to connect two Radeon Instinct MI60 or Radeon Instinct MI50 boards via AMD Infinity Fabric&#x2122; Link GPU interconnect technology.
-
-### New features and enhancements in ROCm 2.3
-
-#### Mem usage per GPU
-Per GPU memory usage is added to rocm-smi.
-Display information regarding used/total bytes for VRAM, visible VRAM and GTT, via the --showmeminfo flag
-
-#### MIVisionX, v1.1 - ONNX
-ONNX parser changes to adjust to new file formats
-
-#### MIGraphX, v0.2
-MIGraphX 0.2 supports the following new features:
-* New Python API
-* Support for additional ONNX operators and fixes that now enable a large set of Imagenet models
-* Support for RNN Operators
-* Support for multi-stream Execution
-* [Experimental] Support for Tensorflow frozen protobuf files
-
-See: [Getting-started:-using-the-new-features-of-MIGraphX-0.2](https://github.com/ROCmSoftwarePlatform/AMDMIGraphX/wiki/Getting-started:-using-the-new-features-of-MIGraphX-0.2) for more details
-
-#### MIOpen, v1.8 - 3d convolutions and int8
-* This release contains full 3-D convolution support and int8 support for inference.
-* Additionally, there are major updates in the performance database for major models including those found in Torchvision.
-
-See: [MIOpen releases](https://github.com/ROCmSoftwarePlatform/MIOpen/releases)
-
-#### Caffe2 -  mGPU support
-Multi-gpu support is enabled for Caffe2.
-
-#### rocTracer library, ROCm tracing API for collecting runtimes API and asynchronous GPU activity traces
-HIP/HCC domains support is introduced in rocTracer library.
-
-#### BLAS -  Int8 GEMM performance, Int8 functional and performance
-Introduces support and performance optimizations for Int8 GEMM, implements TRSV support, and includes improvements and optimizations with Tensile.
-
-#### Prioritized L1/L2/L3 BLAS (functional)
-Functional implementation of BLAS L1/L2/L3 functions
-
-#### BLAS - tensile optimization
-Improvements and optimizations with tensile
-
-#### MIOpen Int8 support
-Support for int8
-
-### New features and enhancements in ROCm 2.2
-
-#### rocSparse Optimization on Vega20
-Cache usage optimizations for csrsv (sparse triangular solve), coomv
-(SpMV in COO format) and ellmv (SpMV in ELL format) are available.
-
-#### DGEMM and DTRSM Optimization
-Improved DGEMM performance for reduced matrix sizes (k=384, k=256)
-
-#### Caffe2
-Added support for multi-GPU training
-
-### New features and enhancements in ROCm 2.1
-
-#### RocTracer v1.0 preview release – 'rocprof' HSA runtime tracing and statistics support -
-Supports HSA API tracing and HSA asynchronous GPU activity including kernels execution and memory copy
-
-#### Improvements to ROCM-SMI tool -
-Added support to show real-time PCIe bandwidth usage via the -b/--showbw flag
-
-#### DGEMM Optimizations -
-Improved DGEMM performance for large square and reduced matrix sizes (k=384, k=256)
-
-### New features and enhancements in ROCm 2.0
-
-#### Adds support for RHEL 7.6 / CentOS 7.6 and Ubuntu 18.04.1
-
-#### Adds support for Vega 7nm, Polaris 12 GPUs
-
-#### Introduces MIVisionX
-* A comprehensive computer vision and machine intelligence libraries, utilities and applications bundled into a single toolkit.
-
-#### Improvements to ROCm Libraries
-* rocSPARSE & hipSPARSE
-* rocBLAS with improved DGEMM efficiency on Vega 7nm
-
-#### MIOpen
-* This release contains general bug fixes and an updated performance database
-* Group convolutions backwards weights performance has been improved
-* RNNs now support fp16
-
-#### Tensorflow multi-gpu and Tensorflow FP16 support for Vega 7nm
-* TensorFlow v1.12 is enabled with fp16 support
-
-#### PyTorch/Caffe2 with Vega 7nm Support
-* fp16 support is enabled
-* Several bug fixes and performance enhancements
-* Known Issue: breaking changes are introduced in ROCm 2.0 which are not addressed upstream yet. Meanwhile, please continue to use ROCm fork at https://github.com/ROCmSoftwarePlatform/pytorch
-
-#### Improvements to ROCProfiler tool
-* Support for Vega 7nm
-
-#### Support for hipStreamCreateWithPriority
-* Creates a stream with the specified priority. It creates a stream on which enqueued kernels have a different priority for execution compared to kernels enqueued on normal priority streams. The priority could be higher or lower than normal priority streams.
-
-#### OpenCL 2.0 support
-* ROCm 2.0 introduces full support for kernels written in the OpenCL 2.0 C language on certain devices and systems.  Applications can detect this support by calling the “clGetDeviceInfo” query function with “parame_name” argument set to “CL_DEVICE_OPENCL_C_VERSION”.  In order to make use of OpenCL 2.0 C language features, the application must include the option “-cl-std=CL2.0” in options passed to the runtime API calls responsible for compiling or building device programs.  The complete specification for the OpenCL 2.0 C language can be obtained using the following link: https://www.khronos.org/registry/OpenCL/specs/opencl-2.0-openclc.pdf
-
-#### Improved Virtual Addressing (48 bit VA) management for Vega 10 and later GPUs
-* Fixes Clang AddressSanitizer and potentially other 3rd-party memory debugging tools with ROCm
-* Small performance improvement on workloads that do a lot of memory management
-* Removes virtual address space limitations on systems with more VRAM than system memory
-
-#### Kubernetes support
-
-### New features and enhancements in ROCm 1.9.2
-#### RDMA(MPI) support on Vega 7nm
-* Support ROCnRDMA based on Mellanox InfiniBand
-
-#### Improvements to HCC
-* Improved link time optimization
-
-#### Improvements to ROCProfiler tool
-* General bug fixes and implemented versioning APIs
-
-### New features and enhancements in ROCm 1.9.2
-#### RDMA(MPI) support on Vega 7nm
-* Support ROCnRDMA based on Mellanox InfiniBand
-
-#### Improvements to HCC
-* Improved link time optimization
-
-#### Improvements to ROCProfiler tool
-* General bug fixes and implemented versioning APIs
-
-#### Critical bug fixes
-
-### New features and enhancements in ROCm 1.9.1
-#### Added DPM support to Vega 7nm
-* Dynamic Power Management feature is enabled on Vega 7nm.
-
-#### Fix for 'ROCm profiling' that used to fail with a “Version mismatch between HSA runtime and libhsa-runtime-tools64.so.1” error
-
-### New features and enhancements in ROCm 1.9.0
-
-#### Preview for Vega 7nm
-* Enables developer preview support for Vega 7nm
-
-#### System Management Interface
-* Adds support for the ROCm SMI (System Management Interface) library, which provides monitoring and management capabilities for AMD GPUs.
-
-#### Improvements to HIP/HCC
-* Support for gfx906
-* Added deprecation warning for C++AMP.  This will be the last version of HCC supporting C++AMP.
-* Improved optimization for global address space pointers passing into a GPU kernel
-* Fixed several race conditions in the HCC runtime
-* Performance tuning to the unpinned copy engine
-* Several codegen enhancement fixes in the compiler backend
-
-#### Preview for rocprof Profiling Tool
-Developer preview (alpha) of profiling tool rocProfiler. It includes a command-line front-end, `rpl_run.sh`, which enables:
-* Cmd-line tool for dumping public per kernel perf-counters/metrics and kernel timestamps
-* Input file with counters list and kernels selecting parameters
-* Multiple counters groups and app runs supported
-* Output results in CSV format
-
-The tool can be installed from the `rocprofiler-dev` package. It will be installed into: `/opt/rocm/bin/rpl_run.sh`
-
-#### Preview for rocr Debug Agent rocr_debug_agent
-The ROCr Debug Agent is a library that can be loaded by ROCm Platform Runtime to provide the following functionality:
-* Print the state for wavefronts that report memory violation or upon executing a "s_trap 2" instruction.
-* Allows SIGINT (`ctrl c`) or SIGTERM (`kill -15`) to print wavefront state of aborted GPU dispatches.
-* It is enabled on Vega10 GPUs on ROCm1.9.
-
-The ROCm1.9 release will install the ROCr Debug Agent library at `/opt/rocm/lib/librocr_debug_agent64.so`
-
-
-#### New distribution support
-
-* Binary package support for Ubuntu 18.04
-
-#### ROCm 1.9 is ABI compatible with KFD in upstream Linux kernels.
-Upstream Linux kernels support the following GPUs in these releases:
-4.17: Fiji, Polaris 10, Polaris 11
-4.18: Fiji, Polaris 10, Polaris 11, Vega10
-
-Some ROCm features are not available in the upstream KFD:
-* More system memory available to ROCm applications
-* Interoperability between graphics and compute
-* RDMA
-* IPC
-
-To try ROCm with an upstream kernel, install ROCm as normal, but do not install the rock-dkms package. Also add a udev rule to control `/dev/kfd` permissions:
-
-```
-    echo 'SUBSYSTEM=="kfd", KERNEL=="kfd", TAG+="uaccess", GROUP="video"' | sudo tee /etc/udev/rules.d/70-kfd.rules
-```
-
-### New features as of ROCm 1.8.3
-
-* ROCm 1.8.3 is a minor update meant to fix compatibility issues on Ubuntu releases running kernel 4.15.0-33
-
-### New features as of ROCm 1.8
-
-#### DKMS driver installation
-
- * Debian packages are provided for DKMS on Ubuntu
- * RPM packages are provided for CentOS/RHEL 7.4 and 7.5 support
- * See the [ROCT-Thunk-Interface](https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/roc-1.8.x) and [ROCK-Kernel-Driver](https://github.com/RadeonOpenCompute/ROCK-Kernel-Driver/tree/roc-1.8.x) for additional documentation on driver setup
-
-#### New distribution support
-
- * Binary package support for Ubuntu 16.04 and 18.04
- * Binary package support for CentOS 7.4 and 7.5
- * Binary package support for RHEL 7.4 and 7.5
-
-#### Improved OpenMPI via UCX support
-
- * UCX support for OpenMPI
- * ROCm RDMA
-
-### New Features as of ROCm 1.7
-
-#### DKMS driver installation
-
- * New driver installation uses Dynamic Kernel Module Support (DKMS)
- * Only amdkfd and amdgpu kernel modules are installed to support AMD hardware
- * Currently only Debian packages are provided for DKMS (no Fedora suport available)
- * See the [ROCT-Thunk-Interface](https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/tree/roc-1.7.x) and [ROCK-Kernel-Driver](https://github.com/RadeonOpenCompute/ROCK-Kernel-Driver/tree/roc-1.7.x) for additional documentation on driver setup
-
-### New Features as of ROCm 1.5
-
-#### Developer preview of the new OpenCL 1.2 compatible language runtime and compiler
-
- * OpenCL 2.0 compatible kernel language support with OpenCL 1.2 compatible
-   runtime
- * Supports offline ahead of time compilation today;
-   during the Beta phase we will add in-process/in-memory compilation.
-
-#### Binary Package support for Ubuntu 16.04
-
-#### Binary Package support for Fedora 24 is not currently available
-
-#### Dropping binary package support for Ubuntu 14.04, Fedora 23
-
-#### IPC support